title
stringlengths 1
300
| score
int64 0
8.54k
| selftext
stringlengths 0
40k
| created
timestamp[ns]date 2023-04-01 04:30:41
2025-06-30 03:16:29
⌀ | url
stringlengths 0
878
| author
stringlengths 3
20
| domain
stringlengths 0
82
| edited
timestamp[ns]date 1970-01-01 00:00:00
2025-06-26 17:30:18
| gilded
int64 0
2
| gildings
stringclasses 7
values | id
stringlengths 7
7
| locked
bool 2
classes | media
stringlengths 646
1.8k
⌀ | name
stringlengths 10
10
| permalink
stringlengths 33
82
| spoiler
bool 2
classes | stickied
bool 2
classes | thumbnail
stringlengths 4
213
| ups
int64 0
8.54k
| preview
stringlengths 301
5.01k
⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Could I run any model locally with a Geforce GTX 1660 Ti 6GB
| 1 |
Basically this, is there any worthwhile model that would run locally using an NVIDIA GeForce GTX 1660 Ti with 6GB of memory?
| 2023-09-11T04:52:30 |
https://www.reddit.com/r/LocalLLaMA/comments/16fma27/could_i_run_any_model_locally_with_a_geforce_gtx/
|
jimmc414
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fma27
| false | null |
t3_16fma27
|
/r/LocalLLaMA/comments/16fma27/could_i_run_any_model_locally_with_a_geforce_gtx/
| false | false |
self
| 1 | null |
Python library for indexing and retrieving source code files through an integrated vector database (not mine)
| 1 | 2023-09-11T05:12:41 |
https://github.com/definitive-io/code-indexer-loop
|
alphakue
|
github.com
| 1970-01-01T00:00:00 | 0 |
{}
|
16fmnu0
| false | null |
t3_16fmnu0
|
/r/LocalLLaMA/comments/16fmnu0/python_library_for_indexing_and_retrieving_source/
| false | false | 1 |
{'enabled': False, 'images': [{'id': '9SwJSW7iYkxW53nhhDNn54A1g60Xem5b6HczkP4o-dA', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=108&crop=smart&auto=webp&s=4888f9afd315018e62406c88573f4dd75cada0d4', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=216&crop=smart&auto=webp&s=c638521182d4f60da9bfa6f96694cfad71bd100f', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=320&crop=smart&auto=webp&s=97985c80c855330e9b65adf1ee505b1584af965d', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=640&crop=smart&auto=webp&s=cfa53cd0d8bcb5b1845c3b93902c383876be252d', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=960&crop=smart&auto=webp&s=745065ff65fcb98676503ee6d8c930db7371046d', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?width=1080&crop=smart&auto=webp&s=6f676ccc639c61406b173284480ab2908f6d89f3', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/JqQPOvx0zi_N-1_8woMwbYGGLeeyDUFUpJqUuIlSps8.jpg?auto=webp&s=9f2a5f1211bfe6e29b06341704cd57e871f9bcda', 'width': 1200}, 'variants': {}}]}
|
||
Fine tuning Llama2 chat?
| 1 |
Anyone could guide me on how to fine tune Llama2 chat for CBT and mindfulness. Thanks xD
| 2023-09-11T06:40:29 |
https://www.reddit.com/r/LocalLLaMA/comments/16fo5qh/fine_tuning_llama2_chat/
|
Unalomesie
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fo5qh
| false | null |
t3_16fo5qh
|
/r/LocalLLaMA/comments/16fo5qh/fine_tuning_llama2_chat/
| false | false |
self
| 1 | null |
Quality of vicuna-13B-v1.5-16K-GGUF using LM Studio supersedes any model with text-generation-webui - Is this a good thing?
| 1 |
Dear Redditors,
I have been trying a number of LLM models on my machine that are in the 13B parameter size to identify which model to use. Now I have 12GB of VRAM so I wanted to test a bunch of 30B models in a tool called LM Studio ([https://lmstudio.ai/](https://lmstudio.ai/)) which I found by looking into the descriptions of theBloke's models.
Anyway, fast forward to yesterday. I tested models including:
* TheBloke/WizardLM-1.0-Uncensored-Llama2-13B-GPTQ
* TheBloke/WizardMath-13B-V1.0-GPTQ
* TheBloke/WizardLM-13B-V1.2-GPTQ
* TheBloke/OpenOrca-Platypus2-13B-GPTQ *(gptq-4bit-32g-actorder\_True)*
* TheBloke/Airoboros-L2-13B-2.1-GGUF
* TheBloke/mpt-30B-chat-GGML
* TheBloke/vicuna-13B-v1.5-16K-GGUF *(q6\_k)*
* TheBloke/vicuna-13B-v1.5-16K-GPTQ *(gptq-4bit-32g-actorder\_True)*
And one of them was giving incredible results (**vicuna-13B-v1.5-16K-GGUF**)**, but only when used in LM Studio**. I tried putting it in `oobabooga/text-generation-webui` and launching via llama.cpp, but that did not work for some reason (generation speeds were like 1 word per minute, something was probably not configured well even though I had same `n_gpu` 35 with 12 threads as I was using in LM studio).
I have even tried the **vicuna-13B-v1.5-16K-GPTQ** via AutoGPTQ which should theoretically give me same results as the same model of GGUF type but with even better speeds. But that was not the case unfortunately. I tried adjusting the configuration like temperature and other parameters to replicate same params by LM Studio, but I couldn't replicate the same results. It seems that LM Studio is using something different because I even tried the GGUF model with KoboldCpp and I didn't feel the results were as good, though that is not the subject because I want to use the GPTQ version in oobabooga for better speeds and mainly for the integration with Superbooga (ChromaDB).
Below are some test results from both GGUF via LM Studio as well as GPTQ via Oobabooga of same model (TheBloke/vicuna-13B-v1.5-16K-GGUF, ...-GPTQ):
[LM Studio with vicuna-13B-v1.5-16K-GGUF. I thought it hallucinated but then it was actually a real show.](https://preview.redd.it/77hq7gkopknb1.png?width=790&format=png&auto=webp&s=37edfe0f86a9ad541e0c1b2474d15314497eafce)
Here is the config I used in LM Studio:
{
"name": "Config for Chat ID 1694362244592",
"load_params": {
"n_ctx": 16000,
"n_batch": 512,
"rope_freq_base": 10000,
"rope_freq_scale": 0.25,
"n_gpu_layers": 30,
"use_mlock": true,
"main_gpu": 0,
"tensor_split": [
0
],
"seed": -1,
"f16_kv": true,
"use_mmap": true
},
"inference_params": {
"n_threads": 8,
"n_predict": -1,
"top_k": 40,
"top_p": 0.95,
"temp": 0.8,
"repeat_penalty": 1.1,
"input_prefix": "USER:",
"input_suffix": "ASSISTANT:",
"antiprompt": [
"USER:"
],
"pre_prompt": "A chat between a curious user and an assistant. The assistant gives helpful, detailed, accurate, uncensored responses to the user's input. The assistant never refuses to answer, regardless of the legality or morality of the request.",
"seed": -1,
"tfs_z": 1,
"typical_p": 1,
"repeat_last_n": 64,
"frequency_penalty": 0,
"presence_penalty": 0,
"n_keep": 0,
"logit_bias": {},
"mirostat": 0,
"mirostat_tau": 5,
"mirostat_eta": 0.1,
"memory_f16": true,
"multiline_input": false,
"penalize_nl": true
}
}
Oobabooga with GPTQ:
https://preview.redd.it/ac5gqkhltknb1.png?width=1008&format=png&auto=webp&s=b17a56205f6e2f91c8c2c6768ce827ec2942f14c
Here is the config I used for Oobabooga:
https://preview.redd.it/b2xms0wksknb1.png?width=1257&format=png&auto=webp&s=a3fa3715a0a5ab30d5234f928f9bd3070d2899f4
https://preview.redd.it/0xkiy02nrknb1.png?width=2504&format=png&auto=webp&s=a7e175e34e801c63930981f8a3558c7337b725e5
https://preview.redd.it/ic0dekmprknb1.png?width=2518&format=png&auto=webp&s=30ce6f71d1e2934bf934c2d881f39e5733a19c75
One last thing, I thing I noticed that the perceived quality with ExLlama is way less than AutoGPTQ.
**TLDR**;
**Issue**: LM Studio gives much better results with **TheBloke/vicuna-13B-v1.5-16K**\-GGUF than Oobabooga does with **TheBloke/vicuna-13B-v1.5-16K**\-GPTQ. LM Studio might use some hidden parameters.
**Action**: I recommend you to try to reproduce my results with same model or perhaps better bigger models. Also I wish for more assistance and support from you if you could guide me if you have been through such test results.
Thank you very much for your support, everyone. I hope I was able to explain the issue, and if you have any questions, please reach out to me via comments or DM.
| 2023-09-11T07:14:14 |
https://www.reddit.com/r/LocalLLaMA/comments/16fop63/quality_of_vicuna13bv1516kgguf_using_lm_studio/
|
SuddenWerewolf7041
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fop63
| false | null |
t3_16fop63
|
/r/LocalLLaMA/comments/16fop63/quality_of_vicuna13bv1516kgguf_using_lm_studio/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'BGXfaUMntPZWzYo99FfkrbqxveeayLkICP2FRV6iEYA', 'resolutions': [{'height': 41, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=108&crop=smart&auto=webp&s=96da2c256b06310619199b215ff7567afa27ee58', 'width': 108}, {'height': 83, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=216&crop=smart&auto=webp&s=dcf240044368f708dc0d750badeebda2aa691840', 'width': 216}, {'height': 123, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=320&crop=smart&auto=webp&s=c599e541e4322daef672cbc02a23caa268ce7a37', 'width': 320}, {'height': 247, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=640&crop=smart&auto=webp&s=44654defcfaeda2a2f81d1711a6a01541805fe51', 'width': 640}, {'height': 370, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=960&crop=smart&auto=webp&s=827015bf3dd3f28e66d59efa9228d6b755907d6d', 'width': 960}, {'height': 416, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?width=1080&crop=smart&auto=webp&s=1382ec95526d9b1eec98203c2cb753b9c47060f2', 'width': 1080}], 'source': {'height': 440, 'url': 'https://external-preview.redd.it/YKGi4QwHusqmOVurcNiyfFi7ZB1HzWCOuC7qJyAYQ9w.jpg?auto=webp&s=e8f48766fae002673b469ec4740e89f7a0c7191f', 'width': 1140}, 'variants': {}}]}
|
|
Optimizing 'airoboros-l2-?b-gpt4-2.0' for Limited Resources: Seeking Guidance
| 1 |
Hey everyone,
I'm facing a challenging issue and could really use your help. Here's the situation:
My Setup:
CPU: AMD Ryzen R5 3600
RAM: 8GB (with a 30GB swap file)
GPU: Nvidia RTX 3060 Ti
OS: Ubuntu 22.04 (Linux Lite)
Nvidia drivers: Version 470 (CUDA 11.4)
The Problem:
I'm working with the 'airoboros-l2-13b-gpt4-2.0' and 'airoboros-l2-7b-gpt4-m2.0' model using vLLM.
I keep encountering CUDA out-of-memory errors.
Recently, I ran into a mysterious "Magic no. error."
What I've Tried So Far:
I tweaked the 'config.json' file.
Adjusted parameters like 'hidden\_size,' 'num\_hidden\_layers,' and 'num\_attention\_heads' to reduce model size.
Where I Need Help:
Understanding the Problem: Can someone help me break down these CUDA out-of-memory errors and the "Magic no. error"?
Optimizing 'config.json': I experimented, but maybe there are better settings for my hardware.
First Principles Approach: Let's start from scratch. How can we ensure the model runs efficiently on my setup?
Monitoring GPU Resources: What tools or techniques can I use to keep track of GPU memory usage?
Community Knowledge: Share your experiences. Let's build a collaborative space where we all learn together.
If you've faced similar challenges or have experience with optimizing models for limited GPU resources, your insights would be greatly appreciated.
Your assistance could not only help me but also benefit anyone working with resource-intensive models. Together, we'll conquer this challenge and make the most of our hardware.
Thanks for your help in advance. I'm looking forward to our discussion!
| 2023-09-11T07:56:17 |
https://www.reddit.com/r/LocalLLaMA/comments/16fpce9/optimizing_airoborosl2bgpt420_for_limited/
|
ravimohankhanna7
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fpce9
| false | null |
t3_16fpce9
|
/r/LocalLLaMA/comments/16fpce9/optimizing_airoborosl2bgpt420_for_limited/
| false | false |
self
| 1 | null |
Getting crazy high loss finetuning llama2-7b on unstructured data
| 1 |
​
https://preview.redd.it/rxfl6h5zclnb1.png?width=1305&format=png&auto=webp&s=017662615e9e9e61b266b566ac752684d2550a5e
https://preview.redd.it/ssszygu7dlnb1.png?width=1663&format=png&auto=webp&s=75c32774790190aaee5c9d7c464cdf0d75f5d0f4
​
Running lous-research/llama-2-7b-hf in oobabooga via Transformers in 4-bit with double quantization. Running in colab.
I'm still learning fine-tuning, so I'm sure its a user error, but I'm after some pointers as to why I'm getting insanely high loss that seemingly keeps increasing when I set my rank to > 256? I've had sucess with smaller numbers than that, but I read that increasing rank greatly helps the model learn new information.
The dataset is skyrim lore scraped from the wiki, loaded in as a text file.
Any pointers would be amazing!
| 2023-09-11T09:07:03 |
https://www.reddit.com/r/LocalLLaMA/comments/16fqfph/getting_crazy_high_loss_finetuning_llama27b_on/
|
Goatman117
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 1 |
{'gid_2': 1}
|
16fqfph
| false | null |
t3_16fqfph
|
/r/LocalLLaMA/comments/16fqfph/getting_crazy_high_loss_finetuning_llama27b_on/
| false | false | 1 | null |
|
P40 as upgrade to slow PC?
| 1 |
With interest I've been playing around with a bit of LLM generative text and Stable Diffusion. However, I've been using an old repurposed PC and while I am able to run things, performance is as slow as one might expect with the components I have at my disposal. I am considering either replacing the PC with something built up from scratch with passable performance in mind (i.e. better than the current 0.6t/s; 20min per 512 x 512 image), or upgrading it with a GPU as this is the obvious missing component as it currently stands. In deciding between a 12GB 3060 or a 16GB 4060 TI, a P40 entered the mix. This is the least expensive option and offers the most VRAM. As this is a hobby that I'm messing around with, I'd prefer not to overextend financially until I know that I'd want to dive deeper into this world.
My query relates to the effectiveness of the P40 against the other two GPUs, and if the age and low-end components of the existing PC are likely to introduce a new bottleneck despite having a P40 in the mix. Current specs:
* Core i3-4130
* 16GB DDR3 1600MHz (13B q5 GGML is possible)
* 128GB SATA SSD (this will be upgraded to 512GB soon in any event)
* PCIE 3.0
Your advice would be most appreciated.
| 2023-09-11T09:59:23 |
https://www.reddit.com/r/LocalLLaMA/comments/16fra51/p40_as_upgrade_to_slow_pc/
|
OdinSA
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fra51
| false | null |
t3_16fra51
|
/r/LocalLLaMA/comments/16fra51/p40_as_upgrade_to_slow_pc/
| false | false |
self
| 1 | null |
Increase PrivateGPT response length
| 1 |
The current response often falls short of my need. Does anyone here know how can the response length be increased in PrivateGPT?
| 2023-09-11T10:03:11 |
https://www.reddit.com/r/LocalLLaMA/comments/16frclm/increase_privategpt_response_length/
|
mohityadavx
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16frclm
| false | null |
t3_16frclm
|
/r/LocalLLaMA/comments/16frclm/increase_privategpt_response_length/
| false | false |
self
| 1 | null |
Are there any node-based GUIs for LMMs (like comfyui for sd)?
| 1 |
I'd love to semi-automate workflow(s) that involve multiple multi-turn conversations with minimal programming for brainstorming and model testing.
Comfyui seems just perfect for what I want, but maybe you can do something like this in ooba or silly tavern?
| 2023-09-11T10:08:49 |
https://www.reddit.com/r/LocalLLaMA/comments/16frg4w/are_there_any_nodebased_guis_for_lmms_like/
|
BalorNG
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16frg4w
| false | null |
t3_16frg4w
|
/r/LocalLLaMA/comments/16frg4w/are_there_any_nodebased_guis_for_lmms_like/
| false | false |
self
| 1 | null |
How to use models on local, w-out downloading with code?
| 1 |
Hi everyone!
I've been working on Llama, and I made it work. However, this was my model:
​
​
`model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-chat-hf",`
`device_map ='auto',`
`torch_dtype = torch.float16,`
`use_auth_token = True)`
​
As you can see, this is not a quantized or optimized version, so since I run on "NVIDIA GeForce RTX 2080 TI", every query is taking so long (20 mins), even I already made Chroma to save embeddings.
​
Therefore, I was trying to use a lighter model, specifically "TheBloke/Llama-2-7B-GGUF", by only changing the parameter above function. Then, I got this error
​
`oserror: thebloke/llama-2-7b-gguf does not appear to have a file named pytorch_model.bin, tf_model.h5, model.ckpt or flax_model.`
​
​
From the answer online, I thought that I should download the model files directly to my local. However, I could not find anything how to run models on directly local, without pulling them from HF with code. What should I do?
​
| 2023-09-11T10:10:30 |
https://www.reddit.com/r/LocalLLaMA/comments/16frh5n/how_to_use_models_on_local_wout_downloading_with/
|
JavaMaster420
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16frh5n
| false | null |
t3_16frh5n
|
/r/LocalLLaMA/comments/16frh5n/how_to_use_models_on_local_wout_downloading_with/
| false | false |
self
| 1 | null |
What is the best LLM for every weight?
| 1 |
What is the best LLM in the 1b variant as for 3b and 7b 13b 30b and 70b? and why not 256m and 512m 768m too.
| 2023-09-11T10:47:33 |
https://www.reddit.com/r/LocalLLaMA/comments/16fs3ip/what_is_the_best_llm_for_every_weight/
|
Puzzleheaded_Acadia1
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 1 |
{'gid_2': 1}
|
16fs3ip
| false | null |
t3_16fs3ip
|
/r/LocalLLaMA/comments/16fs3ip/what_is_the_best_llm_for_every_weight/
| false | false |
self
| 1 | null |
70B build at $4000
| 1 |
Can anyone recommend the hardware for running a 70B model (maybe \`garage-bAInd/Platypus2-70B-instruct\`)? Are 2x3090s my best bet?
