Skip to content

[WIP] IN MY LM trainer add more features fully simar to unsloth https://github.com/unslothai/unsloth see full repo codes and make lm trainer better based on that Finetune Gemma 3n, Qwen3, Llama 4, Phi-4 & Mistral 2x faster with 80% less VRAM! ✨ Fi...#3

Closed
Copilot wants to merge 1 commit into
mainfrom
copilot/fix-005fbb91-6dc2-4b7a-8640-90ffb993ea9e

Conversation

Copilot AI commented Jul 23, 2025

Copy link
Copy Markdown
  • Analyzed current LLM Trainer codebase structure
  • Identified missing Unsloth-like features for enhancement
  • Add quantization support (4-bit/8-bit) using bitsandbytes
  • Implement LoRA/QLoRA for parameter-efficient fine-tuning
  • Add support for latest model architectures (Gemma 3n, Qwen3, Llama 4, Phi-4)
  • Integrate TRL (Transformers Reinforcement Learning) support
  • Add memory optimization features and gradient checkpointing improvements
  • Implement export capabilities (GGUF, Ollama, vLLM formats)
  • Add pre-quantized model loading support
  • Enhance distributed training capabilities
  • Add advanced optimizers (8-bit AdamW)
  • Update documentation with new features

Goal: Transform LLM Trainer to match Unsloth's performance and memory efficiency features, focusing on 2x faster training with 80% less VRAM usage.

Created from VS Code via the GitHub Pull Request extension.


💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

@codeant-ai

codeant-ai Bot commented Jul 23, 2025

Copy link
Copy Markdown

CodeAnt AI is reviewing your PR.

@codeant-ai

codeant-ai Bot commented Jul 23, 2025

Copy link
Copy Markdown

CodeAnt AI finished reviewing your PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants