CodeQwen-7B
FreeReleased on December 2023
FreeFree tier
Inputs: textOutputs: text
About CodeQwen-7B
CodeQwen1.5-7B is a code-specific version of Qwen1.5, a transformer-based decoder-only language model pretrained on 3 trillion tokens of code. It supports a 64K token context length and 92 coding languages, and includes group query attention (GQA) for efficient inference. The model achieves competitive performance across code generation benchmarks, text-to-SQL, bug fixing, and other code-related tasks. It is available as an open-source model on Hugging Face, compatible with Transformers 4.37.0+ and suitable for fine-tuning, code infilling, and code generation.
Key Features
Strong code generation capabilities with competitive benchmark performance
Supports long context understanding and generation up to 64K tokens
Supports 92 coding languages
Group query attention (GQA) for efficient inference
Pretrained on 3 trillion tokens of code
Excellent performance on text-to-SQL and bug fixing tasks
Compatible with Hugging Face Transformers 4.37.0+
Supports fine-tuning, code infilling, and code generation
Open-source model available on Hugging Face
Pros & Cons
Pros
- Strong performance on code generation benchmarks
- Long 64K token context window enables handling large codebases
- Supports 92 programming languages
- Efficient inference via group query attention
- Open-source and freely available
- Compatible with standard Hugging Face tools and pipelines
Cons
- Not suitable for conversational chat out of the box
- Requires Transformers version >= 4.37.0
- Base model may need fine-tuning for specific applications
- Limited official documentation on the model page
Best For
Code generation and completionCode infillingText-to-SQL conversionBug fixing and code repairFine-tuning for domain-specific code tasksLong code understanding and analysis
FAQ
What is CodeQwen1.5-7B?
CodeQwen1.5-7B is a code-specific language model based on Qwen1.5, pretrained on 3 trillion tokens of code. It is a transformer-based decoder-only model with 7 billion parameters.
What context length does CodeQwen1.5-7B support?
It supports long context understanding and generation with a context length of 64K tokens.
How many coding languages does it support?
It supports 92 coding languages.
Is CodeQwen1.5-7B free to use?
Yes, it is an open-source model available for free on Hugging Face.
Can I use CodeQwen1.5-7B for chat?
No, it is a base language model designed for code tasks. The documentation advises against using it for chat and recommends fine-tuning for specific applications.