File size: 3,493 Bytes
2114071 bfd9221 2114071 bfd9221 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 |
---
library_name: transformers
base_model: mistralai/Mistral-7B-v0.1
language:
- en
pipeline_tag: text-generation
tags:
- code
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
> This finetuned model is already merged with Mistral7B (Base model)
> There will be 2 options running this model for inference
> - _Option 1:_ Load base model and use **Peft library** to load parameters of finetuned model on base model
> - _Option 2:_ Load finetuned model straight from this huggingface hub
## Approach 1
### Run Inference on Google Colab
1. First run this code to load the base model which is Mistral-7B-v0.1
```py
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
base_model_id = "mistralai/Mistral-7B-v0.1"
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
base_model = AutoModelForCausalLM.from_pretrained(
base_model_id, # Mistral, same as before
quantization_config=bnb_config, # Same quantization config as before
device_map="auto",
trust_remote_code=True,
use_auth_token=True
)
eval_tokenizer = AutoTokenizer.from_pretrained(base_model_id, add_bos_token=True, trust_remote_code=True)
```
2. After that, we would use Peft library to merge the new parameters that we already finetuned with GAML with this code
```py
from peft import PeftModel
import torch
# ft_model = PeftModel.from_pretrained(base_model, "mistral-gama-finetune_allblocks_newdata/checkpoint-45")
ft_model = PeftModel.from_pretrained(base_model, "Phanh2532/GAML-151-500")
#ft_model3 = PeftModel.from_pretrained(base_model, "mistral-allbloclks//checkpoint-250")
#ft_model.save_pretrained('/content/mistral-allblocksft/')
eval_prompt = "Create a GAML code snippet inspired by water pollution in real life"
model_input = eval_tokenizer(eval_prompt, return_tensors="pt").to("cuda")
ft_model.eval()
with torch.no_grad():
print(eval_tokenizer.decode(ft_model.generate(**model_input, max_new_tokens=2000, repetition_penalty=1.15)[0], skip_special_tokens=True))
print('----------------------------------------------------------------------')
#print(eval_tokenizer.decode(ft_model2.generate(**model_input, max_new_tokens=2000, repetition_penalty=1.15)[0], skip_special_tokens=True))
```
## Approach 2
Run this code snippet
```py
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
# Load Mistral 7B model and tokenizer
model_id = "Phanh2532/GAML-151-500"
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
model = AutoModelForCausalLM.from_pretrained(
model_id,
quantization_config=bnb_config,
device_map="auto",
trust_remote_code=True,
use_auth_token=True
)
# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id, add_bos_token=True, trust_remote_code=True)
with torch.no_grad():
print(tokenizer.decode(model.generate(**model_input, max_new_tokens=2000, repetition_penalty=1.15)[0], skip_special_tokens=True))
print('----------------------------------------------------------------------')
# print(eval_tokenizer.decode(ft_model2.generate(**model_input, max_new_tokens=2000, repetition_penalty=1.15)[0], skip_special_tokens=True))
```
### Framework versions
- PEFT 0.7.2.dev0 |