Shypd.aiThe largest open index of AI toolsMFTCoder is an open-source multi-task fine-tuning framework for Code LLMs that boosts accuracy and efficiency. It supports various models and fine-tuning methods like LoRA and QLoRA, enabling training with minimal resources.
Trending
About
MFTCoder is a high-accuracy and efficiency multi-task fine-tuning framework specifically designed for Code Large Language Models (LLMs). Developed by CodeFuse, an AI Native team within Ant Group, this open-source project facilitates accurate and efficient Multi-task Fine-tuning (MFT) on LLMs, with a particular focus on code-related tasks. It offers two main codebases: MFTCoder-accelerate, which leverages Accelerate and DeepSpeed/FSDP for robust training, and MFTCoder-atorch, based on the ATorch framework for fast distributed training. The framework supports multi-task learning, allowing models to be trained on diverse tasks while maintaining balance and even generalizing to new, unseen tasks. It integrates state-of-the-art open-source models and provides efficient fine-tuning options like LoRA, QLoRA, and full-parameter training, making it suitable for large models with minimal resources. MFTCoder also includes support for SFT (Supervised FineTuning) and releases high-quality code-related instruction datasets.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
