---
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