🚀 Huggingface-TogetherAI LangChain Wrapper

A LangChain integration for DeepSeek-R1 and Meta Llama-3.3-70B-Instruct-Turbo models via Hugging Face's Inference API, enabling seamless interaction with state-of-the-art language models.

✨ Features

• 🚀 Custom LangChain Chat Model – Optimized for Hugging Face + Together AI.
• ⚡ Sync & Async Support – Run queries in synchronous or asynchronous mode.
• 🌊 Streaming Capabilities – Supports token streaming for real-time responses.
• 🛠️ Tool Calling & Structured Output – Enables function calling and JSON outputs.
• 🔧 Configurable Model Parameters – Fine-tune temperature, max tokens, etc.

📦 Installation

pip install huggingface-togetherai 

🚀 Quick Start

from huggingface_togetherai import ChatHuggingFaceTogetherAI

hf_token = "your_huggingface_token"
hf_llm = ChatHuggingFaceTogetherAI(
    model="deepseek-ai/DeepSeek-R1",
    hf_token=hf_token
)

response = hf_llm.invoke("Hi!")
print(response)

🤔 Why Use Huggingface-TogetherAI?

In LangChain, the HuggingFaceEndpoint class is typically used for Hugging Face models:

from langchain_huggingface import HuggingFaceEndpoint
from langchain_huggingface.chat_models import ChatHuggingFace

hf_endpoint = HuggingFaceEndpoint(
    repo_id="deepseek-ai/DeepSeek-R1",
    task="text-generation",
    huggingfacehub_api_token=hf_token
)

langchain_llm = ChatHuggingFace(llm=hf_endpoint)
langchain_llm.invoke("Hello")

However, this results in an error:

The model deepseek-ai/DeepSeek-R1 is too large to be loaded automatically (688GB > 10GB).

✅ The Better Alternative: Huggingface-TogetherAI

With Huggingface-TogetherAI, you can seamlessly use large models without running into memory issues:

from huggingface_togetherai import ChatHuggingFaceTogetherAI

hf_llm = ChatHuggingFaceTogetherAI(
    model="deepseek-ai/DeepSeek-R1",
    hf_token=hf_token,
    other_params...
)

response = hf_llm.invoke("Hello")
print(response)  # Output: '<think>\n\n</think>\n\nHello! How can I assist you today? 😊'

🎉 Good News!

✅ You can leverage all Langchain functionalities for standard LLMs with this package.

📜 License

This project is licensed under the MIT License. See the LICENSE file for more details.