Scholar·PuYu large model full-link open source open system
Large models become popular keywords
Large models have become an important way to develop general artificial intelligence
The open source history of Shusheng·Puyu large model
Scholar Puyu 20B open source large model performance
- Comprehensively leading open source models of similar magnitude (including Llama-33B, Llama2-13B and domestic mainstream 7B and 13B open source models)
- Reaching Llama2-70B level with less than one-third of the parameters
From model to application
Scholar·PuYu full-link open source open system
data
Total data volume 2TB
includes:
- Text data 5 billion documents
- Image-text dataset 22 million files
- Video data exceeds 1000 files
pre-training
fine-tuning
- Incremental training
- Vertical domain knowledge articles, books, codes, etc.
- Supervised fine-tuning
- High-quality dialogue and question-and-answer data
- High-quality dialogue and question-and-answer data
Review
Proposed the OpenCompass evaluation system
in 6 major dimensions with 80+ evaluation sets and 400,000+ evaluation questions
deploy
agent
Limitations:
- Access to the latest information and knowledge
- reliability of response
- mathematical calculations
- Tool usage and interaction