Xwin-LM/XwinCoder-13B

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Nov 13, 2023License:llama2Architecture:Transformer0.0K Open Weights Cold

XwinCoder-13B is a 13 billion parameter instruction-tuned code generation model developed by Xwin-LM, based on the CodeLLaMA architecture. It specializes in programming tasks, achieving 68.8 pass@1 on HumanEval and 60.1 pass@1 on MBPP. This model is designed for robust code generation across various benchmarks, offering strong performance for developers.

Loading preview...

XwinCoder-13B Overview

XwinCoder-13B is an instruction-tuned code generation model from Xwin-LM, built upon the CodeLLaMA architecture. It is part of a series of models, including 7B and 34B variants, specifically designed to excel in coding abilities. The 34B version has demonstrated performance comparable to GPT-3.5-turbo across six coding benchmarks, indicating the family's strong capabilities.

Key Capabilities

  • Code Generation: Optimized for generating code based on instructions.
  • Benchmark Performance: The 13B model achieves a HumanEval pass@1 score of 68.8 and an MBPP pass@1 score of 60.1.
  • Instruction Following: Fine-tuned to understand and execute coding instructions effectively.
  • Comprehensive Evaluation: Evaluated against mainstream coding leaderboards beyond just HumanEval, including MBPP and APPS, to demonstrate real-world applicability.

Good For

  • Software Development: Assisting developers with code generation tasks.
  • Prototyping: Quickly generating code snippets or functions.
  • Educational Tools: Supporting learning and practice in programming.
  • Research: Exploring advanced code generation techniques and model performance.