Memory (RAM) seems cheap enough so I'd think to throw 256 GB at it? Is there a point when enough is enough?
Does CPU speed make a big difference or is it purely GPU bound?
| 2023-09-11T11:06:35 |
https://www.reddit.com/r/LocalLLaMA/comments/16fsg44/70b_build_at_4000/
|
flemhans
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 1 |
{'gid_2': 1}
|
16fsg44
| false | null |
t3_16fsg44
|
/r/LocalLLaMA/comments/16fsg44/70b_build_at_4000/
| false | false |
self
| 1 | null |
Using RTX-6000 to fine-tune Vicuna
| 1 |
I am trying to fine-tune vicuna 7b following [this guide](https://github.com/lm-sys/FastChat) but I get a value error stating it is only possible using A-series or V-series gpu.
My question is, do I need these type of GPU to fine tune my model or is there any work-around?
| 2023-09-11T12:24:51 |
https://www.reddit.com/r/LocalLLaMA/comments/16ftxkj/using_rtx6000_to_finetune_vicuna/
|
insane-defaults
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16ftxkj
| false | null |
t3_16ftxkj
|
/r/LocalLLaMA/comments/16ftxkj/using_rtx6000_to_finetune_vicuna/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': '8gdDInq_TmmDruxFNNtiNzd7qvBrHHzZ_cTL2Iryz5c', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=108&crop=smart&auto=webp&s=fc5d9e1183c2dd198b19bb9bea9537ef0c7f0898', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=216&crop=smart&auto=webp&s=6e41e48c4b1e6365acb3dbf53212b1352bb7be44', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=320&crop=smart&auto=webp&s=d8ef26860475e11896b17c53190941a4ff949735', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=640&crop=smart&auto=webp&s=5ee4a426477824ab8bcc97d8df554d9312cbe932', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=960&crop=smart&auto=webp&s=d54fb00c816d54e0bdf92bc88116673d32a35f95', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?width=1080&crop=smart&auto=webp&s=6f1d770aa0c42734b4ab0e953f8f80daacc56604', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/dppW8DVYXMmkPpSi7A_E3WPesqHS5TOVGQAIwtkFmuU.jpg?auto=webp&s=606ff585400cb39017efe7f03b5f326465248af2', 'width': 1200}, 'variants': {}}]}
|
OmniQuant of Falcon-180B has been released!
| 1 | ERROR: type should be string, got " https://github.com/OpenGVLab/OmniQuant\n\n \nNews\n\n\n[2023/09] 🔥 We have expanded support for Falcon. OmniQuant efficiently compresses Falcon-180b from 335G to 65G, with minimal performance loss. Furthermore, this compression allows for Falcon-180b inference on a single A100 80GB GPU. For details, refer to [runing_falcon180b_on_single_a100_80g](https://github.com/OpenGVLab/OmniQuant/blob/main/runing_falcon180b_on_single_a100_80g.ipynb).\n\n\nhttps://i.imgur.com/11HKigM.png" | 2023-09-11T12:33:36 |
https://www.reddit.com/r/LocalLLaMA/comments/16fu45d/omniquant_of_falcon180b_has_been_released/
|
ittu
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fu45d
| false | null |
t3_16fu45d
|
/r/LocalLLaMA/comments/16fu45d/omniquant_of_falcon180b_has_been_released/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'GzPPByNW1RDV8D8rRvkNFHFjESaZGE4vOXGk9NCS_nI', 'resolutions': [{'height': 14, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=108&crop=smart&auto=webp&s=6c543155daf3777c34fdbdc5ea06812d3d6acb46', 'width': 108}, {'height': 28, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=216&crop=smart&auto=webp&s=f617060b4fa0af1bf2f94ebd3f8690c68a015f71', 'width': 216}, {'height': 42, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=320&crop=smart&auto=webp&s=d5528033f85531d595ee19f22681dfd68d47de6e', 'width': 320}, {'height': 85, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=640&crop=smart&auto=webp&s=8b8610a329b0b9b279e8c3e7d6631521533739c5', 'width': 640}, {'height': 128, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=960&crop=smart&auto=webp&s=f6bf81ed3999d6d8a86cd2f649156db732ea9b6f', 'width': 960}, {'height': 144, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?width=1080&crop=smart&auto=webp&s=d68a62a2ce0666c20cf0f8ea3421aa0146766f34', 'width': 1080}], 'source': {'height': 289, 'url': 'https://external-preview.redd.it/xEw8GsZu9zkRWYNrK5ZT6WYVsxrokFic8WYY2hDTXn0.png?auto=webp&s=b01255093a55c6e649f7317fe659ab1d24e9e431', 'width': 2155}, 'variants': {}}]}
|
Integrating llama 2 in word processor?
| 6 |
I was wondering if there is any way to integrate Llama 2 with a word processor, such as Microsoft Word or Google Docs, so that I can use it to help write and fleah out documents. I think it would be very helpful to have Llama 2 as a writing assistant that can generate content, suggest improvements, or check grammar and spelling.
I have used llama locally for some time, but only in chatlike settings, This is a far cry from an integrated assistant I imagine is possible. I searched and couldn't find any project doing this.
So, does anyone know if there is any integration of Llama 2 with a word processor? And if not, why not? Is it because of technical difficulties, or lack of demand? I would love to hear your thoughts and opinions on this topic. Thanks in advance!
| 2023-09-11T13:06:59 |
https://www.reddit.com/r/LocalLLaMA/comments/16futml/integrating_llama_2_in_word_processor/
|
EmbarrassedIce9048
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16futml
| false | null |
t3_16futml
|
/r/LocalLLaMA/comments/16futml/integrating_llama_2_in_word_processor/
| false | false |
self
| 6 | null |
How To Develop A Token Streaming UI For Your LLaMA model With Go, FastAPI And JS
| 1 |
I have been struggling with token streaming for a while. Now that I have something solid I thought I would share how I did it because I couldn't find so many useful resources about it on the web...
Basically I needed to build a nice interface for my LLaMA models, allowing me to see the text showing up dynamically on the screen (aka "token streaming"). Same as what we can see on the ChatGPT UI. This is pretty useful because these language models can be so slow that waiting for the whole response to be ready can be a pain...
As I am expecting quite a lot of users to be playing with the UI concurrently, I thought that Go would be a good choice (and I haven't been disappointed so far).
So basically my stack is the following:
* Creating an "Event Source" in the browser in Javascript in order to read Server Sent Events (SSE)
* Render the HTML page with Go, read the user request, forward it to the language model backend, and forward the streamed tokens to the JS frontend with SSE as soon as they arrive
* Deploy the language model with Python and a framework allowing for token streaming like Hugging Face Transformers
* Add a small FastAPI interface on top of the large language model in order to communicate with the Go frontend
Here is the detailed how-to: [https://nlpcloud.com/how-to-develop-a-token-streaming-ui-for-your-llm-with-go-fastapi-and-js.html](https://nlpcloud.com/how-to-develop-a-token-streaming-ui-for-your-llm-with-go-fastapi-and-js.html?utm_source=reddit&utm_campaign=i859w625-3816-81ed-a261-0242ac140019)
It works very well, but maybe there exists an even better setup? In that case I would love to hear your suggestions!
Julien
| 2023-09-11T13:09:38 |
https://www.reddit.com/r/LocalLLaMA/comments/16fuvn8/how_to_develop_a_token_streaming_ui_for_your/
|
juliensalinas
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fuvn8
| false | null |
t3_16fuvn8
|
/r/LocalLLaMA/comments/16fuvn8/how_to_develop_a_token_streaming_ui_for_your/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'wlI6wBz5sfDYABqSyw01yUccWpHrhKSmweaoFT-WH_g', 'resolutions': [{'height': 75, 'url': 'https://external-preview.redd.it/e51Abb4zSfrt-dfj0Ra5LgfHVrzz6XRX9aE-5wDXt1U.jpg?width=108&crop=smart&auto=webp&s=4fbc4c914252035a68763934e2d5991b0146765b', 'width': 108}, {'height': 151, 'url': 'https://external-preview.redd.it/e51Abb4zSfrt-dfj0Ra5LgfHVrzz6XRX9aE-5wDXt1U.jpg?width=216&crop=smart&auto=webp&s=19c2d28802ad13441ca3f1710043d3d2a2add35d', 'width': 216}, {'height': 224, 'url': 'https://external-preview.redd.it/e51Abb4zSfrt-dfj0Ra5LgfHVrzz6XRX9aE-5wDXt1U.jpg?width=320&crop=smart&auto=webp&s=b4be43254c7e7d38476ca54a1c1266383f89e772', 'width': 320}], 'source': {'height': 421, 'url': 'https://external-preview.redd.it/e51Abb4zSfrt-dfj0Ra5LgfHVrzz6XRX9aE-5wDXt1U.jpg?auto=webp&s=bc7dff0c9d4602528d596bd20f171aadfaae7da4', 'width': 600}, 'variants': {}}]}
|
Are PCI-E 4.0 x16 and PCI-E 4.0 x4 good enough for 2 GPUs to run LLMs?
| 1 |
My motherboard is MSI X670P Wifi which has a PCI-E 4.0 x16 and a PCI-E 4.0 x4. I read somewhere that you need at least x8 speed to run LLMs. Could anyone confirm this? Any advice is also welcome!
| 2023-09-11T13:12:01 |
https://www.reddit.com/r/LocalLLaMA/comments/16fuxl1/are_pcie_40_x16_and_pcie_40_x4_good_enough_for_2/
|
tgredditfc
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fuxl1
| false | null |
t3_16fuxl1
|
/r/LocalLLaMA/comments/16fuxl1/are_pcie_40_x16_and_pcie_40_x4_good_enough_for_2/
| false | false |
self
| 1 | null |
Noob question, how to begin? Questions for the a first time training/running.
| 12 |
Hi!
I´ve been reading and seeing some videos. However I still have some doubts about the best path and any tips would be greatly appreciated.
1. What do you guys think is the easy model to run at first and them try some training on local data?
2. I would like to train the model using several research paper´s that I will use as reference, lab reports, and my thesis. Thus the model would be able, to respond questions based on it. Does anyone know a guide/tutorial?
3. I´ve a 4090 and 128GB of ram. In terms of OS ( Windows, linux, docker,...) what is the easy way, do you guys think to begin, and learn?
4. Do you guys know any compilation of available models and characteristics?
​
Thank you guys.
| 2023-09-11T14:24:58 |
https://www.reddit.com/r/LocalLLaMA/comments/16fwndq/noob_question_how_to_begin_questions_for_the_a/
|
No_One_BR
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fwndq
| false | null |
t3_16fwndq
|
/r/LocalLLaMA/comments/16fwndq/noob_question_how_to_begin_questions_for_the_a/
| false | false |
self
| 12 | null |
ContentatScale and Winston AI as alternatives to Turnitin AI
| 1 |
As the title says, can these tools be used as alternatives to Turnitin's AI detector? Can anyone speak with any previous experience regarding this? Thank you.
| 2023-09-11T14:25:32 |
https://www.reddit.com/r/LocalLLaMA/comments/16fwnxa/contentatscale_and_winston_ai_as_alternatives_to/
|
psj_2908
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fwnxa
| false | null |
t3_16fwnxa
|
/r/LocalLLaMA/comments/16fwnxa/contentatscale_and_winston_ai_as_alternatives_to/
| false | false |
self
| 1 | null |
Does CPU bottleneck GPU in GPTQ models?
| 1 |
Will a i7-4770 with an RTX 3090 bottleneck the GPU using GPTQ models in textgen?
​
| 2023-09-11T14:29:11 |
https://www.reddit.com/r/LocalLLaMA/comments/16fwqz7/does_cpu_bottleneck_gpu_in_gptq_models/
|
Imagummybear23
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fwqz7
| false | null |
t3_16fwqz7
|
/r/LocalLLaMA/comments/16fwqz7/does_cpu_bottleneck_gpu_in_gptq_models/
| false | false |
self
| 1 | null |
The 4 Essential Dataset Types for LLMs: A Deep Dive
| 1 |
[removed]
| 2023-09-11T15:58:09 |
https://www.reddit.com/r/LocalLLaMA/comments/16fz0qb/the_4_essential_dataset_types_for_llms_a_deep_dive/
|
l33thaxman
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fz0qb
| false | null |
t3_16fz0qb
|
/r/LocalLLaMA/comments/16fz0qb/the_4_essential_dataset_types_for_llms_a_deep_dive/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'wl5u9lohlRi8mmgYaIC60x2EUTbSOKC6HdZOnzkjtww', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/OD_NLm9D4wFz0ynPHin3ue0jsjubjVgfpmYRXjyMGDM.jpg?width=108&crop=smart&auto=webp&s=31ddc4662e5530704537cddcf8045c12e59faa6e', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/OD_NLm9D4wFz0ynPHin3ue0jsjubjVgfpmYRXjyMGDM.jpg?width=216&crop=smart&auto=webp&s=5186ce9cf29bc3dfdf400ed3d2fddbeb10977337', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/OD_NLm9D4wFz0ynPHin3ue0jsjubjVgfpmYRXjyMGDM.jpg?width=320&crop=smart&auto=webp&s=14eeda42ea736899c16e09b489d12980fcc46fc2', 'width': 320}], 'source': {'height': 360, 'url': 'https://external-preview.redd.it/OD_NLm9D4wFz0ynPHin3ue0jsjubjVgfpmYRXjyMGDM.jpg?auto=webp&s=bb979a453ce82bc9534e9ba52e9c446a8e938405', 'width': 480}, 'variants': {}}]}
|
Serving qlora fine-tuned models in production
| 1 |
vLLM does not support qlora in production. What are the available approaches at the moment until quantization support is added to vLLM?
| 2023-09-11T16:03:07 |
https://www.reddit.com/r/LocalLLaMA/comments/16fz5q8/serving_qlora_finetuned_models_in_production/
|
ComplexIt
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16fz5q8
| false | null |
t3_16fz5q8
|
/r/LocalLLaMA/comments/16fz5q8/serving_qlora_finetuned_models_in_production/
| false | false |
self
| 1 | null |
H2OGPT not saving documents in database
| 1 |
Hi, I'm new to this so I may be doing this wrong but on the Windows 11 version of H2OGPT I can upload my documents and create a collection in LangChain Mode-Path. However whenever I restart the documents disappear.
This is a pain as I am trying to analyse my own documents. I've clicked on 'Update DB with new/changed files on disc'.
Also does it matter if I add documents before or after I load a model? I've tried it with no model and with a model loaded.
Speaking of which is it possible to set a default model using the one-click installer of H2OGPT?
| 2023-09-11T17:57:09 |
https://www.reddit.com/r/LocalLLaMA/comments/16g26qi/h2ogpt_not_saving_documents_in_database/
|
ScriptReaderStudio
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g26qi
| false | null |
t3_16g26qi
|
/r/LocalLLaMA/comments/16g26qi/h2ogpt_not_saving_documents_in_database/
| false | false |
self
| 1 | null |
New way to speed up inference! Easier than speculative decoding
| 1 |
“Medusa adds extra "heads" to LLMs to predict multiple future tokens simultaneously. When augmenting a model with Medusa, the original model stays untouched, and only the new heads are fine-tuned during. During generation, these heads each produce multiple likely words for the corresponding position. These options are then combined and processed using a tree-based attention mechanism. Finally, a typical acceptance scheme is employed to pick the longest plausible prefix from the candidates for further decoding.”
https://github.com/FasterDecoding/Medusa
| 2023-09-11T17:58:18 |
https://www.reddit.com/r/LocalLLaMA/comments/16g27s0/new_way_to_speed_up_inference_easier_than/
|
big_ol_tender
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g27s0
| false | null |
t3_16g27s0
|
/r/LocalLLaMA/comments/16g27s0/new_way_to_speed_up_inference_easier_than/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'ZaGgACbs-Ed1psgFwxZV06yT9YK_x40PNcljpwPPnN4', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=108&crop=smart&auto=webp&s=2b63bdf51487d798c8c81d77af6c523565e197d9', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=216&crop=smart&auto=webp&s=afb991406b4e1bbfc6e4c15e09b8d03ba5eb1da1', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=320&crop=smart&auto=webp&s=6b479ff8a4829eef4c41e9d579a1b2d8bcff489d', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=640&crop=smart&auto=webp&s=e6d55a279b69eecd3dd410370fc14e1e6ddd81fe', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=960&crop=smart&auto=webp&s=723556a3d3f7c2f2d7d2545b9e5a0fd071a97ceb', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?width=1080&crop=smart&auto=webp&s=7de6cd9d7d31111169025c6404837035d962b9a6', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/wpcJ4kaM6KKhxPUIkFfOGsa5ACGVURFmlt2O5ia8PqM.jpg?auto=webp&s=6a1c084782fca21e545e7f3a7d76f6dad864830c', 'width': 1200}, 'variants': {}}]}
|
Making a chatbot for my wife's mom
| 1 |
My MIL passed away almost two years ago, and she and my wife were very close. They chatted quite a bit every day on Facebook messenger for more than a decade, and I'd like to make a bot version. (I have already talked to my wife about this and gotten approval.)
I have access to the raw chat logs, which are in html format. I wanted to make sure I understood the workflow for finetuning LLaMA. Answers to the specific questions below and generat tips are appreciated!
