Fetch from nvidia Megatron-LM #5
Open
Conversation
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
feat(MoE): FP8 Support for Multi-Token-Prediction See merge request ADLR/megatron-lm!2950
Fix checkpoint directory bug in distill nightly test Closes #446 See merge request ADLR/megatron-lm!3096
[dist ckpt] Re-attempt !2493 + fixing merge conflicts See merge request ADLR/megatron-lm!2637
ci: Control which checks per test to run See merge request ADLR/megatron-lm!3175
Fix the sync issue in `TemporalAsyncWorker` See merge request ADLR/megatron-lm!3155
Co-authored-by: Chenhan Yu <[email protected]> Co-authored-by: Chen-Han Yu <[email protected]> Co-authored-by: Ye Yu <[email protected]>
Add ModelOpt speculative decoding finetune See merge request ADLR/megatron-lm!2971
Co-authored-by: yaoyu-33 <[email protected]> Co-authored-by: Mcore Bot <[email protected]> Co-authored-by: Chenhan Yu <[email protected]>
Moe fix for Llama4 See merge request ADLR/megatron-lm!3083
…DeepSeek-v3 Co-authored-by: jianbinc <[email protected]>
[custom FSDP] Support EP + FSDP training for DeepSeek-v3 See merge request ADLR/megatron-lm!2910
Fix extra tokens in returned generation Closes dl/JoC/nemo-ci#2075 See merge request ADLR/megatron-lm!3178
Update current scaling supported TE version to 2.2.0.dev0 See merge request ADLR/megatron-lm!3160
Co-authored-by: Shanmugam Ramasamy <[email protected]> Co-authored-by: Mcore Bot <[email protected]> Co-authored-by: Shanmugam Ramasamy <[email protected]> Co-authored-by: Shanmugam Ramasamy <[email protected]> Co-authored-by: Vijay Korthikanti <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]>
Seperate chunk allocator See merge request ADLR/megatron-lm!3121
…inference_context.sequence_len_offset > 0
Revert inference_context.is_decode_only() to inference_context.sequence_len_offset > 0 See merge request ADLR/megatron-lm!3180
…-fusion will throw an exception when topk/num_local_experts is not the power of 2.
[BUG FIX]: fix the bug of indices-to-multihot-fusion will throw an exception when topk/num_local_experts is not the power of 2. See merge request ADLR/megatron-lm!3058
…g global ones with optional local ones for better parallelism flexibility Co-authored-by: Zhiyu Li <[email protected]>
Refactor Inference Process Groups by replacing global ones with optional local ones for better parallelism flexibility See merge request ADLR/megatron-lm!3015
Update te patch to include 1626 See merge request ADLR/megatron-lm!3179
Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]>
Update dataset helper for online video decoding See merge request ADLR/megatron-lm!3367
Do not use eval on arbitrary user input. See merge request ADLR/megatron-lm!3365
tests: Update frozen-checkpoints See merge request ADLR/megatron-lm!3363
…eration Co-authored-by: root <[email protected]>
Consolidate eval methods across train and generation See merge request ADLR/megatron-lm!3375
ci: Auto-restart on nan See merge request ADLR/megatron-lm!3388
…YARN embedding cache Co-authored-by: xuwenc <[email protected]>
perf(mla, experimental): MLA RoPE fusion and YARN embedding cache Closes #429 See merge request ADLR/megatron-lm!2949
Co-authored-by: jianbinc <[email protected]>
Fix custom FSDP float8 tensor set_item See merge request ADLR/megatron-lm!3280
ci: Move queue blocker See merge request ADLR/megatron-lm!3401
Co-authored-by: Mcore Bot <[email protected]>
ci: Improve error-handling of missing logs See merge request ADLR/megatron-lm!3400
Co-authored-by: Mcore Bot <[email protected]>
ci: Control job concurrency See merge request ADLR/megatron-lm!3408
ci: Catch missing logs See merge request ADLR/megatron-lm!3412
ci: Remove tests from A100 See merge request ADLR/megatron-lm!3411
…of ChainedOptimizer
Add an option to skip counting zeros in grad of ChainedOptimizer See merge request ADLR/megatron-lm!3393
Add an interface to set high priority stream groups See merge request ADLR/megatron-lm!3326
Co-authored-by: Chen-Han Yu <[email protected]> Co-authored-by: Chenhan Yu <[email protected]>
Llama4 inference See merge request ADLR/megatron-lm!3241
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.