[Primus] Merge Hybrid Models branch to Primus release v26.2 branch#559
Closed
clairesonglee wants to merge 27 commits intorelease/v26.2from
Closed
[Primus] Merge Hybrid Models branch to Primus release v26.2 branch#559clairesonglee wants to merge 27 commits intorelease/v26.2from
clairesonglee wants to merge 27 commits intorelease/v26.2from
Conversation
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
…32B Configs for MI300X & MI355X (#556) YF: Only SFT related config and Doc changes, bypassing unit CI tests ## Summary This PR introduces post-training documentation and updates Qwen3 32B model configuration files to support AMD MI300X and MI355X accelerators. --- ## Changes ### 📘 Documentation - **Added `posttraining.md`** - New comprehensive guide for post-training workflows - Covers setup instructions, configuration details, and usage examples - **Updated `docs/README.md`** - Added a new section referencing post-training documentation - Improved documentation organization and navigation --- ### ⚙️ Configuration Updates - **Updated Qwen3_32B model YAML configs** - Added/modified configurations optimized for: - MI300X - MI355X - Adjusted parameters for compatibility and stable execution --- ## Validation - Verified updated configs load and execute successfully on MI300X and MI355X environments - Confirmed documentation links and structure render correctly --- ## Checklist - [x] Added `posttraining.md` - [x] Updated `docs/README.md` - [x] Modified Qwen3_32B YAML configs - [x] Verified changes locally
) Override Megatron build_tokenizer to support custom tokenizer types with HuggingFace Hub IDs - Fixes Llama2Tokenizer failing with Hub IDs in new architecture - All custom types now work consistently in legacy and new architectures --------- Co-authored-by: HuangWei-95 <weihuan@amd.com> Co-authored-by: Xiaoming-AMD <Xiaoming.Peng@amd.com>
- Update MegatronPretrainTrainer.run_train() to detect model_type from backend_args - Conditionally import pretrain_mamba or pretrain_gpt based on model_type - Pass model_type to get_model_provider() to use correct builder (mamba_builder vs gpt_builder) - Restore core runtime support for megatron as intended by commit cfe8cc0 - Fixes 'specialize for HybridStack' error when using core runtime with hybrid models
add env for TestMegatronTrainerDeterministic ci test Co-authored-by: HuangWei-95 <weihuan@amd.com>
Update all references to the Primus Docker base image across documentation, configuration files, CI/CD workflows, and example scripts to use the latest v26.1 release.
…odel_type detection
clairesonglee
requested review from
Xiaoming-AMD and
wenxie-amd
as code owners
clairesonglee
force-pushed
the
hybrid/release/v26.2
branch
from
bc47a41 to
1808caa
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.