🚀 Llama 2 Coder 🦙👩💻
Llama 2 Coder是基於Llama-2 7b模型,使用QLoRA方法和PEFT庫,在CodeAlpaca 20k指令數據集上進行微調得到的模型,可有效輔助編碼工作。
🚀 快速開始
示例代碼
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
model_id = "mrm8488/llama-2-coder-7b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id).to("cuda")
def create_prompt(instruction):
system = "You are a coding assistant that will help the user to resolve the following instruction:"
instruction = "### Instruction: " + instruction
return system + "\n" + instruction + "\n\n" + "### Solution:" + "\n"
def generate(
instruction,
max_new_tokens=128,
temperature=0.1,
top_p=0.75,
top_k=40,
num_beams=4,
**kwargs,
):
prompt = create_prompt(instruction)
print(prompt)
inputs = tokenizer(prompt, return_tensors="pt")
input_ids = inputs["input_ids"].to("cuda")
attention_mask = inputs["attention_mask"].to("cuda")
generation_config = GenerationConfig(
temperature=temperature,
top_p=top_p,
top_k=top_k,
num_beams=num_beams,
**kwargs,
)
with torch.no_grad():
generation_output = model.generate(
input_ids=input_ids,
attention_mask=attention_mask,
generation_config=generation_config,
return_dict_in_generate=True,
output_scores=True,
max_new_tokens=max_new_tokens,
early_stopping=True
)
s = generation_output.sequences[0]
output = tokenizer.decode(s)
return output.split("### Solution:")[1].lstrip("\n")
instruction = """
Edit the following XML code to add a navigation bar to the top of a web page
<html>
<head>
<title>CliBrAIn</title>
</head>
"""
print(generate(instruction))
✨ 主要特性
- 基於強大的Llama-2 7b模型,具備優秀的語言理解和生成能力。
- 使用QLoRA方法和PEFT庫在CodeAlpaca 20k指令數據集上進行微調,專注於編碼輔助場景。
📚 詳細文檔
模型描述 🧠
Llama-2是Meta開發並公開發布的一系列預訓練和微調的生成式文本模型,參數規模從70億到700億不等。其中,經過微調的Llama-2-Chat模型針對對話場景進行了優化,在大多數測試基準上優於開源聊天模型,在人工評估的有用性和安全性方面,與ChatGPT和PaLM等一些流行的閉源模型相當。
訓練和評估數據 📚
CodeAlpaca_20K數據集包含20K條指令跟隨數據,用於微調Code Alpaca模型。
訓練超參數 ⚙
optim="paged_adamw_32bit",
num_train_epochs = 2,
eval_steps=50,
save_steps=50,
evaluation_strategy="steps",
save_strategy="steps",
save_total_limit=2,
seed=66,
load_best_model_at_end=True,
logging_steps=1,
learning_rate=2e-4,
fp16=True,
bf16=False,
max_grad_norm=0.3,
warmup_ratio=0.03,
group_by_length=True,
lr_scheduler_type="constant"
訓練結果 🗒️
步驟 |
訓練損失 |
驗證損失 |
50 |
0.624400 |
0.600070 |
100 |
0.634100 |
0.592757 |
150 |
0.545800 |
0.586652 |
200 |
0.572500 |
0.577525 |
250 |
0.528000 |
0.590118 |
評估結果 📊
待完成
引用
@misc {manuel_romero_2023,
author = { {Manuel Romero} },
title = { llama-2-coder-7b (Revision d30d193) },
year = 2023,
url = { https://huggingface.co/mrm8488/llama-2-coder-7b },
doi = { 10.57967/hf/0931 },
publisher = { Hugging Face }
}
📄 許可證
本項目採用apache-2.0
許可證。