8000 feat: MS-Swift Megatron by bradhilton · Pull Request #407 · OpenPipe/ART · GitHub
[go: up one dir, main page]

Skip to content

Conversation

bradhilton
Copy link
Collaborator

No description provided.

cursoragent and others added 6 commits September 16, 2025 00:06
Configure Qwen3 MoE for LoRA SFT with Megatron-SWIFT.

Co-authored-by: bhilton <bhilton@wandb.com>
Enhanced the setup and run sections in config.yaml by adjusting indentation and ensuring proper execution of Python scripts for data generation and model testing. This improves readability and maintainability of the configuration.
Modified config.yaml to rename the model and adjust GPU resources. Added a new Dockerfile to set up the environment with necessary dependencies for ms-swift-megatron, including SSH server configuration and preinstalled packages for SkyPilot.
Modified to-hf.sh, to-mcore.sh, and train.sh to use the new model version Qwen3-235B-A22B-Instruct-2507 and expanded CUDA_VISIBLE_DEVICES to include more GPUs. Adjusted dataset and training parameters in train.sh for improved performance.
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
0