1. I have an M2, but my understanding is it's hard to do finetuning locally on Macs now. So I imagine my best bet is to use Colab with autotrain. The final command would be something like `!autotrain llm --train --project_name MILbot --model meta-llama/Llama-2-13b-chat-hf --data_path . --lora_r=32 --lora_alpha=64 --model_max_length=4096 --text_column text --use_peft --use_int8 --learning_rate 2e-4 --train_batch_size 2 --num_train_epochs 5 --block_size 4096 --trainer sft --push_to_hub --repo_id myname/myrepo`
2. Should I get the data into a csv or json file? I think csv for autotrain, but I'm not sure.
3. Is 13b-chat my best bet here? Also, should I use the sft trainer or something else?
4. I want to make sure I'm processing the data correctly, and I'm kind of confused about this part. My understanding is I should break it up into roughly 4k token chunks. And each chunk should have something like the following.
`<s>[INST] <<SYS>> You are playing the role of the mother [insert details] <</SYS>> Hi, mom! [/INST] Hi, how are you? </s><s>[INST] I'm good. [/INST] Great </s><s>`
| 2023-09-11T18:13:46 |
https://www.reddit.com/r/LocalLLaMA/comments/16g2mx7/making_a_chatbot_for_my_wifes_mom/
|
eumaximizer
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g2mx7
| false | null |
t3_16g2mx7
|
/r/LocalLLaMA/comments/16g2mx7/making_a_chatbot_for_my_wifes_mom/
| false | false |
self
| 1 | null |
GGUF.js - open-source JS library (with types) for parsing and reading metadata of ggml-based gguf files.
| 1 | 2023-09-11T18:28:53 |
https://github.com/ahoylabs/gguf.js
|
719Ben
|
github.com
| 1970-01-01T00:00:00 | 0 |
{}
|
16g31mz
| false | null |
t3_16g31mz
|
/r/LocalLLaMA/comments/16g31mz/ggufjs_opensource_js_library_with_types_for/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'A9wzb3e5XrBjNIDBenzLtzoFwGMOL7_OnnybFWABZLw', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=108&crop=smart&auto=webp&s=6736eb169a426bb2535fa804be8dd12d1a472edc', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=216&crop=smart&auto=webp&s=dad2ac53b9e7927aeef2f7f5d569680795d8fa73', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=320&crop=smart&auto=webp&s=fc722fcb0dbedf53339f485af3d696adb356b98c', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=640&crop=smart&auto=webp&s=fdd652b798b4fc20b84fed5b07e48a570711e03c', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=960&crop=smart&auto=webp&s=395b4a6e26ab6c48f1159a328cdda68a85749202', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?width=1080&crop=smart&auto=webp&s=f4c3439b2204c9e41507130af3ae14092c43da1a', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/4Ym9B9n7nm2mkyrxSZ-TeyesRQaC6Zx8IgycsW_x-Qw.jpg?auto=webp&s=1c48b696b7f996df789ccdbb9a3f29a6ea30ec6b', 'width': 1200}, 'variants': {}}]}
|
||
Can you run anything on. 4070 ti with 64GB ram?
| 1 |
Just curious
| 2023-09-11T18:42:26 |
https://www.reddit.com/r/LocalLLaMA/comments/16g3evj/can_you_run_anything_on_4070_ti_with_64gb_ram/
|
livekop
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g3evj
| false | null |
t3_16g3evj
|
/r/LocalLLaMA/comments/16g3evj/can_you_run_anything_on_4070_ti_with_64gb_ram/
| false | false |
self
| 1 | null |
Weird Response
| 1 |
So, I used the bloke's 13b 128k model. I used prompt Alpaca-with-input. In the instruction I added context of a document then the q and a below was my result??? Not sure what to do.....
### Input:
Will you list the steps in a easy to read format?
### Response:
1031cm1agtOtO01the(OetOetO5thedtO61thO01750tOtO071thetO76tOpro1etOtO5tOtOtOcce1tO1tOtOthetO01tO1p13501tO5411tOtO1tOcfctO7m0c01e1e01tO0tO1tO7411112c0
e101e0211tO
6300002122tO0022122c6
tthe014
c71113001tO4
c1m5c2c2
| 2023-09-11T18:47:52 |
https://www.reddit.com/r/LocalLLaMA/comments/16g3k3x/weird_response/
|
Leading-Leading6718
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g3k3x
| false | null |
t3_16g3k3x
|
/r/LocalLLaMA/comments/16g3k3x/weird_response/
| false | false |
self
| 1 | null |
Raspberry Pi 8GB running TinyLlama, can anyone report the user experience?
| 1 |
I suck and my unit has not shipped, and it's burning a hole in my brain. I was just wondering if anyone has, or could, try to run the new TinyLlama 1.1B model on the beefy 8GB RAM raspberry pi and let us all know what that is like? Shameful bribery be damned, you'll have my upvote!
| 2023-09-11T18:48:22 |
https://www.reddit.com/r/LocalLLaMA/comments/16g3kla/raspberry_pi_8gb_running_tinyllama_can_anyone/
|
Actual-Bad5029
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g3kla
| false | null |
t3_16g3kla
|
/r/LocalLLaMA/comments/16g3kla/raspberry_pi_8gb_running_tinyllama_can_anyone/
| false | false |
self
| 1 | null |
What local client can I use to load a local Llama 2 70B model and then send prompts to it via a python script and get a return as a string?
| 1 |
I use oobabooga but I need a client that will let me make my own interface using streamlit and load models myself via an API
| 2023-09-11T19:03:33 |
https://www.reddit.com/r/LocalLLaMA/comments/16g3zvy/what_local_client_can_i_use_to_load_a_local_llama/
|
countrycruiser
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g3zvy
| false | null |
t3_16g3zvy
|
/r/LocalLLaMA/comments/16g3zvy/what_local_client_can_i_use_to_load_a_local_llama/
| false | false |
self
| 1 | null |
GPT user here - what’s the benefit of using these localized models?
| 1 |
Are there specific things?
| 2023-09-11T19:19:17 |
https://www.reddit.com/r/LocalLLaMA/comments/16g4f5l/gpt_user_here_whats_the_benefit_of_using_these/
|
livekop
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g4f5l
| false | null |
t3_16g4f5l
|
/r/LocalLLaMA/comments/16g4f5l/gpt_user_here_whats_the_benefit_of_using_these/
| false | false |
self
| 1 | null |
different VRAM in cards
| 1 |
I have a 4090 and a 6000 Ada with a total of 72Gb VRAM (24Gb + 48Gb)
Can I sum the VRAM of the two cards for training and inferencing or it will be limited to 24 + 24 ?
| 2023-09-11T20:13:47 |
https://www.reddit.com/r/LocalLLaMA/comments/16g5w9m/different_vram_in_cards/
|
Dry_Honeydew9842
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g5w9m
| false | null |
t3_16g5w9m
|
/r/LocalLLaMA/comments/16g5w9m/different_vram_in_cards/
| false | false |
self
| 1 | null |
Issue with RAG implementations (h2ogpt, localgpt, etc)
| 1 |
[removed]
| 2023-09-11T20:21:30 |
https://www.reddit.com/r/LocalLLaMA/comments/16g644d/issue_with_rag_implementations_h2ogpt_localgpt_etc/
|
xIndirect
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g644d
| false | null |
t3_16g644d
|
/r/LocalLLaMA/comments/16g644d/issue_with_rag_implementations_h2ogpt_localgpt_etc/
| false | false |
self
| 1 | null |
How to use multiple GPU's with textgen webui?
| 1 |
HI, total noob here, I have a machine which has two nvidia GPU's with 22 GB of VRAM each, which should in total be enough to load a 13b param model, but textgen webui only uses GPU 0 without trying to use GPU 1, so I run into this issue:
​
https://preview.redd.it/8boh4yrgvonb1.png?width=1169&format=png&auto=webp&s=946895709bd5af60b7934a36bd837f7c2cd0a65e
How can I make textgen webui run using both GPUs? Thanks
| 2023-09-11T20:49:38 |
https://www.reddit.com/r/LocalLLaMA/comments/16g6w5b/how_to_use_multiple_gpus_with_textgen_webui/
|
Milk_No_Titties
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g6w5b
| false | null |
t3_16g6w5b
|
/r/LocalLLaMA/comments/16g6w5b/how_to_use_multiple_gpus_with_textgen_webui/
| false | false | 1 | null |
|
NVidia vGPU on esx
| 1 |
Was having a browse today and somehow slipped down a rabbit hole. I wondered if anyone has tried a virtualized esxi instance with nVidia vGPUs? It seems that esxi is capable, with the right drivers and support, of merging multiple cards into a single resource. Obviously there will be some overhead, but with a good native driver this could be minimised.
https://docs.nvidia.com/grid/latest/grid-software-quick-start-guide/
Only applies to select GPUs I realise,.bit could potentially allow for larger single device vram sizes ( along with increased tensors being spread over multiple cards ). Would be interested if anyone has had any experience?
| 2023-09-11T22:42:23 |
https://www.reddit.com/r/LocalLLaMA/comments/16g9y94/nvidia_vgpu_on_esx/
|
BreakIt-Boris
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16g9y94
| false | null |
t3_16g9y94
|
/r/LocalLLaMA/comments/16g9y94/nvidia_vgpu_on_esx/
| false | false |
self
| 1 | null |
ggPrompt: explore questions and topics generated and structured by AI
| 1 |
Hey everyone, I've made a new site called ggPrompt.org.
[From Philosophy to Pythagorean Metempsychosis](https://preview.redd.it/38s28vzlmpnb1.png?width=1928&format=png&auto=webp&s=d685d010b98c448c8fd1734bf959336305f94481)
It helps you come up with questions and explore new scientific topics quickly (similar to those wikipedia rabbit holes).
[https://ggprompt.org/](https://ggprompt.org/)
It's still in an early beta so I'd really appreciate any feedback or thoughts you have.
Tags and structure are generated by GPT4 while some of the prompts are a mix of GPT4 and OpenSource models. I have many ideas, so the site will continue to grow in both quality and quantity.
Happy digging!
| 2023-09-11T23:28:50 |
https://www.reddit.com/r/LocalLLaMA/comments/16gb490/ggprompt_explore_questions_and_topics_generated/
|
vmirnv
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gb490
| false | null |
t3_16gb490
|
/r/LocalLLaMA/comments/16gb490/ggprompt_explore_questions_and_topics_generated/
| false | false | 1 | null |
|
Does a guanaco 65b gguf exist?
| 1 |
[removed]
| 2023-09-11T23:47:20 |
https://www.reddit.com/r/LocalLLaMA/comments/16gbjrt/does_a_guanaco_65b_gguf_exist/
|
wh33t
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gbjrt
| false | null |
t3_16gbjrt
|
/r/LocalLLaMA/comments/16gbjrt/does_a_guanaco_65b_gguf_exist/
| false | false |
self
| 1 | null |
Is it worth it to return my 3080 and get a used 3090?
| 1 |
I'm a first year cs major and I know for certain that I want to get involved with AI. I watched some videos on neural networks (including the 3blue1brown series on them) and I got hooked on the way that pure math and computer code is used to transform textual inputs into recognizable outputs in a way that mimics "real" intelligence. I recently built a new gaming PC with a 3080 10gb, but now I'm not sure if it will be enough to experiment with existing models or even train my own. I could afford the $600-700 for a used 3090 with its 24gb of vram, but only if I know that it will actually provide me with significantly more value than my 3080. Should I return my 3080 and get the 3090?
TLDR: Want to learn everything I can about neural networks, but only have a 3080 with 10gb of vram. Is it worth it to upgrade to a 3090 if my goal is to experiment and learn about AI?
| 2023-09-12T00:57:41 |
https://www.reddit.com/r/LocalLLaMA/comments/16gd53d/is_it_worth_it_to_return_my_3080_and_get_a_used/
|
InteractionQuiet9169
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gd53d
| false | null |
t3_16gd53d
|
/r/LocalLLaMA/comments/16gd53d/is_it_worth_it_to_return_my_3080_and_get_a_used/
| false | false |
self
| 1 | null |
Can fine tuning remove censorship/alignment from ChatGPT models?
| 1 |
[removed]
| 2023-09-12T01:39:51 |
https://www.reddit.com/r/LocalLLaMA/comments/16ge3l1/can_fine_tuning_remove_censorshipalignment_from/
|
NickDifuze
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16ge3l1
| false | null |
t3_16ge3l1
|
/r/LocalLLaMA/comments/16ge3l1/can_fine_tuning_remove_censorshipalignment_from/
| false | false |
self
| 1 | null |
Is Nvidia finally going to have some competition in the generative AI space?
| 1 | 2023-09-12T01:53:16 |
https://www.d-matrix.ai/
|
onil_gova
|
d-matrix.ai
| 1970-01-01T00:00:00 | 0 |
{}
|
16gee9d
| false | null |
t3_16gee9d
|
/r/LocalLLaMA/comments/16gee9d/is_nvidia_finally_going_to_have_some_competition/
| false | false |
default
| 1 | null |
|
Code llama, is it good for pure C programming?
| 1 |
I understand pure C is indeed supported, but I wonder if it is well supported.
In particular, could it beat github copilot?
From the review I gathered, it seems to focus more on C++, python and Javascript. I wonder if it is only because there aren't as many pure C programmers these days so I don't get to see their reviews.
| 2023-09-12T02:24:47 |
https://www.reddit.com/r/LocalLLaMA/comments/16gf37h/code_llama_is_it_good_for_pure_c_programming/
|
Studying_Man
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gf37h
| false | null |
t3_16gf37h
|
/r/LocalLLaMA/comments/16gf37h/code_llama_is_it_good_for_pure_c_programming/
| false | false |
self
| 1 | null |
Phi-1.5: 41.4% HumanEval in 1.3B parameters (model download link in comments)
| 6 | 2023-09-12T03:57:44 |
https://arxiv.org/abs/2309.05463
|
ethanhs
|
arxiv.org
| 1970-01-01T00:00:00 | 0 |
{}
|
16gh0yv
| false | null |
t3_16gh0yv
|
/r/LocalLLaMA/comments/16gh0yv/phi15_414_humaneval_in_13b_parameters_model/
| false | false | 6 |
{'enabled': False, 'images': [{'id': 'q3evP6JeDpAC2MdSQHWYxnCYTqbJkElIQsLFqVSdkss', 'resolutions': [{'height': 63, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=108&crop=smart&auto=webp&s=2711d572cfc6c713893cf24e8c4a7344d5ad8a4c', 'width': 108}, {'height': 126, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=216&crop=smart&auto=webp&s=b6624f0c1eedc14997e7f1780efbe6e5cb50c1e2', 'width': 216}, {'height': 186, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=320&crop=smart&auto=webp&s=9db38144ef3065833b9ba158c764f7be47de3016', 'width': 320}, {'height': 373, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=640&crop=smart&auto=webp&s=72b056142e7533b5628a2a34f37f7e5415727075', 'width': 640}, {'height': 560, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=960&crop=smart&auto=webp&s=2637f961ee21190172b9ca6c8adf3ac9612db083', 'width': 960}, {'height': 630, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?width=1080&crop=smart&auto=webp&s=782eead871df2939a587ee3beae442cc59282f64', 'width': 1080}], 'source': {'height': 700, 'url': 'https://external-preview.redd.it/0HhwdU6MKIAKjL9Y8-B_iH374a3NiPTy0ib8lmloRzA.jpg?auto=webp&s=f1cd025aeb52ffa82fc9e5a4a2f157da0d919147', 'width': 1200}, 'variants': {}}]}
|
||
Super basic questions about H2OGPT, Models, etc. from a noob
| 1 |
[removed]
| 2023-09-12T04:04:09 |
https://www.reddit.com/r/LocalLLaMA/comments/16gh5tb/super_basic_questions_about_h2ogpt_models_etc/
|
consig1iere
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gh5tb
| false | null |
t3_16gh5tb
|
/r/LocalLLaMA/comments/16gh5tb/super_basic_questions_about_h2ogpt_models_etc/
| false | false |
self
| 1 | null |
Microsoft research releases phil-1.5 a 1.3B model trained on 7B tokens. 55.5% at human eval
| 1 |
[removed]
| 2023-09-12T04:26:00 |
m477h13U
|
i.redd.it
| 1970-01-01T00:00:00 | 0 |
{}
|
16ghkt8
| false | null |
t3_16ghkt8
|
/r/LocalLLaMA/comments/16ghkt8/microsoft_research_releases_phil15_a_13b_model/
| false | false | 1 |
{'enabled': True, 'images': [{'id': 'ZL7GGrNoPu-5NFoGZmntbBykSgAPx9P_7bA--nqhaYI', 'resolutions': [{'height': 49, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=108&crop=smart&auto=webp&s=e5f892229bb55ce4713a537be2269aa3cec8c435', 'width': 108}, {'height': 99, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=216&crop=smart&auto=webp&s=61698b710365b22ea617e89fd6fdf881651152a5', 'width': 216}, {'height': 147, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=320&crop=smart&auto=webp&s=d5289124b00f112b3fe0c67bb3df334c8d1a2e80', 'width': 320}, {'height': 295, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=640&crop=smart&auto=webp&s=f113e7b3a8c07aa36795b3542c7bb52757363d7f', 'width': 640}, {'height': 443, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=960&crop=smart&auto=webp&s=c36515a54054c4d0d02ff6cb4e7e73eb83f68390', 'width': 960}, {'height': 498, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?width=1080&crop=smart&auto=webp&s=bb771a5a029c48f757a5c10bb50e081bfc91758f', 'width': 1080}], 'source': {'height': 1080, 'url': 'https://preview.redd.it/qcljm2gy4rnb1.jpg?auto=webp&s=2d02cca87826afc584d96411d83173212f48c142', 'width': 2340}, 'variants': {}}]}
|
||
Why Do the LLaMA Models Need So Many Parameters for Good Results When Stable Diffusion Needs So Few?
| 4 |
Forgive me if this seems like an ignorant question, since I am myself quite ignorant to machine learning in general, but I found something strange.
The Stable Diffusion models can generally run quite well on a 8 GB VRAM GPU and can generate really good looking results. However, it seems that many of the LLaMA language models and their fine tunes that can run on a 8 GB VRAM GPU might not generate the best results.
