File size: 3,241 Bytes
a422f49 b251617 a422f49 b251617 a422f49 2284226 a422f49 a33d8cd e069b7b a33d8cd e069b7b |
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 98 99 100 101 102 103 104 105 106 107 108 |
---
license: apache-2.0
library_name: peft
tags:
- code
datasets:
- mlabonne/CodeLlama-2-20k
base_model: meta-llama/Llama-2-7b-hf
---
# 🦙💻 CodeLlama
📝 [Article](https://towardsdatascience.com/fine-tune-your-own-llama-2-model-in-a-colab-notebook-df9823a04a32) |
💻 [Colab](https://colab.research.google.com/drive/1PEQyJO1-f6j0S_XJ8DV50NkpzasXkrzd?usp=sharing) |
📄 [Script](https://gist.github.com/mlabonne/b5718e1b229ce6553564e3f56df72c5c)
<center><img src="https://i.imgur.com/yTPNIZj.png" width="300"></center>
`CodeLlama-7b` is a Llama 2 version of [**CodeAlpaca**](https://github.com/sahil280114/codealpaca).
## 🔧 Training
This model is based on the `llama-2-7b-chat-hf` model, fine-tuned using QLoRA on the [`mlabonne/CodeLlama-2-20k`](https://huggingface.co/datasets/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`](https://gist.github.com/mlabonne/b5718e1b229ce6553564e3f56df72c5c) script as follows:
``` bash
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
``` python
# 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
|