🦙💻 CodeLlama

📝 Article | 💻 Colab | 📄 Script

<center><img src="https://i.imgur.com/yTPNIZj.png" width="300"></center>

CodeLlama-7b is a Llama 2 version of CodeAlpaca.

🔧 Training

This model is based on the llama-2-7b-chat-hf model, fine-tuned using QLoRA on the mlabonne/CodeLlama-2-20k dataset. It was trained on an RTX 3090 and can be used for inference.

It was trained using this custom finetune_llama2.py script as follows:

python finetune_llama2.py --dataset_name=mlabonne/CodeLlama-2-20k --new_model=mlabonne/codellama-2-7b --bf16=True --learning_rate=2e-5

<center><img src="https://i.imgur.com/5Qx7Kzo.png"></center>

💻 Usage

# pip install transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/codellama-2-7b"
prompt = "Write Python code to generate an array with all the numbers from 1 to 100"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

sequences = pipeline(
    f'<s>[INST] {prompt} [/INST]',
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
    max_length=200,
)
for seq in sequences:
    print(f"Result: {seq['generated_text']}")

Ouput:

Here is a Python code to generate an array with all the numbers from 1 to 100:

 ```
 numbers = []
 for i in range(1,101):
     numbers.append(i)
 ```

This code generates an array with all the numbers from 1 to 100 in Python. It uses a loop that iterates over the range of numbers from 1 to 100, and for each number, it appends that number to the array 'numbers'. The variable 'numbers' is initialized to a list, and its length is set to 101 by using the range of numbers (0-99).

```## Training procedure


The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16
### Framework versions

- PEFT 0.5.0.dev0

- PEFT 0.5.0.dev0
## Training procedure


The following `bitsandbytes` quantization config was used during training:
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16