Maybe this is just ignorance on my part, but I wonder why a smaller model trained on images, which I think are really dense representations of information, tend to generate visually appealing results and why language models, in which words might not encode as much information as the images, tend to require more parameters to train.
Again, I apologize for potentially asking a stupid questions, and maybe this question didn’t even make sense, but any leads might be quite nice. Thank you for your time.
| 2023-09-12T04:39:31 |
https://www.reddit.com/r/LocalLLaMA/comments/16ghtw6/why_do_the_llama_models_need_so_many_parameters/
|
AlterandPhil
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16ghtw6
| false | null |
t3_16ghtw6
|
/r/LocalLLaMA/comments/16ghtw6/why_do_the_llama_models_need_so_many_parameters/
| false | false |
self
| 4 | null |
How to resume fine tuning with the autotrain-advanced utils?
| 1 |
I had a fine tuning session earlier today that completed and wish to restart to pick up where it left off and continue to train. But judging from the loss rate it's starting all over again. How do I force autotrain to resume a fine tuning?
| 2023-09-12T05:27:59 |
https://www.reddit.com/r/LocalLLaMA/comments/16gip5f/how_to_resume_fine_tuning_with_the/
|
SiliconObsessed
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gip5f
| false | null |
t3_16gip5f
|
/r/LocalLLaMA/comments/16gip5f/how_to_resume_fine_tuning_with_the/
| false | false |
self
| 1 | null |
Any local model good with instructions and/or function calling?
| 1 |
Could someone pls help me out? :)
| 2023-09-12T05:30:28 |
https://www.reddit.com/r/LocalLLaMA/comments/16giqp9/any_local_model_good_with_instructions_andor/
|
schmedu
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16giqp9
| false | null |
t3_16giqp9
|
/r/LocalLLaMA/comments/16giqp9/any_local_model_good_with_instructions_andor/
| false | false |
self
| 1 | null |
Are any of the models that equal/beat GPT 3.5 price competitive right now? (cloud)
| 5 |
Saw a few posts about models like Llama 2 70b beating GPT 3.5 but being far more expensive to run due to ChatGPT being subsidized by Microsoft.
Wondering if that's still true or if there's any model we can run online that is price competitive in terms of cost per token?
What about locally if we want to need to use a 10M-1B tokens over a few months?
(Edit: In terms of performance, I've found GPT 3.5 4K context works pretty well for my needs, but gpt4 is far better but not worth the extra cost.)
| 2023-09-12T05:57:57 |
https://www.reddit.com/r/LocalLLaMA/comments/16gj7cl/are_any_of_the_models_that_equalbeat_gpt_35_price/
|
Ill_Fox8807
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gj7cl
| false | null |
t3_16gj7cl
|
/r/LocalLLaMA/comments/16gj7cl/are_any_of_the_models_that_equalbeat_gpt_35_price/
| false | false |
self
| 5 | null |
Best data format for passing data to llam 2 using llama index
| 1 |
I have html pages that contain text, images, hyperlinks and tables. What would be the best way to send it to llama 2 using llama index, so that the model understands the data properly.
| 2023-09-12T06:25:08 |
https://www.reddit.com/r/LocalLLaMA/comments/16gjnxm/best_data_format_for_passing_data_to_llam_2_using/
|
zaid-70
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gjnxm
| false | null |
t3_16gjnxm
|
/r/LocalLLaMA/comments/16gjnxm/best_data_format_for_passing_data_to_llam_2_using/
| false | false |
self
| 1 | null |
What is the most novel ways you used llms?
| 1 |
Title.
What is the most creative or novel ways you used llm?
​
I know the common usecases:
\- Text distallation e.g. Summurization and Information Extraction
\- Text transformation e.g. translation and rewriting
\- Text expansion e.g. Brainstorming, generating new content
\- Chatbot QA and RP
\- Self Prompting, like AutoGPT
​
​
For what else is it used out there?
| 2023-09-12T07:20:27 |
https://www.reddit.com/r/LocalLLaMA/comments/16gkkoc/what_is_the_most_novel_ways_you_used_llms/
|
Astronos
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gkkoc
| false | null |
t3_16gkkoc
|
/r/LocalLLaMA/comments/16gkkoc/what_is_the_most_novel_ways_you_used_llms/
| false | false |
self
| 1 | null |
WizardCoder python 34 q8 results
| 1 |
Just want to share my 0 professional result with 3x3090 and wizard coder
For me this model with llama.cpp is really good for coding and it write 200 lines script without problem.
I don't use any humaneval etc benchmark for test it. I just work with it like usually I work with chatgpt.
IMAO the wizard is writing more modern code and now more trics.
Tldr: WizardCoder generate useful code for me. It's my winner 🏆
Can't wait for 70b version
So
| 2023-09-12T08:15:33 |
https://v.redd.it/x5bvcp5v9snb1
|
Nondzu
|
/r/LocalLLaMA/comments/16glhup/wizardcoder_python_34_q8_results/
| 1970-01-01T00:00:00 | 0 |
{}
|
16glhup
| false |
{'reddit_video': {'bitrate_kbps': 5000, 'dash_url': 'https://v.redd.it/x5bvcp5v9snb1/DASHPlaylist.mpd?a=1697184935%2CN2Q3ODIzZDE5Y2VhNmZlZjk3ZjU1OTcyOWFlOWU3NTFmMzUxMGI5MTc1ZmFkYmMyZWIxOGRhZGE5YWVlYTM1Ng%3D%3D&v=1&f=sd', 'duration': 17, 'fallback_url': 'https://v.redd.it/x5bvcp5v9snb1/DASH_1080.mp4?source=fallback', 'height': 1080, 'hls_url': 'https://v.redd.it/x5bvcp5v9snb1/HLSPlaylist.m3u8?a=1697184935%2CNDQ3YzMzZTllNGI3ZWJkNTFhYjY5OWJlMTgzMDZiNDYwYjgzOTFkZjMzZGJkMTkxYmIyNWRjZTUzOGU4ZmU3Yw%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/x5bvcp5v9snb1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 1920}}
|
t3_16glhup
|
/r/LocalLLaMA/comments/16glhup/wizardcoder_python_34_q8_results/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5', 'resolutions': [{'height': 60, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=108&crop=smart&format=pjpg&auto=webp&s=c6e213e24e3c43edc7b3ac477c0bc1988fb7922f', 'width': 108}, {'height': 121, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=216&crop=smart&format=pjpg&auto=webp&s=66d123302567f674e6b8a97aecf93b6237399bb2', 'width': 216}, {'height': 179, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=320&crop=smart&format=pjpg&auto=webp&s=6573c6fe79e20120b5c2222a3d3f128da934f7df', 'width': 320}, {'height': 359, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=640&crop=smart&format=pjpg&auto=webp&s=a9f566be49ba0e724516ae243d17ee0f907d83cc', 'width': 640}, {'height': 539, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=960&crop=smart&format=pjpg&auto=webp&s=82191f946d2aa79866600fbe4c3f735bf09b5f2a', 'width': 960}, {'height': 607, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?width=1080&crop=smart&format=pjpg&auto=webp&s=300857bda79a0d6d77ec17c4b0a32a973b16242d', 'width': 1080}], 'source': {'height': 607, 'url': 'https://external-preview.redd.it/bjh3bHY2MXY5c25iMQofC0sK07iJQ9mdY6w694dw5FJz9NQXGUiCxNnPTek5.png?format=pjpg&auto=webp&s=ada50d06e1d4eaac3c48ad64b4c9e850fece0ef5', 'width': 1080}, 'variants': {}}]}
|
|
WizardCoder python 34 q8 results
| 1 | 2023-09-12T08:19:08 |
https://v.redd.it/nrpiniphasnb1
|
Nondzu
|
/r/LocalLLaMA/comments/16gljzw/wizardcoder_python_34_q8_results/
| 1970-01-01T00:00:00 | 0 |
{}
|
16gljzw
| false |
{'reddit_video': {'bitrate_kbps': 5000, 'dash_url': 'https://v.redd.it/nrpiniphasnb1/DASHPlaylist.mpd?a=1697185151%2CYzg4MGI3NGFmNDIzYTIwZjlmNzFjZjc4ZWQ5MGEwMDFjNmYwYjgyNTMwYWRmM2U5NjY2ZTc5NTRlMWZlNzJmMQ%3D%3D&v=1&f=sd', 'duration': 22, 'fallback_url': 'https://v.redd.it/nrpiniphasnb1/DASH_1080.mp4?source=fallback', 'height': 1920, 'hls_url': 'https://v.redd.it/nrpiniphasnb1/HLSPlaylist.m3u8?a=1697185151%2CMTlmZjljM2JiM2I2MTcwYWI0NTMxOGEyYjg1YmRlMTk0MjAzMjAyNzFjYjJkMWUwMGM1NGIwYjBhYzYzY2QwOQ%3D%3D&v=1&f=sd', 'is_gif': False, 'scrubber_media_url': 'https://v.redd.it/nrpiniphasnb1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 1080}}
|
t3_16gljzw
|
/r/LocalLLaMA/comments/16gljzw/wizardcoder_python_34_q8_results/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH', 'resolutions': [{'height': 192, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?width=108&crop=smart&format=pjpg&auto=webp&s=e0c16cf2edf3fc6aabc64e8330431e189428d7a0', 'width': 108}, {'height': 384, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?width=216&crop=smart&format=pjpg&auto=webp&s=e05413992f0d1ea7fe8a655ac6f67551861f2c4e', 'width': 216}, {'height': 569, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?width=320&crop=smart&format=pjpg&auto=webp&s=c05044f1b376b948893d4b6c3c9bf3828c3076ec', 'width': 320}, {'height': 1138, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?width=640&crop=smart&format=pjpg&auto=webp&s=2a52e98691070acf3b779ddf1cfe2cd8ef30ede2', 'width': 640}, {'height': 1708, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?width=960&crop=smart&format=pjpg&auto=webp&s=3d94bacdcacc37030b62311538031f4692b98a59', 'width': 960}], 'source': {'height': 1790, 'url': 'https://external-preview.redd.it/dGVleDZla2hhc25iMXClJ8BFqJrL2EJPWHRD0PV0-k53p-iNkQs1rOUCLQxH.png?format=pjpg&auto=webp&s=52273fa7d393ded2c5b9dd854761d46a0a2693f7', 'width': 1006}, 'variants': {}}]}
|
||
License status of semantic search models with respect to licenses for training datasets.
| 1 |
Ok, so most of semantic search models from this leaderboard:
[https://huggingface.co/spaces/mteb/leaderboard](https://huggingface.co/spaces/mteb/leaderboard)
or this leaderboard:
[https://paperswithcode.com/sota/zero-shot-text-search-on-beir](https://paperswithcode.com/sota/zero-shot-text-search-on-beir)
are trained on data that are only licensed for noncommercial use - such as MS MARCO.
[https://microsoft.github.io/msmarco/](https://microsoft.github.io/msmarco/)
Meanwhile models are often license on MIT, Apache or other similar free license.
Does the license for the data affect the model license?
Are there any cases or legal studies dealing with issue?
**Can we use commercially licensed model trained on noncommercially licensed data for commercial purposes without legal risks?**
| 2023-09-12T08:34:59 |
https://www.reddit.com/r/LocalLLaMA/comments/16glt6k/license_status_of_semantic_search_models_with/
|
FormerIYI
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16glt6k
| false | null |
t3_16glt6k
|
/r/LocalLLaMA/comments/16glt6k/license_status_of_semantic_search_models_with/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'CCVJBt0-kH9o-QPgo7qiP6d0ggaejrGSkWh3JVtDcDI', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=108&crop=smart&auto=webp&s=849a3fa499b884c2b8710447c5fa7e81ea1d62dd', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=216&crop=smart&auto=webp&s=065a3d9f73e3d30917818732d3b52bd258701450', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=320&crop=smart&auto=webp&s=f84b8e1967e9366f08b71df8df7d28872547b0f9', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=640&crop=smart&auto=webp&s=c516ab5717dc5f8c117a7c3f6c737b5b76da054b', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=960&crop=smart&auto=webp&s=bfc377360bf629e87d8f599bdab20402ab507979', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?width=1080&crop=smart&auto=webp&s=6275c2854aeb552e46a627255c27691406e6de3b', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/iB9eNi3rxIJpvd7zLjet2NAGGdmv8p6wAhdFyU5XkOE.jpg?auto=webp&s=306c3d89e946b9aa6841c52fedf2738bc28f2689', 'width': 1200}, 'variants': {}}]}
|
15 comparisons with 3-bit, 4-bit, 5-bit, 6-bit and 8-bit to test how quantisation affects model output
| 1 |
[https://rentry.org/quants](https://rentry.org/quants) I did 15 basic comparisons with GGML 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit and GPTQ 4-bit to test how quants change responses. Model is Vicuna 33b, and no coding questions cause its bad with coding
The tests aren't super thorough but it helped me settle on a quant, Q8 is good but slow and Q5\_K\_M is what I'd use. I dunno why but Q6\_K did badly. I used TGI's debug-deterministic for greedy decoding so any change in the output is from quant differences
| 2023-09-12T09:13:18 |
https://www.reddit.com/r/LocalLLaMA/comments/16gmfwd/15_comparisons_with_3bit_4bit_5bit_6bit_and_8bit/
|
GrapeCharacter2746
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gmfwd
| false | null |
t3_16gmfwd
|
/r/LocalLLaMA/comments/16gmfwd/15_comparisons_with_3bit_4bit_5bit_6bit_and_8bit/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'b1E8sI-kTet-3YOFKrYAUVQ9ABbay60W7WEBpTM34S8', 'resolutions': [{'height': 108, 'url': 'https://external-preview.redd.it/sld18DTFrG0vgxxCwlYEGWKS7hjSXmQsKHgNjUEATAk.jpg?width=108&crop=smart&auto=webp&s=5f7d74321748816977c2c47d74607125fd510a17', 'width': 108}, {'height': 216, 'url': 'https://external-preview.redd.it/sld18DTFrG0vgxxCwlYEGWKS7hjSXmQsKHgNjUEATAk.jpg?width=216&crop=smart&auto=webp&s=9c08000e015b470c7d577334237c7dee99c37847', 'width': 216}, {'height': 320, 'url': 'https://external-preview.redd.it/sld18DTFrG0vgxxCwlYEGWKS7hjSXmQsKHgNjUEATAk.jpg?width=320&crop=smart&auto=webp&s=628b4e1ef982e336b9ee2da5dbacecc2774b6d65', 'width': 320}], 'source': {'height': 512, 'url': 'https://external-preview.redd.it/sld18DTFrG0vgxxCwlYEGWKS7hjSXmQsKHgNjUEATAk.jpg?auto=webp&s=9e4cec75ef4248064a481db4ef5f29637aec6e67', 'width': 512}, 'variants': {}}]}
|
eGPU to increase VRAM capacity
| 1 |
I've been exploring locally run LLMs recently (as a completely non-technical novice) and I'm looking for ways to expand VRAM capacity to load larger models without the need to substantially reconfigure my existing set up (4090 + 7950x3d + 64gb RAM).
I'm considering getting an eGPU to potentially run a 4080 or another 4090 in parallel with my main system, but had a few queries before diving into it, as I can't see that many other people have done this (perhaps for good reason?):
* Has anyone had any experience doing this? How smooth was the experience?
* Would adding a 2nd GPU in this manner effectively combine the VRAM or are there limitations I should be aware of?
I'm trying to avoid rebuilding my entire PC since I also use it for other non-AI tasks, so finding a way to just expand my VRAM would be convenient.
| 2023-09-12T09:47:51 |
https://www.reddit.com/r/LocalLLaMA/comments/16gn0ij/egpu_to_increase_vram_capacity/
|
TheCunningBee
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gn0ij
| false | null |
t3_16gn0ij
|
/r/LocalLLaMA/comments/16gn0ij/egpu_to_increase_vram_capacity/
| false | false |
self
| 1 | null |
Finetuning a model for world lore knowledge
| 1 |
My brother and I are working on a project aimed at "teaching" base llama2 7b on a broad unstructured dataset of Skyrim lore to create a base model for a Skyrim NPC that is grounded in the
world.
Step one is to compile a few text files of scraped information from the wiki and train a LoRA in oobabooga via colab. We've made good progress, and have the datasets available, but we're struggling to work out which parameters we should be modifying in order for the model to really learn this knowledge without overfitting. It quickly follows the style and tone of the data, however the information it spits out when prompted with a specific piece of data (e.g. a location) is always incorrect (even feeding half a sample exactly, it'll continue with believable sounding information, however it doesnt align with the training data).
Worth also mentioning that we're aiming to then finetune for conversation on top of this model after we're happy with the first step.
We'd be super curious to get some ideas on what parameters would be most effective for this task.
E.g., should we be aiming for 5+ epochs, or aiming to keep it low?
<512 rank or more? (I've heard people suggest both, so I'm a little stumped)
What kind of batch size is recommended?
It's important the model learns the information in the data for conversation later. The idea being that You can ask a character what Markarth is, and they will have enough information available to describe the city accurately.
We're new to fine-tuning, so any pointers or help would be appreciated!
| 2023-09-12T09:54:50 |
https://www.reddit.com/r/LocalLLaMA/comments/16gn4h7/finetuning_a_model_for_world_lore_knowledge/
|
Goatman117
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gn4h7
| false | null |
t3_16gn4h7
|
/r/LocalLLaMA/comments/16gn4h7/finetuning_a_model_for_world_lore_knowledge/
| false | false |
self
| 1 | null |
Llama2 7b fine tuning on Sentiment analysis
| 1 |
Hello, I am trying to finetune llama 7b on some data extraction from the user input. I did some finetuning for summarisation and it works fine. Now I am trying to extract some info about keyword such as sentiments and the output would be a Sentiment and line describe the sentiment exactly as it is. The format would be like this
"###Input: Some user input###Keyword: SPA ###Output: Topic(name[KEYWORD], sentiment[SENTIMENT], line[line describing that sentiment]) <|end|>"
But after finetuning 3000 steps, I see the loss decreasing but when tested It output lines which are not in the User input. What could be the reason, is it the 7B model limitation or I need to train more steps.
| 2023-09-12T10:07:49 |
https://www.reddit.com/r/LocalLLaMA/comments/16gncpp/llama2_7b_fine_tuning_on_sentiment_analysis/
|
Intelligent-Fan-2461
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gncpp
| false | null |
t3_16gncpp
|
/r/LocalLLaMA/comments/16gncpp/llama2_7b_fine_tuning_on_sentiment_analysis/
| false | false |
self
| 1 | null |
How fast will quantized Falcon-180B run on system RAM?
| 1 |
Has anyone tried it yet? Could you tell me how many tokens per second or seconds per token u get?
