Awesome Chinese LLM

URL: https://github.com/HqWu-HITCS/Awesome-Chinese-LLM
Category: Themed Directories
Tags: ai, llm, open-source

Overview

Awesome Chinese LLM is a curated, "awesome-style" directory of open-source Chinese large language model resources. It emphasizes smaller-scale models that are feasible for private deployment and lower-cost training, and aggregates base models, domain-specific fine-tunes, applications, datasets, and tutorials related to Chinese LLMs.

Features

1. Focus and Scope

• Concentrates on Chinese-language LLMs and related tooling.
• Prioritizes smaller-scale models that:
  • Can be run by individuals or small teams.
  • Are suitable for private deployment.
  • Have lower training and deployment costs.
• Includes resources across the full ecosystem:
  • Base / foundation models.
  • Domain-specific fine-tuned models.
  • Applications built on top of LLMs.
  • Datasets for training and evaluation.
  • Tutorials and learning materials.

2. Base Model Overview Table

Provides a comparative overview of commonly used base models, including:

• ChatGLM family

  • Variants: ChatGLM / ChatGLM2 / ChatGLM3 / ChatGLM4 (Base & Chat)
  • Parameters: ~6B
  • Training tokens: ~1T / 1.4T
  • Max sequence length: 2K / 32K
  • Commercial use: Allowed

• LLaMA family

  • Variants: LLaMA / LLaMA2 / LLaMA3 (Base & Chat)
  • Parameters: 7B / 8B / 13B / 33B / 70B
  • Training tokens: ~1T / 2T
  • Max sequence length: 2K / 4K
  • Commercial use: Partially allowed (depends on version/license)

• Baichuan

  • Variants: Baichuan / Baichuan2 (Base & Chat)
  • Parameters: 7B / 13B
  • Training tokens: ~1.2T / 1.4T
  • Max sequence length: 4K
  • Commercial use: Allowed

• Qwen (通义千问)

  • Variants: Qwen / Qwen1.5 / Qwen2 / Qwen2.5 (Base, Chat, VL)
  • Parameters: 7B / 14B / 32B / 72B / 110B
  • Training tokens: ~2.2T / 3T / 18T
  • Max sequence length: 8K / 32K
  • Commercial use: Allowed

• BLOOM

  • Variants: BLOOM
  • Parameters: 1B / 7B / 176B-MT
  • Training tokens: ~1.5T
  • Max sequence length: 2K
  • Commercial use: Allowed

• Aquila

  • Variants: Aquila / Aquila2 (Base / Chat)
  • Parameters: 7B / 34B
  • Max sequence length: 2K
  • Commercial use: Allowed

• InternLM

  • Variants: InternLM / InternLM2 / InternLM2.5 (Base / Chat / VL)
  • Parameters: 7B / 20B
  • Max sequence length: up to 200K
  • Commercial use: Allowed

• Mixtral

  • Variants: Base & Chat
  • Parameters: 8×7B (Mixture-of-Experts)
  • Max sequence length: 32K
  • Commercial use: Allowed

• Yi

  • Variants: Base & Chat
  • Parameters: 6B / 9B / 34B
  • Training tokens: ~3T
  • Max sequence length: up to 200K
  • Commercial use: Allowed

• DeepSeek

  • Variants: Base & Chat
  • Parameters: 1.3B / 7B / 33B / 67B
  • Max sequence length: 4K
  • Commercial use: Allowed

• XVERSE

  • Variants: Base & Chat
  • Parameters: 7B / 13B / 65B / A4.2B
  • Training tokens: ~2.6T / 3.2T
  • Max sequence length: 8K / 16K / 256K
  • Commercial use: Allowed

3. Structured Directory of Resources

The repository is organized as an "awesome list" with a table of contents that includes (among others):

• 1. 模型 (Models)
  • 1.1 文本 LLM 模型 (Text LLMs)
  • 1.2 多模态 LLM 模型 (Multimodal LLMs)

(Additional sections for applications, datasets, and tutorials exist but are not fully visible in the provided excerpt.)

4. Scale and Community Activity

• Tracks and collects 100+ Chinese LLM-related open-source resources.
• Hosted as a public GitHub repository, allowing community contributions via pull requests.
• As of the snapshot, shows significant community interest (stars and forks), indicating active maintenance and ecosystem relevance.

5. Contribution Guidelines (High-Level)

• Encourages contributions of:
  • New open-source models.
  • Applications built on Chinese LLMs.
  • Datasets and tutorials.
• Requests contributors to follow a consistent format, including:
  • Repository link.
  • Star count.
  • Concise introduction/description.

Pricing

• This is an open-source, free GitHub directory.
• No pricing plans or paid tiers are indicated.