- For now, it's only simple commands. The LLM has already proven it can do several tool calls in a single query. I remember now that those minecraft agents did create scripts themselves? It's definitely something I can explore, too.
- No special training or tokenization at the moment. It's just a raw LLM. I think a finetuned LLM would improve quality, massively.
- Yes, only certain characters. I'd hate for the "dragon slayer quest" giver to figure out they could kill the dragon themselves and go get killed. The agent has tools to interact with "fixed dialogue" npcs, though, like shops and quest givers. I've yet to see them do it efficiently, though.
Rickard Edén
neph1
AI & ML interests
applying, dataset creation, finetuning, loras
Recent Activity
published
a model
1 day ago
neph1/1950s_scifi_wan2.2_t2v
updated
a model
1 day ago
neph1/1950s_scifi_wan2.2_t2v
published
a model
3 days ago
neph1/80s_fantasy_movies_wan_2.2
Organizations
replied to
their
post
4 days ago
posted
an
update
4 days ago
Post
3117
I'm building a mmo-ish RPG with LLM agents that can (hopefully) complete player tasks, as an experiment. I've started documenting my progress here: https://huggingface.co/blog/neph1/rpg-llm-agents
Let me know if you want to see more of it.
Let me know if you want to see more of it.
-
- 4 replies
posted
an
update
4 months ago
Post
2978
I know Hunyuan Video is yesterday's jam, but in case you're looking for some cinematic LoRA's (and don't like civitai for some reason), I've uploaded my most popular ones to hf. They are:
1980s fantasy: neph1/1980s_Fantasy_Movies_Hunyuan_Video_Lora
1950s scifi: neph1/50s_scifi_hunyuan_video_lora
1920s horror: neph1/1920s_horror_hunyuan_video_lora
1980s fantasy: neph1/1980s_Fantasy_Movies_Hunyuan_Video_Lora
1950s scifi: neph1/50s_scifi_hunyuan_video_lora
1920s horror: neph1/1920s_horror_hunyuan_video_lora
posted
an
update
7 months ago
Post
1836
There's a new version of the Swedish instruct model, bellman. Due to 'popular demand' (at least as opposed to 'no demand'), I based it off the latest mistral 7b, v0.3. The v0.2 seems to be the most popular of the bunch, despite being quite old by now. Why, I don't know. Must be a link in some old reddit post that is drawing clicks. :)
Anyway, here it is:
neph1/bellman-mistral-7b-instruct-v0.3
You can also try it out (on cpu), here:
neph1/bellman
Anyway, here it is:
neph1/bellman-mistral-7b-instruct-v0.3
You can also try it out (on cpu), here:
neph1/bellman
posted
an
update
8 months ago
Post
1209
For those interested in game development I've released an experimental finetune of Qwen2.5-Coder for Unity.
neph1/Qwen2.5-Coder-7B-Instruct-Unity
It's using a mix of open source datasets + one specifically made for this (also OS) with multiple responses.
Also thinking about making a code completion model, or one to have more architectural discussions with.
neph1/Qwen2.5-Coder-7B-Instruct-Unity
It's using a mix of open source datasets + one specifically made for this (also OS) with multiple responses.
Also thinking about making a code completion model, or one to have more architectural discussions with.
- 1 reply
posted
an
update
10 months ago
Post
614
Bellman, the Swedish finetune, has once again returned in his biggest incarnation yet, at 12b. Based on Mistral-Nemo-Instruct:
neph1/Mistral-Nemo-Instruct-bellman-12b
posted
an
update
about 1 year ago
Post
798
Bellman Swedish finetune based on llama3.1 8b is now available:
neph1/llama-3.1-instruct-bellman-8b-swedish
More quants and fp16 are coming. Working out some issues with colab..
neph1/llama-3.1-instruct-bellman-8b-swedish
More quants and fp16 are coming. Working out some issues with colab..
posted
an
update
about 1 year ago
Post
612
I've noticed some people are still downloading
neph1/sd-seer-griffin-3b
Should I make an update based on a more modern architecture? (griffin-3b is llama (1!))
Should I make an update based on a more modern architecture? (griffin-3b is llama (1!))
- 1 reply
posted
an
update
about 1 year ago
Post
776
First real version of bellman based on llama 3 instruct 8b has been released!
neph1/llama-3-instruct-bellman-8b-swedish
Close to 16k examples, including 250 rows from my translated codefeedback dataset and a number of non-copyrighted stories.
Two quants are up in the gguf folder. I'll work on adding more quants in the coming days.
neph1/llama-3-instruct-bellman-8b-swedish
Close to 16k examples, including 250 rows from my translated codefeedback dataset and a number of non-copyrighted stories.
Two quants are up in the gguf folder. I'll work on adding more quants in the coming days.
reacted to
stas's
post with 👍
over 1 year ago
Post
Do you have a hidden massive storage leak thanks to HF hub models and datasets revisions adding up and not getting automatically deleted?
Here is how to delete all old revisions and only keeping
In terminal A:
Do not answer the prompt and proceed with my instructions.
(note your tmp file will have a different path, so adjust it below)
In terminal B:
The perl one-liner uncommented out all lines that had
Now go back to terminal A and hit: N, Y, Y, so it looks like:
Done.
If you messed up with the prompt answering you still have
For more details and additional techniques please see https://github.com/stas00/ml-engineering/tree/master/storage#huggingface-hub-caches
Here is how to delete all old revisions and only keeping
main in a few quick steps and no tedious manual editing.In terminal A:
$ pip install huggingface_hub["cli"] -U
$ huggingface-cli delete-cache --disable-tui
File to edit: /tmp/tmpundr7lky.txt
0 revisions selected counting for 0.0. Continue ? (y/N)Do not answer the prompt and proceed with my instructions.
(note your tmp file will have a different path, so adjust it below)
In terminal B:
$ cp /tmp/tmpedbz00ox.txt cache.txt
$ perl -pi -e 's|^#(.*detached.*)|$1|' cache.txt
$ cat cache.txt >> /tmp/tmpundr7lky.txtThe perl one-liner uncommented out all lines that had
(detached) in it - so can be wiped out. And then we pasted it back into the tmp file huggingface-cli expects to be edited.Now go back to terminal A and hit: N, Y, Y, so it looks like:
0 revisions selected counting for 0.0. Continue ? (y/N) n
89 revisions selected counting for 211.7G. Continue ? (y/N) y
89 revisions selected counting for 211.7G. Confirm deletion ? (Y/n) yDone.
If you messed up with the prompt answering you still have
cache.txt file which you can feed again to the new tmp file it'll create when you run huggingface-cli delete-cache --disable-tui again.For more details and additional techniques please see https://github.com/stas00/ml-engineering/tree/master/storage#huggingface-hub-caches