back

Qwen 2.5 72B

Qwen

Alibaba · 72B · Dense

Alibaba's flagship open model

581.6K downloads 916 likes 2024-09 128K context

Use Cases

chat multilingual reasoning code

Quantization Options

Quant Bits VRAM Quality Status
Q2_K 2 23.6 GB low
Q3_K_M 3 32.8 GB moderate
Q4_K_M 4 37.4 GB good
Q5_K_M 5 46.6 GB good
Q6_K 6 55.8 GB excellent
Q8_0 8 74.3 GB excellent
F16 16 148 GB lossless

About this model

Qwen2.5 is the latest series of Qwen large language models. For Qwen2.5, a range of base language models and instruction-tuned models are released, with sizes ranging from 0.5 to 72 billion parameters. Qwen2.5 introduces the following improvements over Qwen2:

  • It possesses significantly more knowledge and has greatly enhanced capabilities in coding and mathematics, due to specialized expert models in these domains.
  • It demonstrates significant advancements in instruction following, long-text generation (over 8K tokens), understanding structured data (e.g., tables), and generating structured outputs, especially in JSON format. It is also more resilient to diverse system prompts, improving role-play and condition-setting for chatbots.
  • It supports long contexts of up to 128K tokens and can generate up to 8K tokens.
  • It offers multilingual support for over 29 languages, including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, Arabic, and more.

Please note: all models except the 3B and 72B are released under the Apache 2.0 license, while the 3B and 72B models are under the Qwen license.

References

GitHub

Blog post

HuggingFace