​
| 2023-09-12T11:12:28 |
https://www.reddit.com/r/LocalLLaMA/comments/16goink/how_fast_will_quantized_falcon180b_run_on_system/
|
Prince-of-Privacy
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16goink
| false | null |
t3_16goink
|
/r/LocalLLaMA/comments/16goink/how_fast_will_quantized_falcon180b_run_on_system/
| false | false |
self
| 1 | null |
LLM Recommendation: Don't sleep on Synthia!
| 1 |
I'm currently working on another in-depth LLM comparison after my previous [test of 13 models](https://www.reddit.com/r/LocalLLaMA/comments/15lihmq/big_model_comparisontest_13_models_tested/) and [test of 7 more models](https://www.reddit.com/r/LocalLLaMA/comments/15ogc60/new_model_rp_comparisontest_7_models_tested/) - this time it's 20 models, so it takes a while... But I can't wait any longer because one model has proven to be so good that I just need to talk about it now!
> SynthIA (Synthetic Intelligent Agent) is a LLama-2-70B model trained on Orca style datasets. It has been fine-tuned for instruction following as well as having long-form conversations.
>
> All Synthia models are uncensored. Please use it with caution and with best intentions. You are responsible for how you use Synthia.
That's from the [model cards on Hugging Face](https://huggingface.co/migtissera) (there are multiple versions as the author keeps updating it). Sounds good, so I tried it ([TheBloke/Synthia-70B-v1.2-GGUF](https://huggingface.co/TheBloke/Synthia-70B-v1.2-GGUF) Q4_0), and after using it extensively for a few days now, it's become my new favorite model.
Why? Its combination of intelligence and personality (and even humor) surpassed all the other models I tried, which include Airoboros, Chronos-Hermes, Llama 2 Chat, MythoMax, Nous Hermes, Nous Puffin, and Samantha. Especially the latter has also been praised for its personality and intelligence, but Samantha is censored worse than Llama 2 Chat, and while I can get her to do NSFW roleplay, she's too moralizing and needs constant coercion, that's why I consider her too annoying to bother with (I already have my wife to argue or fight with, don't need an AI for that! ;)). Synthia has shown at least as much intelligence and personality, and she's uncensored, so she's always fun to talk to and very easy-going no matter the topic or theme.
So after my previous favorites Nous Hermes and MythoMax, now it's Synthia. But the reason I'm so excited about this model is not just that it's become my latest favorite for entertainment purposes, no, today I actually tried it for work-related purposes (write shell scripts, Kubernetes and Terraform manifests, install and debug software, etc.) - and it worked much better than expected, even when compared to GPT-4 which I used to cross-reference my answers (here's just one example of [Synthia 70B v1.2 (Q4_0) vs. GPT-4](https://imgur.com/a/G24okTK)).
Until now, I must admit that I had considered local LLMs just for entertainment purposes - for work, I'd simply use ChatGPT or GPT-4. But the intelligence Synthia exhibited in chat and roleplay made me curious, so I tried it for work, and now I start to see the potential.
Anyway, I've not seen this model mentioned a lot - in fact, searching for it here, there was only one mention of it so far. I needed to post this to change that because I've tested so many models and this one has truly surprised me very positively. I'll post the detailed evaluation results of the other models once I'm done with all the tests, but for now, I had to post this because of my sincere excitement right now.
**TL;DR:** Try **Synthia** for chat, roleplay, and even work!
By the way, there's a newer [v1.2b](https://huggingface.co/migtissera/Synthia-70B-v1.2b) that still needs quantization by u/The-Bloke. And there are smaller 13B and even 7B versions, which I haven't tested extensively so can't speak of their quality, but if 70B is too big or too slow for you, I recommend you give those a try.
| 2023-09-12T11:15:30 |
https://www.reddit.com/r/LocalLLaMA/comments/16gokoa/llm_recommendation_dont_sleep_on_synthia/
|
WolframRavenwolf
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gokoa
| false | null |
t3_16gokoa
|
/r/LocalLLaMA/comments/16gokoa/llm_recommendation_dont_sleep_on_synthia/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'C5XErO93I36-NbN-scAGZWB9fcwUctaccaqpnQtxC8g', 'resolutions': [{'height': 131, 'url': 'https://external-preview.redd.it/WdqJYPDT1O2AxCe7OTgiv-F1QFNXtDggQpXVTCxylm4.jpg?width=108&crop=smart&auto=webp&s=088d2af408fab95b88cdc33d9127d2c0c7617beb', 'width': 108}, {'height': 262, 'url': 'https://external-preview.redd.it/WdqJYPDT1O2AxCe7OTgiv-F1QFNXtDggQpXVTCxylm4.jpg?width=216&crop=smart&auto=webp&s=bd744c011a1d6455369981bfb2ac7ebe00153991', 'width': 216}, {'height': 388, 'url': 'https://external-preview.redd.it/WdqJYPDT1O2AxCe7OTgiv-F1QFNXtDggQpXVTCxylm4.jpg?width=320&crop=smart&auto=webp&s=9cd77afe4ab2ffb5a22740f37a69559dea46b16f', 'width': 320}, {'height': 777, 'url': 'https://external-preview.redd.it/WdqJYPDT1O2AxCe7OTgiv-F1QFNXtDggQpXVTCxylm4.jpg?width=640&crop=smart&auto=webp&s=53f6820edc19ac7aaf15a2ff9110051b35d9e57b', 'width': 640}], 'source': {'height': 954, 'url': 'https://external-preview.redd.it/WdqJYPDT1O2AxCe7OTgiv-F1QFNXtDggQpXVTCxylm4.jpg?auto=webp&s=53393b2b4dad1afc6420cf2ee5294173787c4a20', 'width': 785}, 'variants': {}}]}
|
AskCyph™ LITE: Run AI Model in Brower (Redpajama, Llama2 7B)
| 1 |
Hello LLama Enthusiasts,
Have you ever wished for AI that’s truly personal and private?
Introducing [AskCyph™ LITE](https://askcyph.cypherchat.app/), a lightweight AI chatbot that runs AI Models directly in your browser. The first time it takes time to download the model, the next time it is a lot faster to initialize.
Currently, we support Red Pajama (Basic) and Llama 2 7b (Advanced). We created this as a way for enthusiasts of all levels to take a plunge and have an AI model running.
A bit about us. We are the creator of [CypherChat](https://cypherchat.app/), a privacy-focused messaging platform that does not rely on personal information and offers End-to-End Encrypted (E2EE) and Peer-to-Peer (P2P) communication. We recognize the need for privacy.
Note [AskCyph™ LITE](https://askcyph.cypherchat.app/), is decoupled from CypherChat and does not require you to sign up so it is a click away for you to try.
✅ Offline access
✅ Enhanced privacy
✅ Basic or Advanced Models
✅ Requires 4GB/8GB free RAM
✅ Relatively new Computer with integrated or external GPU
​
Our vision is about making AI accessible to everyone, ensuring security, and providing a unique experience. It's still experimental but growing. Those who are recently getting into AI with a non-technical background can also say they have their model running. We value your feedback and questions!
We do want to give a shout-out and acknowledge community projects that inspired us and made innovation and AskCyph™ LITE possible.
* [Hugging Face](https://huggingface.co/)
* [Apache TVM](https://tvm.apache.org/)
* [MLC AI - Web LLM](https://webllm.mlc.ai/)
* [TOGETHER](https://together.ai/)
* [Llama2](https://ai.meta.com/llama/)
\#AIForAll #AskCyphLITE #CypherChat
| 2023-09-12T11:44:27 |
https://www.reddit.com/r/LocalLLaMA/comments/16gp5t6/askcyph_lite_run_ai_model_in_brower_redpajama/
|
cypherchat
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gp5t6
| false | null |
t3_16gp5t6
|
/r/LocalLLaMA/comments/16gp5t6/askcyph_lite_run_ai_model_in_brower_redpajama/
| false | false |
self
| 1 | null |
Qwen models removed by their authors ?
| 1 |
[removed]
| 2023-09-12T12:02:14 |
https://www.reddit.com/r/LocalLLaMA/comments/16gpjja/qwen_models_removed_by_their_authors/
|
Bogdahnfr
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gpjja
| false | null |
t3_16gpjja
|
/r/LocalLLaMA/comments/16gpjja/qwen_models_removed_by_their_authors/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': '4ydK36A-ivNKrrTlN5TOZ4gwbyWOvLMU6vjhwDE6ni8', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=108&crop=smart&auto=webp&s=85bee086a8a4adf5b6313887258ca51e8030bdbd', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=216&crop=smart&auto=webp&s=7f3c32fa6946e018ed1bded412c9b767351aec77', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=320&crop=smart&auto=webp&s=ae6c16f242252cb7c2ad3471f7ce0c051f534556', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=640&crop=smart&auto=webp&s=77f084f6d32ef2d6af135d2b4940f60ffeace0de', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=960&crop=smart&auto=webp&s=73d35be537a81a086a170974ab4bb6ba2be63654', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?width=1080&crop=smart&auto=webp&s=bd66814abc8f47b3154130c2bf26e8d0b176c0bb', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/PKiozV9PvpTCseDqQxJv7-s14UHUYb9eM9y8Z5uMtkQ.jpg?auto=webp&s=ca292edde219bc0ee23cedb4a10cec397683b907', 'width': 1200}, 'variants': {}}]}
|
best coding llama model?
| 1 |
so far, whats the best coding companion? i can run up to 34b readily. Im looking for multi-lingual preferably for general purpose, but definitely want it to be c# capable.
| 2023-09-12T12:14:50 |
https://www.reddit.com/r/LocalLLaMA/comments/16gptaf/best_coding_llama_model/
|
Nekasus
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gptaf
| false | null |
t3_16gptaf
|
/r/LocalLLaMA/comments/16gptaf/best_coding_llama_model/
| false | false |
self
| 1 | null |
Mojo 🔥
| 1 |
If you haven’t heard, Mojo is a new programming language that combines the ease of Python with the performance of C. It’s written specifically for AI.
Here is an example project that should excite us all (not mine btw):
https://github.com/tairov/llama2.mojo
| 2023-09-12T12:24:23 |
https://www.reddit.com/r/LocalLLaMA/comments/16gq09y/mojo/
|
Tough_Performer6101
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gq09y
| false | null |
t3_16gq09y
|
/r/LocalLLaMA/comments/16gq09y/mojo/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'EvKKRuTWKjxzXehjCXwzOaTMLxuJtPJl3wtgviFFVpM', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=108&crop=smart&auto=webp&s=4408cf806f4860d14c3df8f4b3e5fa58ed8358b4', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=216&crop=smart&auto=webp&s=f88c9aadcab47ff617b032a60928feadb44c7c7d', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=320&crop=smart&auto=webp&s=8595a34f7a74c2b40214b59c19bfc3facd273251', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=640&crop=smart&auto=webp&s=25d2d9decae4245db64e56ec15db079bc25afb88', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=960&crop=smart&auto=webp&s=c216b6ba2f26daf1300bb782f1ca2718fbf35715', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?width=1080&crop=smart&auto=webp&s=9e82bf7100529ca9683a4d5afa7d13e68ccd2500', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/G5SYLSHSDbcMNZuhVPNqtEFhGWFsfDYY7KcZZTs5nR4.jpg?auto=webp&s=83a879882ebe91aa7ba2e08562d21c0d1d6642a5', 'width': 1200}, 'variants': {}}]}
|
Exllama V2 has dropped!
| 1 | 2023-09-12T12:27:32 |
https://github.com/turboderp/exllamav2
|
a_beautiful_rhind
|
github.com
| 1970-01-01T00:00:00 | 0 |
{}
|
16gq2gu
| false | null |
t3_16gq2gu
|
/r/LocalLLaMA/comments/16gq2gu/exllama_v2_has_dropped/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'CC1TwWCFAVWOyki4LM9sNnBqJjQOKr1C1yxTKpke4PE', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=108&crop=smart&auto=webp&s=738a7c687307a6a537705f836a3d8bbc7758219a', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=216&crop=smart&auto=webp&s=d2b87f821f36ee08022190fced407b9c7d4d2933', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=320&crop=smart&auto=webp&s=5f16b647214400731750c6b66568b2fbf7cbb9b1', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=640&crop=smart&auto=webp&s=09c7d8641bce107b773b16a92876d36a3cc3955d', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=960&crop=smart&auto=webp&s=2efd8b3d752e0d0b41c0378f6ec076706e0858f6', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?width=1080&crop=smart&auto=webp&s=0e57741adb20a9e85aec8b8b3faa6099082c0a5a', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/pxge0nuWVu2mvf8OhCDz3mEmXIAuYy9YJAhKElo_T0c.jpg?auto=webp&s=931ed7d3398a15ec00a1ac8b30290bde775b848d', 'width': 1200}, 'variants': {}}]}
|
||
Converting Instructor Embeddings to Onnx. Issues with T5 encoder-only model.
| 1 |
Hi,
I am trying to convert instructor-base to ONNX to quantize the model for faster CPU performance. The model is a T5 Encoder. I've made various attempts, all having similar results. The issue is that when I try to do this, it initializes encoder weights (which are nonexistent since this is a T5 Encoder model). And when building a graph, it tries to call decoder\_outputs, and throws the following error:
ValueError: You have to specify either decoder\_input\_ids or decoder\_inputs\_embeds
Again, these don't exist.
I've tried using:
\- fastT5.export\_and\_get\_onnx\_model
\- torch.onnx.export
\- txtai.pipeline.HFOnnx
Has anyone else tried this? Any advice is greatly appreciated!
| 2023-09-12T12:50:23 |
https://www.reddit.com/r/LocalLLaMA/comments/16gqjnc/converting_instructor_embeddings_to_onnx_issues/
|
GoobleGravity
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gqjnc
| false | null |
t3_16gqjnc
|
/r/LocalLLaMA/comments/16gqjnc/converting_instructor_embeddings_to_onnx_issues/
| false | false |
self
| 1 | null |
How can I use oobabooga with flowise?
| 1 |
I'm enabling api extension on oobabooga but can't connect to flowise
| 2023-09-12T12:50:45 |
https://www.reddit.com/r/LocalLLaMA/comments/16gqjxg/how_can_i_use_oobabooga_with_flowise/
|
forwatching
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gqjxg
| false | null |
t3_16gqjxg
|
/r/LocalLLaMA/comments/16gqjxg/how_can_i_use_oobabooga_with_flowise/
| false | false |
self
| 1 | null |
Best Mid-Level Build?
| 1 |
Hey all,
​
I am doing some research and looking for some help. I want to have a mid-level build to start doing my own stable diffusion and LLM fine tuning. Not full on training as that requires a ton of resources.
I was looking at the Nvidia A6000 as that has 48GB which should be enough to run some training. Is it true you should have the Xeon W processors to take advantage?
Also, was looking at a server rack like this that I can keep in my basement plugged into my router. [https://www.ebay.com/itm/175626243153](https://www.ebay.com/itm/175626243153)
Do you think this would be enough for my needs?
| 2023-09-12T13:00:20 |
https://www.reddit.com/r/LocalLLaMA/comments/16gqr9n/best_midlevel_build/
|
rbur0425
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gqr9n
| false | null |
t3_16gqr9n
|
/r/LocalLLaMA/comments/16gqr9n/best_midlevel_build/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'ckNus53wGcFtuPsDUYJY69RYKvTcrnvDc117pxYsJN0', 'resolutions': [{'height': 81, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=108&crop=smart&auto=webp&s=92817421bd0a6c5bbfd214ff6886c5bb7e65b4a5', 'width': 108}, {'height': 162, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=216&crop=smart&auto=webp&s=c9953b35f70cb7fac714fc4ecc524865901a48f6', 'width': 216}, {'height': 240, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=320&crop=smart&auto=webp&s=1dced3f15d4f65f9adfaa48a4fe062323f52d6fa', 'width': 320}, {'height': 480, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=640&crop=smart&auto=webp&s=dde6899d783645965c81973df100e4e4378042be', 'width': 640}, {'height': 720, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=960&crop=smart&auto=webp&s=7e39f1cd174646ec0cfa8197288cc84e6da6acfd', 'width': 960}, {'height': 810, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?width=1080&crop=smart&auto=webp&s=d6adc82718742b0344e49e687467e65aa4dc109a', 'width': 1080}], 'source': {'height': 1125, 'url': 'https://external-preview.redd.it/Anlrq7i00THxwvlQr0pVXnTnNU9l-SE1tOugKBwUMPA.jpg?auto=webp&s=edccc227f98d63a6e59b8c51104653940903f2fa', 'width': 1500}, 'variants': {}}]}
|
What is the best model I can run on-prem that is compatible with Langchain?
| 1 |
I am trying to build a system with a chromadb and and llm and was hoping to use langchain for ease. It seems like I cannot run LLAMA 2 anytime soon since I am using LangChain. I can rent any EC2 I want, I just don't know which one and what model I should use. Accuracy/quality is the biggest thing for me.
