OptimalScale
[New Feature][Roadmap requested]Do you have a roadmap right now?
Is your feature request related to a problem? Please describe.
I am very interested in this project, you guys did a great job. And I wonder if there is a roadmap provided?
For instance:
- Is there a plan to implement the following functionalities in this project: model compression, quantization, and pruning
- Is there a plan to support other fine-tuning algorithms, such as QLORA、freeze、p-tuning? ...
Describe the solution you'd like A clear and concise roadmap should be helpful.
I have created a PR for support for QLoRA (Added QLoRA support for Decoder transformers with tune_strategy "Normal" #613)
Once it is reviewed hopefully it get's merged.
Hi, Here are some important features we are going to implement:
- Efficient fine-tuning: e.g., p-tuning, and other adapters
- Efficient inference: e.g., speculative decoding, vllm
- Tool and Agent: e.g., ToolLLM
- Model compression: quantization, distillation, pruning
As for the timeline, we are working on Efficient inference and Model compression right now and plan to focus on fine-tuning and tool in Octomber.
Welcome to contribute to LMFlow! Thanks!
