🤗 HuggingFace • 🤖 ModelScope • ✡️ WiseModel
👾 Discord • 🐤 Twitter • 💬 WeChat
📝 Paper • 🙌 FAQ • 📗 Learning Hub

• Intro
• News
• Quick Start
• Web Demo
• Deployment
• Fine-tuning
• API
• License

Yi-1.5 is an upgraded version of Yi. It is continuously pre-trained on Yi with a high-quality corpus of 500B tokens and fine-tuned on 3M diverse fine-tuning samples.

Compared with Yi, Yi-1.5 delivers stronger performance in coding, math, reasoning, and instruction-following capability, while still maintaining excellent capabilities in language understanding, commonsense reasoning, and reading comprehension.

Yi-1.5 comes in 3 model sizes: 34B, 9B, and 6B. For model details and benchmarks, see Model Card.

• 2024-05-13: The Yi-1.5 series models are open-sourced, further improving coding, math, reasoning, and instruction-following abilities.

• Make sure Python 3.10 or a later version is installed.

• Set up the environment and install the required packages.

  pip install -r requirements.txt

• Download the Yi-1.5 model from Hugging Face, ModelScope, or WiseModel.

This tutorial runs Yi-1.5-34B-Chat locally on an A800 (80G).

from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = '<your-model-path>'

tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False)

# Since transformers 4.35.0, the GPT-Q/AWQ model can be loaded using AutoModelForCausalLM.
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",
    torch_dtype='auto'
).eval()

# Prompt content: "hi"
messages = [
    {"role": "user", "content": "hi"}
]

input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(input_ids.to('cuda'), eos_token_id=tokenizer.eos_token_id)
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)

# Model response: "Hello! How can I assist you today?"
print(response)

Prerequisites: Before deploying Yi-1.5 models, make sure you meet the software and hardware requirements.

Prerequisites: Download the lastest version of vLLM.

1. Start the server with a chat model.

   python -m vllm.entrypoints.openai.api_server  --model 01-ai/Yi-1.5-9B-Chat  --served-model-name Yi-1.5-9B-Chat

2. Use the chat API.

• HTTP

  curl https://localhost:8000/v1/chat/completions \
      -H "Content-Type: application/json" \
      -d '{
          "model": "Yi-1.5-9B-Chat",
          "messages": [
              {"role": "system", "content": "You are a helpful assistant."},
              {"role": "user", "content": "Who won the world series in 2020?"}
          ]
      }'

• Python client

  from openai import OpenAI
  # Set OpenAI's API key and API base to use vLLM's API server.
  openai_api_key = "EMPTY"
  openai_api_base = "https://localhost:8000/v1"
  
  client = OpenAI(
      api_key=openai_api_key,
      base_url=openai_api_base,
  )
  
  chat_response = client.chat.completions.create(
      model="Yi-1.5-9B-Chat",
      messages=[
          {"role": "system", "content": "You are a helpful assistant."},
          {"role": "user", "content": "Tell me a joke."},
      ]
  )
  print("Chat response:", chat_response)

We have deployed a Yi-1.5-34B-Chat Space on Huggingface.

Or you can build it locally by yourself, as follows:

python demo/web_demo.py -c <your-model-path>

You can use LLaMA-Factory, Swift, XTuner, and Firefly for fine-tuning. These frameworks all support fine-tuning the Yi series models.

Yi APIs are OpenAI-compatible and provided at Yi Platform. Sign up to get free tokens, and you can also pay-as-you-go at a competitive price. Additionally, Yi APIs are also deployed on Replicate and OpenRouter.

The code and weights of the Yi-1.5 series models are distributed under the Apache 2.0 license.

[ Back to top ⬆️ ]