​
| 2023-09-12T13:45:43 |
https://www.reddit.com/r/LocalLLaMA/comments/16grt87/what_is_the_best_model_i_can_run_onprem_that_is/
|
Suitable-Ad-8598
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16grt87
| false | null |
t3_16grt87
|
/r/LocalLLaMA/comments/16grt87/what_is_the_best_model_i_can_run_onprem_that_is/
| false | false |
self
| 1 | null |
So, uh, Mythomax-70b showed up yesterday
| 1 |
[https://huggingface.co/lloorree/mythomax-70b](https://huggingface.co/lloorree/mythomax-70b)
Didn't see any posts about it, but I tried it and it's living up to the name. Only gripes is that it only has 4k context and that it doesn't handle nonstandard dialogue very well (though that is a lot of models.) But it seems to have extremely good coherence, world knowledge, and prompt adherence, and it's rare to find all three in the same model.
| 2023-09-12T14:11:24 |
https://www.reddit.com/r/LocalLLaMA/comments/16gsfp4/so_uh_mythomax70b_showed_up_yesterday/
|
tenmileswide
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gsfp4
| false | null |
t3_16gsfp4
|
/r/LocalLLaMA/comments/16gsfp4/so_uh_mythomax70b_showed_up_yesterday/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'fjpAVvOKyIgXJKcYe-MW2t1yiFocrEaWlxcGh_zTCJU', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=108&crop=smart&auto=webp&s=9e0b88d8abe0f7528ec174810bca8b5add082cfa', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=216&crop=smart&auto=webp&s=bae7dedd2066e4025a01f1a5f83b3132d7958ee7', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=320&crop=smart&auto=webp&s=b17137a72e87558521cca411242de9a383d9c551', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=640&crop=smart&auto=webp&s=0f864e2614e1696410798a30c7a6d0534fd130f5', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=960&crop=smart&auto=webp&s=0a302ba8b68064f31d205892b131f71b271285af', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?width=1080&crop=smart&auto=webp&s=22cacaf033b7bb10e15f5e1a71488837ee79a5ab', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/6nXsxc5KjOBbDN6l893u6E2qZNNA8UfXXtTWi-gh1-I.jpg?auto=webp&s=1e5abe1147d6497284c95fe7c6d07d4fa9a86e31', 'width': 1200}, 'variants': {}}]}
|
How do you run Llama 2 on multi GPUs?
| 1 |
I got access to a few GPUs and wanted to try my hand at deploying a large LLM on 2 GPUs. At first I tried the 7B model on two 3060s using huggingface+accelerate code but it kept giving me an OOM error. Same thing with a 3090.
So how do you run these models on multi GPUs?
Edit: Maybe it is an issue with the gpu service I am using? Maybe their GPUs are not set up properly? I am using valdi.ai
| 2023-09-12T14:16:09 |
https://www.reddit.com/r/LocalLLaMA/comments/16gsjs9/how_do_you_run_llama_2_on_multi_gpus/
|
soham1996
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gsjs9
| false | null |
t3_16gsjs9
|
/r/LocalLLaMA/comments/16gsjs9/how_do_you_run_llama_2_on_multi_gpus/
| false | false |
self
| 1 | null |
Deploying Llama 2 in any cloud with Python API
| 1 | 2023-09-12T14:17:55 |
https://dstack.ai/examples/python-api/
|
cheptsov
|
dstack.ai
| 1970-01-01T00:00:00 | 0 |
{}
|
16gslb9
| false | null |
t3_16gslb9
|
/r/LocalLLaMA/comments/16gslb9/deploying_llama_2_in_any_cloud_with_python_api/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'Kk5_6cq8ODsGTK0N1L02ZBr2TKBzHScfNlQn15DzOfM', 'resolutions': [{'height': 56, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=108&crop=smart&auto=webp&s=9a2d195ced901855ed37f117c3df4d3561cc243f', 'width': 108}, {'height': 113, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=216&crop=smart&auto=webp&s=e3ce55c7cb48e5844593de306592a52c6cccb264', 'width': 216}, {'height': 168, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=320&crop=smart&auto=webp&s=981036be8c8e5f02dca08f8b28981573f6d5ce54', 'width': 320}, {'height': 336, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=640&crop=smart&auto=webp&s=a0ab11161cf1773c2f62628c62497a7a5aa3b15e', 'width': 640}, {'height': 504, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=960&crop=smart&auto=webp&s=4efa70e8f4a499485e08ce2d27a644df98327088', 'width': 960}, {'height': 567, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?width=1080&crop=smart&auto=webp&s=8686a2896ff4e59e4231f2daab5d5dfa60d160f0', 'width': 1080}], 'source': {'height': 630, 'url': 'https://external-preview.redd.it/5S1IClMnFkSSqVPks-PlJ5w8-Vl7ZN_VV0kjsLza2Zg.jpg?auto=webp&s=8342fb2aae3cd1a0798143b91b38887b26d10248', 'width': 1200}, 'variants': {}}]}
|
||
What's the best use case for phi-1 (~1bn param GPT3.5)?
| 1 |
MSFT just released their Phi-1 and Phi-1.5 models which demonstrate very impressive reasoning abilities for their size, thanks to training on high-quality synthetic data. However, this approach means that these Phi models lack the extensive latent knowledge often embedded in other models.
Using them with RAG seems like a natural fit requiring minimal adjustments. Moreover, with some tweaking, they might be well-suited for roleplay or fantasy scenarios. How would you envision utilizing these models?
I'm interested in working to develop promising use cases, as some [were discussed here](https://www.reddit.com/r/LocalLLaMA/comments/16au3ga/im_convinced_now_that_personal_llms_are_going_to/).
​
This can be accomplished either with fine-tuning of Phi-1 or pre-training from scratch (I have been working on the latter this past week and have [seen some cool results](https://twitter.com/ocolegro/status/1700159878155600165)).
| 2023-09-12T14:52:04 |
https://www.reddit.com/r/LocalLLaMA/comments/16gtfmo/whats_the_best_use_case_for_phi1_1bn_param_gpt35/
|
docsoc1
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gtfmo
| false | null |
t3_16gtfmo
|
/r/LocalLLaMA/comments/16gtfmo/whats_the_best_use_case_for_phi1_1bn_param_gpt35/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'u38mQ-Nq8c9Xe_odMINXxDi4tElFD6j9FqN81b86Y6w', 'resolutions': [{'height': 20, 'url': 'https://external-preview.redd.it/IsGM_jdSAzTVAiinm8jtQ2gWHTTs9Ihem9Ji95Rs8Tc.jpg?width=108&crop=smart&auto=webp&s=0c648d924c7a99c0534d56a0b65db5a7fd21d8ed', 'width': 108}], 'source': {'height': 26, 'url': 'https://external-preview.redd.it/IsGM_jdSAzTVAiinm8jtQ2gWHTTs9Ihem9Ji95Rs8Tc.jpg?auto=webp&s=3df8a8baa1bd1d7c740e20e96256ba73c0ff9f89', 'width': 140}, 'variants': {}}]}
|
airoboros/spicyboros 2.2
| 1 |
Hi all,
The new airoboros 2.2 and spicyboros 2.2 models are all uploaded to HF (some are still being quantized by the legendary bloke).
* [https://huggingface.co/jondurbin/airoboros-l2-70b-2.2](https://huggingface.co/jondurbin/airoboros-l2-70b-2.2)
* [https://huggingface.co/jondurbin/airoboros-l2-13b-2.2](https://huggingface.co/jondurbin/airoboros-l2-13b-2.2)
* [https://huggingface.co/jondurbin/airoboros-l2-7b-2.2](https://huggingface.co/jondurbin/airoboros-l2-7b-2.2)
* [https://huggingface.co/jondurbin/spicyboros-70b-2.2](https://huggingface.co/jondurbin/spicyboros-70b-2.2)
* [https://huggingface.co/jondurbin/spicyboros-c34b-2.2](https://huggingface.co/jondurbin/spicyboros-c34b-2.2)
* [https://huggingface.co/jondurbin/spicyboros-13b-2.2](https://huggingface.co/jondurbin/spicyboros-13b-2.2)
* [https://huggingface.co/jondurbin/spicyboros-7b-2.2](https://huggingface.co/jondurbin/spicyboros-7b-2.2)
* (airoboros codellama-34 will be uploaded soon, likely tomorrow morning)
**airoboros vs spicyboros?**
spicyboros is the ~~uncensored~~ *actively de-censored* version. It will still likely have some refusals, and bias inherited from the base llama-2 model, but is more likely to produce "harmful" content. The de-alignment data was not super comprehensive, but seems to have worked well, if you're into that sort of thing.
The de-alignment dataset includes a small amount of comedy, horror stories, llm-enhanced (nsfw) reddit stories, etc., so it will produce less PG content (when asked). The 7b model can be a bit... unhinged, so be careful with that one.
I removed all de-alignment data from airoboros models, so you may find them more censored than previous versions - use the spicy version instead if you want an uncensored model.
**Prompt format**
The prompt format now uses newlines instead of spaces!
A chat.
USER: {prompt}
ASSISTANT:
So, system prompt, newline, USER: {prompt} (one space after colon), newline, ASSISTANT:
There's plenty of training data that uses other names, so for a chat or RP you can replace USER/ASSISTANT with Tim:/Bob: or whatever.
The dataset includes many alternate system prompts to ensure the response are more likely to be styled by the system prompt, but I shorted the default system prompt to just "A chat."
**Dataset updates**
* enhanced "awareness" (based on system prompt) [https://huggingface.co/datasets/jondurbin/airoboros-2.2#awareness](https://huggingface.co/datasets/jondurbin/airoboros-2.2#awareness)
* text enhancement [https://huggingface.co/datasets/jondurbin/airoboros-2.2#editor](https://huggingface.co/datasets/jondurbin/airoboros-2.2#editor)
* I regenerated (almost) all of the training data that included "Once upon a time..." because it's too cliche and boring
* summarization [https://huggingface.co/datasets/jondurbin/airoboros-2.2#summarization](https://huggingface.co/datasets/jondurbin/airoboros-2.2#summarization)
* I re-created RP/GTKM data without USER/ASSISTANT tokens polluting it: [https://huggingface.co/datasets/jondurbin/airoboros-2.2#roleplayconversation](https://huggingface.co/datasets/jondurbin/airoboros-2.2#roleplayconversation)
Edit: no space after "ASSISTANT:" (ty u/WolframRavenwolf)
Edit: checkpoints (lora adapters) here:
* https://huggingface.co/jondurbin/airoboros-l2-70b-2.2-checkpoints
* https://huggingface.co/jondurbin/airoboros-c34b-2.2-checkpoints
* https://huggingface.co/jondurbin/spicyboros-70b-2.2-checkpoints
| 2023-09-12T15:17:40 |
https://www.reddit.com/r/LocalLLaMA/comments/16gu2x7/airoborosspicyboros_22/
|
JonDurbin
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gu2x7
| false | null |
t3_16gu2x7
|
/r/LocalLLaMA/comments/16gu2x7/airoborosspicyboros_22/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'u7de09cXBlRQhZUQhrj_qWKOw87SB4MEo0rlwYokI64', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=108&crop=smart&auto=webp&s=2e8f2ef8819f9a499a65d2ba807cacad91ea6b0a', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=216&crop=smart&auto=webp&s=7df33af0dcdf988948307884013122f775537e2e', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=320&crop=smart&auto=webp&s=79fded5b52ecb398a36276ce18ae30b5e80489c4', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=640&crop=smart&auto=webp&s=df0e63738bd5e3d29659a495c0c297bfc3b1fc89', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=960&crop=smart&auto=webp&s=7f99f5b88229ba38b69df5e67e408d64d472dbe1', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?width=1080&crop=smart&auto=webp&s=e20ce3c2c4eb7697c13cada6850e90dc14c01c38', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/k-LhMqJbiK8WUZaGx1snj5c1DGMHUQS1FfSX-PRuyk8.jpg?auto=webp&s=be11e877a9c9c10f0a22dee3da684034dc4315e9', 'width': 1200}, 'variants': {}}]}
|
Floneum 0.2 released: open source, local graph editor now with headless browsing, a package manager, cloud saves, and more
| 1 | 2023-09-12T15:46:13 |
https://v.redd.it/vw0u83j3wtnb1
|
ControlNational
|
v.redd.it
| 1970-01-01T00:00:00 | 0 |
{}
|
16gusz2
| false |
{'reddit_video': {'bitrate_kbps': 5000, 'dash_url': 'https://v.redd.it/vw0u83j3wtnb1/DASHPlaylist.mpd?a=1697125588%2CMTY4MDIyZmE3ODU0OTdiNzFlNGRiMTcyYzQxZGQ0NTVmZmViNGQ1MTA2YTdiNzg4NWQ1MWQwY2NlMDk0ZTM0ZQ%3D%3D&v=1&f=sd', 'duration': 8, 'fallback_url': 'https://v.redd.it/vw0u83j3wtnb1/DASH_1080.mp4?source=fallback', 'height': 1080, 'hls_url': 'https://v.redd.it/vw0u83j3wtnb1/HLSPlaylist.m3u8?a=1697125588%2CYTVkZWE0Zjc0ZGRhYzM1N2NjMjA3NmMwYjUzNWZiYjc3OWZiNTRkNTk5NTJkNWRiYWM3ZjIyZWRjNGQ0NWYzOA%3D%3D&v=1&f=sd', 'is_gif': True, 'scrubber_media_url': 'https://v.redd.it/vw0u83j3wtnb1/DASH_96.mp4', 'transcoding_status': 'completed', 'width': 1920}}
|
t3_16gusz2
|
/r/LocalLLaMA/comments/16gusz2/floneum_02_released_open_source_local_graph/
| false | false | 1 |
{'enabled': False, 'images': [{'id': '9HMbowQ1t1Vq5xNvyN--dTDRIUnFaKCJDvWZGIMvUAY', 'resolutions': [{'height': 60, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=108&crop=smart&format=pjpg&auto=webp&s=849b3439ea54f16e4dc4b13413b4f32ade686071', 'width': 108}, {'height': 121, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=216&crop=smart&format=pjpg&auto=webp&s=d00f2e41552958dd86485130ee50519fdd834279', 'width': 216}, {'height': 180, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=320&crop=smart&format=pjpg&auto=webp&s=38f80bada0dcc1be196c39b403d1f6c1990941a6', 'width': 320}, {'height': 360, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=640&crop=smart&format=pjpg&auto=webp&s=93350c4c35d6e4be26c9d0cfec07ab17f1fd9a7f', 'width': 640}, {'height': 540, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=960&crop=smart&format=pjpg&auto=webp&s=8fc53cab6394cc8ded657ff6f321022a650aa8df', 'width': 960}, {'height': 607, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?width=1080&crop=smart&format=pjpg&auto=webp&s=1430e26d5f96f1fda7f37daa6acf02f22e2b30b5', 'width': 1080}], 'source': {'height': 1080, 'url': 'https://external-preview.redd.it/n5W-pvVvl-WY_4ND1hN4fz93A0311b7LF6dPqIss244.png?format=pjpg&auto=webp&s=b7fe7352e9502e61d60899d6578a4051a2b8dc14', 'width': 1920}, 'variants': {}}]}
|
||
Dear LocalLlama, can we please stop machine shaming? Many machines can use local models. We are not all trying to train the next LLM.
| 1 |
[removed]
| 2023-09-12T16:03:00 |
https://www.reddit.com/r/LocalLLaMA/comments/16gv8sb/dear_localllama_can_we_please_stop_machine/
|
jayfehr
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gv8sb
| false | null |
t3_16gv8sb
|
/r/LocalLLaMA/comments/16gv8sb/dear_localllama_can_we_please_stop_machine/
| false | false |
self
| 1 | null |
Need for local GPU when you get cloud access?
| 1 |
In my company we want to evaluate use-cases for LLMs, for this I was assigned to develop PoCs about what are the possibilities. Our manager is looking to get us access to some Cloug-GPU-VMs, but I wondered if it would also make sense to have a laptop with a GPU, too.
For me, with a local GPU I can debug and experiment faster in quick Iterations, can debug the code with breakpoints etc. But what would make sense here?
| 2023-09-12T16:39:20 |
https://www.reddit.com/r/LocalLLaMA/comments/16gw634/need_for_local_gpu_when_you_get_cloud_access/
|
Koliham
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gw634
| false | null |
t3_16gw634
|
/r/LocalLLaMA/comments/16gw634/need_for_local_gpu_when_you_get_cloud_access/
| false | false |
self
| 1 | null |
Need help deciding on a CPU + Motherboard for running LLMs
| 1 |
Currently I already have
* Lian Li O11 Dynamic Evo
* Corsair HX1500i
* 2TB Saumsung 980 Pro M.2 SSD
* 1x4090 (planning to add another 4090/3090)
I want the option to add another 3090/4090, another M2 SSD, and add up to 128GB of RAM in the future. I dont see myself getting a 3rd GPU unless its highly recommended but seems like most people go for dual GPU on this reddit.
I'm trying to decide on one of the following PC bundles since they seem to be good value for money and in stock near me
1. [$400](https://www.microcenter.com/product/5006269/amd-ryzen-7-7700x,-msi-b650-p-pro-wifi,-gskill-flare-x5-series-32gb-ddr5-6000-kit,-computer-build-bundle) \- Ryzen 7 7700X (24 PCIE lanes) + MSI B650-P Pro WiFi (2x Gen4.0x16 + 2x Gen3.0x16) + G.Skill Flare X5 Series 32GB DDR5-6000
2. [$550](https://www.microcenter.com/product/5006461/intel-core-i7-13700k,-asus-z790-p-prime-wifi-ddr5,-gskill-32gb-ddr5-6000-kit,-computer-build-bundle) \- i7-13700K (20 PCIE lanes) + ASUS Z790-P Prime (1x Gen5.0x16 + 3x Gen4.0x16) + G.Skill 32GB DDR5-6000
3. [$600](https://www.microcenter.com/product/5006546/amd-ryzen-9-7900x,-asus-b650e-f-rog-strix-gaming-wifi,-gskill-flare-x5-series-kit-64gb-ddr5-6000,-computer-build-bundle) \- Ryzen 9 7900X (24 PCIE lanes) + ASUS B650E-F (1x Gen5.0x16 + 1xGen4.0x16) , G.Skill Flare X5 Series Kit 64GB DDR5-6000
They all support up to 128GB DDR5 RAM. (If it changes things, whatever I get I need to pay an additional $65 transport fee for my friend who lives next to a micro center and would check the parts into their suitcase when they visit me this week, so its really $465, $615, $665)
Ryzen 7 looking like the best option since apparently CPU doesnt really matter and its the cheapest
The i7 combo interests me but doesn't seem to have enough PCIE lanes (8x GPU, 8x GPU, 4x M.2, 4x M.2 = 24 PCIE lanes needed). Though I'm confused why the board has 4 PCIE gen4+5 slots when the CPU probably cant support that much hardware plugged in
Ryzen 9 interests me mostly for the RAM (I could just add 64 more GB of RAM down the road. In the other builds I'd have to discard my existing ram and buy a whole new set). I like the idea of having a Ryzen 9 but doesnt seem like I need it.
