Instructions to use SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct")
model = AutoModelForCausalLM.from_pretrained("SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct

SGLang

How to use SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct with Docker Model Runner:
```
docker model run hf.co/SYSU-MUCFC-FinTech-Research-Center/ZhiLu-2-8B-Instruct
```

ZhiLu-2是一款基于llama3微调的中文消费金融领域对话大模型。与ZhiLu相比，ZhiLu-2性能有了显著提升。我们不仅收集了全新的高质量指令数据进行对齐，还创新性地设计了独特的数据合成方法，并将大量合成数据应用于模型训练。通过这些努力，ZhiLu-2在性能上取得了显著的突破，展示了卓越的性能。

训练细节在第二版的智鹿训练中，我们引入了全新的指令微调数据，并且加入了合成数据。我们相信，合成数据的使用将带来意想不到的惊喜效果。以下是一些重要的训练细节：

🚀 高效训练

我们使用llama-factory作为训练框架，并配备多块A100显卡，通过DeepSpeed（ds）实现数据并行、模型并行、管道并行和张量并行等优化技术。在微调方法的选择上，我们对Full-Rank FT、LORA、BAdam、LoRA+和DoRA进行了详细比较，评估了各方法在训练时间、显卡占用、推理时间和模型性能等多项指标上的表现。最终，我们决定采用DoRA进行微调，以获得最佳的性价比和性能。

⚡ 加速技术

为了提高资源的利用率并缩短训练时间，我们采用了以下两项关键技术：

Packing FlashAttention-2 🔒 安全性与对齐

我们使用 DPO 来训练模型。DPO具有使用便捷、成效快速的优势，可以达到近似RLHF的偏好对齐效果，确保输出的安全和无害。

🛡️ 避免灾难性遗忘

为了防止训练后模型的灾难性遗忘，并平衡模型在各个任务上的能力，我们使用了 merging 技术。

🌱 自我进化

通过设计新的框架，我们使模型能够自我生成训练数据，从而实现自我进化。

更多细节参见：ZhiLu-2-github仓库

Downloads last month: 24

Safetensors

Model size

8B params

Tensor type

BF16