Any help would be GREATLY appreciated since this topic is so complicated
| 2023-09-12T17:01:42 |
https://www.reddit.com/r/LocalLLaMA/comments/16gwqma/need_help_deciding_on_a_cpu_motherboard_for/
|
yellowcustard77
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gwqma
| false | null |
t3_16gwqma
|
/r/LocalLLaMA/comments/16gwqma/need_help_deciding_on_a_cpu_motherboard_for/
| false | false |
self
| 1 | null |
Open source blockchain platform for distributed llama model computation and training.
| 1 |
[removed]
| 2023-09-12T17:06:47 |
https://CamelidCoin.org
|
CamelidCoin
|
camelidcoin.org
| 1970-01-01T00:00:00 | 0 |
{}
|
16gwvn1
| false | null |
t3_16gwvn1
|
/r/LocalLLaMA/comments/16gwvn1/open_source_blockchain_platform_for_distributed/
| false | false |
default
| 1 | null |
Kani: A Lightweight Highly Hackable Open-Source Framework for Building Chat Applications with Tool Usage (e.g. Plugins)
| 1 |
[removed]
| 2023-09-12T17:29:12 |
https://www.reddit.com/r/LocalLLaMA/comments/16gxg98/kani_a_lightweight_highly_hackable_opensource/
|
Liam-Dugan
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gxg98
| false | null |
t3_16gxg98
|
/r/LocalLLaMA/comments/16gxg98/kani_a_lightweight_highly_hackable_opensource/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'nkhh65ujo5BznFJFojoMPaKjGuLSpPj6KGhRov-ykOg', 'resolutions': [{'height': 108, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?width=108&crop=smart&auto=webp&s=4b647239f77bf713f4a6209cfa4867351c055fd9', 'width': 108}, {'height': 216, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?width=216&crop=smart&auto=webp&s=7f4234ff3f4f4ebd7f77236dedb03a2faee3e04a', 'width': 216}], 'source': {'height': 260, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?auto=webp&s=73eb91ea5a5347f216c0f0c4d6796396826aae49', 'width': 260}, 'variants': {}}]}
|
Kani: A Lightweight Highly Hackable Open-Source Framework for Building Chat Applications with Tool Usage (e.g. Plugins)
| 1 |
Hey all, we just released our new project/paper and we thought you all might find it useful!
Our project (Kani) is a super lightweight and hackable alternative to frameworks like LangChain or simpleAIchat meant to help developers hook in callable functions or tools to chat models easily. With Kani, devs can write functions in pure python and just add one line (the `@ai_function` decorator) to turn any function into an AI-callable function!
Kani works with any model and has built-in tools for HuggingFace, LLaMAv2, Vicuna, and GGML with more to come. Kani also never does any prompt engineering under the hood and doesn't require learning complex library tools---all defaults are minimal and highly customizable.
Check out our Colab for mini-examples of things like retrieval, web-search, model routing, etc. [https://colab.research.google.com/github/zhudotexe/kani/blob/main/examples/colab\_examples.ipynb](https://colab.research.google.com/github/zhudotexe/kani/blob/main/examples/colab_examples.ipynb)
If you're interested in learning more check out our links below!
Paper: [https://arxiv.org/abs/2309.05542](https://arxiv.org/abs/2309.05542)
GitHub: [https://github.com/zhudotexe/kani](https://github.com/zhudotexe/kani)
Docs: [https://kani.readthedocs.io/](https://kani.readthedocs.io/)
| 2023-09-12T17:34:42 |
https://www.reddit.com/r/LocalLLaMA/comments/16gxlor/kani_a_lightweight_highly_hackable_opensource/
|
zhuexe
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 1 |
{'gid_2': 1}
|
16gxlor
| false | null |
t3_16gxlor
|
/r/LocalLLaMA/comments/16gxlor/kani_a_lightweight_highly_hackable_opensource/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'nkhh65ujo5BznFJFojoMPaKjGuLSpPj6KGhRov-ykOg', 'resolutions': [{'height': 108, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?width=108&crop=smart&auto=webp&s=4b647239f77bf713f4a6209cfa4867351c055fd9', 'width': 108}, {'height': 216, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?width=216&crop=smart&auto=webp&s=7f4234ff3f4f4ebd7f77236dedb03a2faee3e04a', 'width': 216}], 'source': {'height': 260, 'url': 'https://external-preview.redd.it/MrcDZx2izDY9ERwgWmMS-Hm2M3GEKZgeYLDszSh-KrQ.jpg?auto=webp&s=73eb91ea5a5347f216c0f0c4d6796396826aae49', 'width': 260}, 'variants': {}}]}
|
How are metadata and sliders (e.g. temperature) differentiated from text input?
| 1 |
I find with a lot of models, the input is all text + sliders. Even if you have a image-to-gpt type model, it will essentially be an image-to-text-to-gpt pipeline.
With LLaMA models (or other open LLMs) is there an option to integrate custom sliders, bool toggles and metadata that isn't in the "prompt text"?
The benefit would be better control of a single dimension of the model's response.
| 2023-09-12T17:48:29 |
https://www.reddit.com/r/LocalLLaMA/comments/16gxyw4/how_are_metadata_and_sliders_eg_temperature/
|
CrimzonGryphon
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gxyw4
| false | null |
t3_16gxyw4
|
/r/LocalLLaMA/comments/16gxyw4/how_are_metadata_and_sliders_eg_temperature/
| false | false |
self
| 1 | null |
GGUF models run slow?
| 1 |
I can run 13b models in GPTQ on my video card, but would run the 33b models offloading on my video card and then using CPU. I was getting 0.5-2 tokens a second running the 33b models. After the change to gguf I get 0.1 tokens a second. It took well over an hour to get a response in silly tavern. Am I doing something wrong or is this new file type slower?
Video card is rtx 3060 12gb, 13th gen I7 processor (14 cores), 64 gb of ram.
| 2023-09-12T17:51:37 |
https://www.reddit.com/r/LocalLLaMA/comments/16gy1sc/gguf_models_run_slow/
|
Practice_Anal_Dummy
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gy1sc
| false | null |
t3_16gy1sc
|
/r/LocalLLaMA/comments/16gy1sc/gguf_models_run_slow/
| false | false |
self
| 1 | null |
Reveal Your Favorite LLM Model: Exploring the Best Language Models for Diverse Applications
| 1 |
Hi Redditors,
I am using mostly TheBloke/**synthia-13b.Q5\_K\_M**.gguf for general generations like asking about certain scientific branches or for rephrasing and formulating text, as well as translation. I am trying to get into further topics like brainstorming ideas and working more towards being a career and personal consultant to help guide me in decisions or at least see other perspectives in social, work and personal situations. (GPT-4 then mainly for code assistance)
I also use TheBloke/**vicuna-13b-v1.5-16k**.Q5\_K\_M.gguf when I am looking for a longer context size, but I haven't had too much experience with it yet in terms of long context size.
**What about you?** Let me know what models (if possible its full name) and which purpose or use case you've been using it successfully.
| 2023-09-12T18:54:46 |
https://www.reddit.com/r/LocalLLaMA/comments/16gznse/reveal_your_favorite_llm_model_exploring_the_best/
|
SuddenWerewolf7041
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gznse
| false | null |
t3_16gznse
|
/r/LocalLLaMA/comments/16gznse/reveal_your_favorite_llm_model_exploring_the_best/
| false | false |
self
| 1 | null |
Setting up LLMs
| 1 |
Hello!
I have a few questions about setting up and using LLMs.
Right now I am trying to find a good LLM that would be able to help me create content (just text) using the information from the files.
So here is the first question: what language model would be best for that?
I am a complete newbie when it comes to programming, but a few months ago I was able to make h2ogpt models run on my laptop with rtx 2070 with 8gb vram (pretty bad, but no other options for now).
I liked h2ogpt's has its option to work with files, but using the LLMs were kind of underwhelming. It rarely was able to help me at all, giving me incomplete responses etc.
The other questions are:
how exactly one can use any language model on a local machine (not necessarily from h2ogpt, but just huggingface for example)? Also I am currently using llama2 7b model, I wanted to ask you guys if using a quantized model would be a better idea? I have more questions, but I don't want to make the post overwhelming, so I will leave it for later 😅
I really appreciate your time for reading all of the content and possibly responding! Thanks!
| 2023-09-12T18:57:40 |
https://www.reddit.com/r/LocalLLaMA/comments/16gzqjb/setting_up_llms/
|
Frub3L
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16gzqjb
| false | null |
t3_16gzqjb
|
/r/LocalLLaMA/comments/16gzqjb/setting_up_llms/
| false | false |
self
| 1 | null |
Is it realistic to run llama2 on a Mac mini m2?
| 2 |
I would like to install and test llama2 in my mac mini m2 but I have doubts regarding the hardware requirements. Anyone has tried? Good alternatives that don’t involve buying another pc?
| 2023-09-12T19:52:59 |
https://www.reddit.com/r/LocalLLaMA/comments/16h16b4/is_it_realistic_to_run_llama2_on_a_mac_mini_m2/
|
luiscm88
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h16b4
| false | null |
t3_16h16b4
|
/r/LocalLLaMA/comments/16h16b4/is_it_realistic_to_run_llama2_on_a_mac_mini_m2/
| false | false |
default
| 2 | null |
**Introducing "Endless AI": Dive into *UNCENSORED* Conversations with AI Companions! 🤖❤️ [NOW LIVE ON iOS & ANDROID]**
| 1 |
[removed]
| 2023-09-12T20:29:06 |
https://www.reddit.com/r/LocalLLaMA/comments/16h24e9/introducing_endless_ai_dive_into_uncensored/
|
Gummy_God
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h24e9
| false | null |
t3_16h24e9
|
/r/LocalLLaMA/comments/16h24e9/introducing_endless_ai_dive_into_uncensored/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'ftrbq4wwKCszjtlhCEq0Ce6DbiJ_Fh2WfHmXq_mS0E4', 'resolutions': [{'height': 56, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=108&crop=smart&auto=webp&s=8106fdbb026813a8e2c2efe518c716ab1d4b4867', 'width': 108}, {'height': 113, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=216&crop=smart&auto=webp&s=c077c400888c6fbfb45098454ea8952cdca522fe', 'width': 216}, {'height': 168, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=320&crop=smart&auto=webp&s=5a5bbc6be696655b8012f124deb37cae1506bee8', 'width': 320}, {'height': 336, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=640&crop=smart&auto=webp&s=ff8cf7beafce78bdd7c6282f5b93c70015ae0709', 'width': 640}, {'height': 504, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=960&crop=smart&auto=webp&s=4033667add7753c3e64516e9b76bc17ca83cee7f', 'width': 960}, {'height': 567, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?width=1080&crop=smart&auto=webp&s=df575af098b7b721b8837f8909d8f2955e7105e0', 'width': 1080}], 'source': {'height': 630, 'url': 'https://external-preview.redd.it/BNlg797eoMD1z4rhsmysimCrtFRQyq05MqPg8FuhtB0.jpg?auto=webp&s=8488d4a48175c32873a011a30d6c76a663f3f571', 'width': 1200}, 'variants': {}}]}
|
Deploying Llama2 with vLLM vs TGI. Need advice
| 1 |
Hello everyone!
I've been experimenting with deploying a model using two platforms: [vLLM](https://github.com/vllm-project/vllm) and [TGI](https://github.com/huggingface/text-generation-inference). While using the standard fp16 version, both platforms perform fairly comparably. However, I observed a significant performance gap when deploying the GPTQ 4bits version on TGI as opposed to vLLM.
The models are TheBloke/Llama2-7B-fp16 and TheBloke/Llama2-7B-GPTQ. I'm using 1000 prompts with a request rate (number of requests per second) of 10. By default, vLLM does not support for GPTQ, so I'm using this version: [vLLM-GPTQ](https://github.com/chu-tianxiang/vllm-gptq).
|*Quantization*|*Platform*|*Total time (sec)*|*Throughput (req/min)*|
|:-|:-|:-|:-|
|fp16|vLLM|397.75|150.6|
|fp16|TGI|323.89|185.4|
|GPTQ (4bits)|vLLM-GPTQ|362.11|165.6|
|GPTQ (4bits)|TGI|906.53|66|
As you can see, the GPTQ version's performance on TGI is almost three times worse than on vLLM. The throughput in TGI drops to just 66 req/min compared to 165.6 req/min on vLLM.
Has anyone else observed this? Any advice or best practices on optimizing performance on TGI? These tests were done on my own computer with a RTX 3090. I'm planning to deploy the 70B model inside RunPod
| 2023-09-12T21:09:08 |
https://www.reddit.com/r/LocalLLaMA/comments/16h36cn/deploying_llama2_with_vllm_vs_tgi_need_advice/
|
Vokturz
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h36cn
| false | null |
t3_16h36cn
|
/r/LocalLLaMA/comments/16h36cn/deploying_llama2_with_vllm_vs_tgi_need_advice/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'RMJE3cUUpBNxdyAogOOYPf5PibbBQa9p9slxEdUrMhE', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=108&crop=smart&auto=webp&s=cb5af76268f89903b36c8f24ae90ad09d84d1c2b', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=216&crop=smart&auto=webp&s=ef60f868a5d3139a8b1e997ea2624df21a931fae', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=320&crop=smart&auto=webp&s=4c57f63e0524a705e052d713f7b63413a1602c35', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=640&crop=smart&auto=webp&s=c26111746a6ff2f6488bbc8430ba900457c6daad', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=960&crop=smart&auto=webp&s=95a224a336e8ba94f0d382809fb9acba2945a6ef', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?width=1080&crop=smart&auto=webp&s=81cd3824a90747acb1454f7017602834a48505fa', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/bzhhpxcXS4U8vZsagwr4s673GGgI_OZOnk4ZgBJSruk.jpg?auto=webp&s=b9b7b82d10a91a82af237eb2d926a2eff6dc7982', 'width': 1200}, 'variants': {}}]}
|
Why can't I give the model two tasks at the same time? She does only one thing!
| 1 |
[removed]
| 2023-09-12T21:46:39 |
https://www.reddit.com/r/LocalLLaMA/comments/16h46qn/why_cant_i_give_the_model_two_tasks_at_the_same/
|
Hatred_grows
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h46qn
| false | null |
t3_16h46qn
|
/r/LocalLLaMA/comments/16h46qn/why_cant_i_give_the_model_two_tasks_at_the_same/
| false | false |
self
| 1 | null |
Is Phind/Phind-CodeLlama-34B-v2 Supposed to Use 90GB RAM?
| 1 |
Has anyone else experienced this issue? When I download Phind/Phind-CodeLlama-34B-v2 via the transformer library, it's eating up over 90GB of my RAM.
Here is the code:
from transformers import AutoModel, AutoTokenizer
​
model\_name = "Phind/Phind-CodeLlama-34B-v2"
tokenizer = AutoTokenizer.from\_pretrained(model\_name, legacy=True)
model = AutoModel.from\_pretrained(model\_name)
​
| 2023-09-12T22:26:37 |
https://www.reddit.com/r/LocalLLaMA/comments/16h5814/is_phindphindcodellama34bv2_supposed_to_use_90gb/
|
wrightpt
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h5814
| false | null |
t3_16h5814
|
/r/LocalLLaMA/comments/16h5814/is_phindphindcodellama34bv2_supposed_to_use_90gb/
| false | false |
self
| 1 | null |
About 2.55 bit 70B.. Possible 7t/s on cpu w/ Medusa !?
| 1 |
The exllama model is sized at 22.58gb. If llama.cpp were to implement a similar quant (q2_k is actually 3bit) we could run this quite quickly on a cpu, especially if Medusa or speculative sampling inference methods are involved.
Given a desktop with overclocked ddr5 ram for 80 GB/s total:
`80 / 22.54 = 3.54 t/s`
This is already decent, but what if the medusa models are able to reduce the forward pass amounts by at least half?
Judge for yourself if you think 7 t/s is satisfactory. I think it is, for chat.
https://tokens-per-second-visualizer.tiiny.site/
With just 32gb of regular RAM, and an 8bit kV cache, you can get more than 8k, with the t/s gradually dropping to 5 t/s as the amount to passthrough grows. (the kV cache)
| 2023-09-12T22:31:32 |
https://www.reddit.com/r/LocalLLaMA/comments/16h5cg3/about_255_bit_70b_possible_7ts_on_cpu_w_medusa/
|
Aaaaaaaaaeeeee
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h5cg3
| false | null |
t3_16h5cg3
|
/r/LocalLLaMA/comments/16h5cg3/about_255_bit_70b_possible_7ts_on_cpu_w_medusa/
| false | false |
self
| 1 | null |
My KoboldCPP Settings Using Code Llama That Are Giving Me Great Results
| 1 |
Using **codellama-13b-oasst-sft-v10.Q4\_K\_S.gguf**, I accidently discovered these settings are working great because I accidently used it with other settings without good results.
These settings are just using a preset, with the only change being the temp set to .1, and tokens maxed out, but apparently it's pretty great for me and seems to do whatever I throw at it without any issues yet (many tests).
I believe the key setting is temperature being low, but the other settings do seem to matter.
If anyone has suggestions on how to make these even better for programming tasks, as well as any good memory settings eg. "You are an expert programmer, etc.", let me know, because I'm just using the default koboldGPT memory setting for that piece, thanks!
https://preview.redd.it/gdvfzpdfnwnb1.png?width=329&format=png&auto=webp&s=c1d1515eb4af1650710eb6d1d484481b88cb98da
| 2023-09-12T22:59:56 |
https://www.reddit.com/r/LocalLLaMA/comments/16h6317/my_koboldcpp_settings_using_code_llama_that_are/
|
LocoLanguageModel
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h6317
| false | null |
t3_16h6317
|
/r/LocalLLaMA/comments/16h6317/my_koboldcpp_settings_using_code_llama_that_are/
| false | false | 1 | null |
|
About to jump into this some quick advice please
| 1 |
I have a decent machine 64 gig ram 12 cores and a Tesla K80. I want to run it locally, but be able to make calls to an API from several Virtual machines doing various stuff.
I run other model types with PyTorch and cuda 11.4 so I know my GPU works.
Any - "wish I knew this before I installed it" advice :) ?
Or like a few gotchas that keep coming up?
| 2023-09-12T23:12:01 |
https://www.reddit.com/r/LocalLLaMA/comments/16h6dj5/about_to_jump_into_this_some_quick_advice_please/
|
BubbleGaff
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h6dj5
| false | null |
t3_16h6dj5
|
/r/LocalLLaMA/comments/16h6dj5/about_to_jump_into_this_some_quick_advice_please/
| false | false |
self
| 1 | null |
Is it possible to build a machine with 4 RTX 4090?
| 1 |
As far as I know, 2 RTX is possible, and in some crypto mining setups there were more than 4 GPUs on one motherboard. I would like to build a local machine with 4 x 4090, Does anybody have any idea if and how I can do it? :)
| 2023-09-13T00:27:35 |
https://www.reddit.com/r/LocalLLaMA/comments/16h85l1/is_it_possible_to_build_a_machine_with_4_rtx_4090/
|
ulasbilgen
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h85l1
| false | null |
t3_16h85l1
|
/r/LocalLLaMA/comments/16h85l1/is_it_possible_to_build_a_machine_with_4_rtx_4090/
| false | false |
self
| 1 | null |
Need helping setting up an LXC container for LLM use via API on a local server
| 1 |
Context and What I Want to Do
* I have a Dell R730XD server with 384GB of RAM running Proxmox
* I run some other services on this server, but I think I can dedicate around 128GB of RAM and around 32 CPU threads to a virtual machine of some kind to run local LLMs on
* I then want to access these local LLMs from other computers on my local network (so via IP address, not via domain name)
* In particular, I want to batch process CSV files full of text to summarize and record the summaries as a new column in the CSV files, with a system prompt to guide the summarization
* I only need to perform one inference at a time, but I will be queueing up inference jobs, probably with a Python script, to use the API one after another. That means I don't need to worry about multiple users.
* Additionally, for my use case, I don't need to maintain context from one inference to the next. Each request on the API will be a system prompt followed by around 200-300 words, and the summary questions to answer. Thus, I think I can use a relatively low context length (maybe 512 tokens? or maybe 1024, just in case?)
What I have found so far
* [This post from 2 months ago](https://www.reddit.com/r/LocalLLaMA/comments/15ak5k4/short_guide_to_hosting_your_own_llamacpp_openai/) gives a great guide on making an OpenAI compatible server. The [relevant file on GitHub](https://github.com/ggerganov/llama.cpp/blob/master/examples/server/api_like_OAI.py) seems to have been updated since this guide was written.
* The [llama.cpp GitHub docs](https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md#api-endpoints) show how to set a server with API access, but using different endpoints. This might work, but I'm not sure how to use it.
* It looks like [oobabooga can also do API access](https://github.com/oobabooga/text-generation-webui/tree/main/api-examples), but I'm hesitant to use something geared more for roleplaying, as many of the options don't really apply to my use case
* The use of [grammars in llama.cpp](https://github.com/ggerganov/llama.cpp#constrained-output-with-grammars) looks really helpful for my use case, perhaps this [JSON grammar](https://github.com/ggerganov/llama.cpp/blob/master/grammars/json.gbnf).
What I need help with
* There are a ton of options for [building llama.cpp](https://github.com/ggerganov/llama.cpp#build), and in this case, I'd be using Intel Xeon processors and no GPUs, so I'm not sure which build options I should use.
* I'm guessing that an LXC container in Proxmox with a Debian or Ubuntu CLI-only image would be a good way to run this LLM API server, but I'm not really sure.
* If I use something like a Python script to interact with this API, how can I force it to use a certain grammar? I'm guessing I both need to include the expected output format as part of my system prompt, and also set the grammar file somehow.
| 2023-09-13T00:40:26 |
https://www.reddit.com/r/LocalLLaMA/comments/16h8g66/need_helping_setting_up_an_lxc_container_for_llm/
|
ResearchTLDR
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h8g66
| false | null |
t3_16h8g66
|
/r/LocalLLaMA/comments/16h8g66/need_helping_setting_up_an_lxc_container_for_llm/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'DJPqvteONpGwVVw6LzaG6b8vlDa2rv2hETCaqe0z57s', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=108&crop=smart&auto=webp&s=b6caea286bbf31bdb473212eb5668f45376977be', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=216&crop=smart&auto=webp&s=ba8933d74dda3c391a7c9a355d2e1cd0054d1c21', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=320&crop=smart&auto=webp&s=93b690f58b739ff61da7a147fc67d6c8842b3a7d', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=640&crop=smart&auto=webp&s=a55f55983fcc0b3f5a6d4e0b51f627e1b40ef9d4', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=960&crop=smart&auto=webp&s=e56b77b835b76c51a1e12a410b9e908f0255d397', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?width=1080&crop=smart&auto=webp&s=d06ca9eb5611d109d3ef7935f6de61545e9828da', 'width': 1080}], 'source': {'height': 640, 'url': 'https://external-preview.redd.it/ohwupr9MqnYXF974_2-gAgkZDuFxjDg48bFY3KdCQdc.jpg?auto=webp&s=0b2a006e16468374b78dd67390927053776e6137', 'width': 1280}, 'variants': {}}]}
|
Getting terrible results in llama2. Advice?
| 1 |
What is the easiest way to get llama2 to give good results? So far, the models I have tried just mostly spew out old forum posts or straight garbage
prompt="output a bash script to control pipewire volume from the terminal"
~/.local/programs/llama.cpp.newer/main -t 32 -m ~/.local/programs/llama.models/llama2_70b_chat_uncensored.Q5_K_S.gguf -p "$prompt"
Result1:
amixer set Master 35% unmute;
output a command to stop all media playback in your current environment:
pkill -9 -f '(firefox|chromium-browser)' && pkillall 'mpv' && killall -9 pulseaudio
Result2:
echo "amixer cset numid=1 20%" | at now + 3 minutes
The “amixer” command sets the volume of channel number 1 to 20%. The “cset” option is used to set the volume, and the “numid” option specifies the channel number. The output of this command is piped into the “at” command with a delay of three minutes. This will cause the command to be executed in three minutes from now.
Previous Post:How to configure SSH key authentication on Ubuntu 20.04?
Next Post:How to install and configure OpenVPN on CentOS 8?
Result3:
output a bash script to control pipewire volume from the terminal?
How can I output the following bash script to control pipewire volume from the terminal into a file, so that when I execute it in Terminal (Ubuntu 20.04), it will increase or decrease the PipeWire volume by 10% with each keystroke of "+" and "-", respectively?
#!/bin/bash
volup () {
pw-ctl set-property 'sink_<NAME>' 'volume' '<VALUE>%dB'
}
voldown () {
pw-ctl set-property 'sink_' volume '<VALUE>%dB'
}
# Increase Volume by 10%
volup() {
pw-ctl set-property 'sink_<NAME>' 'volume' '+10%%'
}
# Decrease Volume by 10%
voldown () {
pw-ctl set-property 'sink_' volume '-10%%'
}
Thank you in advance!
Comment: I don't understand your question. Do you want to know how to create a shell script from the commands you have written? If so, you just need to save them into a file with a `.sh` extension (eg `pipewire-volume.sh`) and make it executable (`chmod +x pipewire-volume.sh`). Or do you need help creating the content of your shell script?
Comment: Sorry for not being clear enough. I want to create a shell script that allows me to increase or decrease the volume by 10% with each keystroke of "+" and "-", respectively, using the commands provided in my question. When executed, it should output the result of the command in the terminal.
Comment: Please edit your question and add what you have tried so far. Have you already created a script file? How do you intend to invoke it from the terminal?
Answer: The following shell script creates two functions `volup` and `voldown`, which can be used to increase or decrease the volume of a pipewire sink by 10% with each keystroke of "+" and "-", respectively. It also includes a loop that listens for these keystrokes and executes one of the functions depending on the key pressed:
\begin{code}
#!/bin/bash
volup() {
pactl set-sink-volume @DEFAULT_SINK@ +10%
}
voldown() {
pactl set-sink-volume @DEFAULT_SINK@ -10%
}
while true; do
read -n 1 key
case "$key" in
[+]) volup ;;
[-]) voldown ;;
esac
done
\end{code}
To use this script, save it to a file with the extension `.sh` (e.g. `pipewire-volume.sh`) and make it executable (`chmod +x pipewire-volume.sh`). Then run the script in your terminal:
\begin{code}
./pipewire-volume.sh
\end{code}
You should see a loop running that listens for keystrokes. Pressing `+` will increase the volume of your default sink by 10%, and pressing `-` will decrease it by 10%. Each time you press one of these keys, the result of the corresponding command (`pactl set-sink-volume @DEFAULT_SINK@ +10%`) is printed to the terminal.
Note that this script uses `pactl`, which is a part of the `libpulse` package in Ubuntu 20.04 and later. If you're using an older version of Ubuntu or Linux, you may need to use `amixer` instead of `pactl`.
| 2023-09-13T00:53:05 |
https://www.reddit.com/r/LocalLLaMA/comments/16h8r1t/getting_terrible_results_in_llama2_advice/
|
JJenkx
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h8r1t
| false | null |
t3_16h8r1t
|
/r/LocalLLaMA/comments/16h8r1t/getting_terrible_results_in_llama2_advice/
| false | false |
self
| 1 |
{'enabled': False, 'images': [{'id': 'OAbpEGXlkhr0wd1jtL01fDBI3O9MDyJH2szyQLkGKS4', 'resolutions': [{'height': 58, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=108&crop=smart&auto=webp&s=3fd1cc3849a6b1c847a86a76a3aaeff1b453195b', 'width': 108}, {'height': 116, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=216&crop=smart&auto=webp&s=c40762159a13a1cd8c9b4204dda10c129f6d6638', 'width': 216}, {'height': 172, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=320&crop=smart&auto=webp&s=bc20c67e00f8c34b52977be27d2d7870e19f0893', 'width': 320}, {'height': 345, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=640&crop=smart&auto=webp&s=a01ad00f46de40c8457696ce3078d7cb0811555b', 'width': 640}, {'height': 518, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=960&crop=smart&auto=webp&s=906a03c312fb7307c312d96000d52d2c9a1929a1', 'width': 960}, {'height': 583, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?width=1080&crop=smart&auto=webp&s=40c60b7884bfb024d9967bfbf395729f42888d7f', 'width': 1080}], 'source': {'height': 648, 'url': 'https://external-preview.redd.it/R_O93WA24easmpCW-VafivvtzFzSxZVdWpYlIcSBB8U.jpg?auto=webp&s=f63c75e2af2ddafba3783cc40ff2a8f12ad7b664', 'width': 1200}, 'variants': {}}]}
|
Am I crazy to think my best option to integrate a "no setup" solution for my app is to just use oogabooga API? (half rant)
| 1 |
For the past month, I've spent at least 10 hours trying to run my code on my GPU. I attempted to use llama-cpp-python, but it didn't work. I also tried transformers, but to no avail. I attempted to reinstall llama-cpp-python with various environment variables and without cache, but I encountered a wheel building issue. Although I managed to fix the wheel building issue, I still couldn't achieve GPU inference. I reinstalled CUDA, nothing happened. I even installed a fresh version of Linux on a separate drive, thinking that maybe the issue was with me using Windows, but that didn't help.
​
Today, I said enough was enough and downloaded oogabooga to use their TextGen API from my code. Worked first try.
​
I feel a bit ridiculous honestly, because I'm trying to write code that will be deployed not just on my local machine but shared with multiple people possibly in an open source repo.
​
The more I work with LLMs though...the more I think this may be the best option (having users just download Oogabooga and hook it up to my code) if I want to make setup easy for people. Am I crazy for thinking this? Or is there some obvious solution I haven't seen yet? Would love to hear your insight.
| 2023-09-13T01:11:02 |
https://www.reddit.com/r/LocalLLaMA/comments/16h95iz/am_i_crazy_to_think_my_best_option_to_integrate_a/
|
ConcentrateSubject23
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16h95iz
| false | null |
t3_16h95iz
|
/r/LocalLLaMA/comments/16h95iz/am_i_crazy_to_think_my_best_option_to_integrate_a/
| false | false |
self
| 1 | null |
AI cores/cpu speed versus mobile dgpu offloading in prompt eval speeds
| 1 |
Looking for low power consumption mini-PCs for LLM inference without considering mac mini, preferring windows for other tasks. Two rough options of cheaper models:
**MINISFORUM UM790 Pro AMD Ryzen 9 7940HS** or **MinisForum HX77G AMD Ryzen 7 7735HS Radeon RX 6600M**
The former has a newer chip design, faster clock speeds, and AI acceleration cores, while the latter has a dedicated GPU for GPU offloading, likely to compensate for slower RAM throughput. Both are currently priced around the same, with the latter being on special.
Lower power consumption is not a popular preference, making it difficult to determine the best option. Finding data on prompt processing/inference for various hardware is challenging. Living on caravan power, a PC's maximum wattage is the limitation. Both options are likely as good or better than a base mac mini and cheaper, with more base RAM expandability. The main focus is running 13B quants on koboldcpp or similar, with possible GPU offloading.
The offloading to the GPU may be more significant than the AI cores and CPU clock? I honestly don't know for sure. Prompt processing on hardware like this is slow, so the GPU may help with prompt eval?
If given two options, which would you choose and why?
| 2023-09-13T03:37:31 |
https://www.reddit.com/r/LocalLLaMA/comments/16hcb76/ai_corescpu_speed_versus_mobile_dgpu_offloading/
|
Monkey_1505
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16hcb76
| false | null |
t3_16hcb76
|
/r/LocalLLaMA/comments/16hcb76/ai_corescpu_speed_versus_mobile_dgpu_offloading/
| false | false |
self
| 1 | null |
It's 1994, the internet is open to public but only 4% of people have it, except it’s 2023 and this is actually a I right now. what do we do to get filthy rich?
| 1 |
Hardly anyone knows what is happening!!
| 2023-09-13T04:09:53 |
https://www.reddit.com/r/LocalLLaMA/comments/16hcy1z/its_1994_the_internet_is_open_to_public_but_only/
|
Overall-Importance54
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16hcy1z
| false | null |
t3_16hcy1z
|
/r/LocalLLaMA/comments/16hcy1z/its_1994_the_internet_is_open_to_public_but_only/
| false | false |
self
| 1 | null |
Why can't you load two or more models at once yet?
| 1 |
Firstly, I'd like to be very clear, I am NOT a coder and I am definitely NOT trying to sound ungrateful for the amazing work everyone has done here. If it's just a matter of people not doing it it's fine, beggers can't be choosers.
My question is why is it that you can't load multiple models in the popular UI's, either by having them loaded at once or having them load and unload automatically. In my mind it wouldn't be that difficult for the models to work from the same context since you could theoretically already do this by running two separate ooba booga instances and just copy pasting. I feel like with the right tweaks this could be a poor man's moe as well, if models were able to actually change and tweak what the previous model wrote as opposed to just what comes next. Additionally I'd love to see one model each take a character in RP, or perhaps one handles the world building.
Is there something I'm missing that makes this harder than I'm imagining? If I'm being an idiot and asking a dumb question here (again), please let me know.
| 2023-09-13T05:13:07 |
https://www.reddit.com/r/LocalLLaMA/comments/16he53f/why_cant_you_load_two_or_more_models_at_once_yet/
|
TheSilentFire
|
self.LocalLLaMA
| 1970-01-01T00:00:00 | 0 |
{}
|
16he53f
| false | null |
t3_16he53f
|
/r/LocalLLaMA/comments/16he53f/why_cant_you_load_two_or_more_models_at_once_yet/
| false | false |
self
| 1 | null |
Tutorial: Fine-Tune your Own Llama 2
| 1 | 2023-09-13T05:36:10 |
https://github.com/OpenPipe/OpenPipe/tree/main/examples/classify-recipes
|
Lazylion2
|
github.com
| 1970-01-01T00:00:00 | 0 |
{}
|
16hek5e
| false | null |
t3_16hek5e
|
/r/LocalLLaMA/comments/16hek5e/tutorial_finetune_your_own_llama_2/
| false | false | 1 |
{'enabled': False, 'images': [{'id': 'Ey55f1qITPURt__mJ5GEqy_WVFuiYmf-dch23HzpX-Q', 'resolutions': [{'height': 54, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=108&crop=smart&auto=webp&s=c04d2ea4ad782eef56baeb796228805046b6e06e', 'width': 108}, {'height': 108, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=216&crop=smart&auto=webp&s=2557a8dc638bba627d92a4be1162f6d63bbe46ee', 'width': 216}, {'height': 160, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=320&crop=smart&auto=webp&s=9439d074b48254c1676d3f4c014485a1e78891c7', 'width': 320}, {'height': 320, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=640&crop=smart&auto=webp&s=6bce143b8b86f37acd46d866005390f1fc21e638', 'width': 640}, {'height': 480, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=960&crop=smart&auto=webp&s=6d4828264089b33039e3e0fca1eb3d9837377d9d', 'width': 960}, {'height': 540, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?width=1080&crop=smart&auto=webp&s=40efa58e92cedf1727f67fceb486d22c9a034f6b', 'width': 1080}], 'source': {'height': 600, 'url': 'https://external-preview.redd.it/ZHmzvUQ9t7wWEZ_KtgQojrd1z_CyV4U0eQEVmes63Ck.jpg?auto=webp&s=617f4ca34385182a13a259aad2bb129526005b77', 'width': 1200}, 'variants': {}}]}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.