From NVIDIA Megatron-LM for visibility #18
Conversation
This file contains 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
…oder, encoder-decoder) to be compatible with all 3 TE backends Co-authored-by: Huy Vu2 <[email protected]> Co-authored-by: root <[email protected]>
Updating all T5 attention masks (encoder, decoder, encoder-decoder) to be compatible with all 3 TE backends See merge request ADLR/megatron-lm!2273
Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]> Co-authored-by: root <[email protected]>
Add hierarchical cp comm group See merge request ADLR/megatron-lm!2279
Add missing arg to save_checkpoint call See merge request ADLR/megatron-lm!2351
NVLM example scripts See merge request ADLR/megatron-lm!2306
ci: Re-enable llava tests See merge request ADLR/megatron-lm!2348
ci: Retry download assets See merge request ADLR/megatron-lm!2357
… ckpt-format when epp>1 Co-authored-by: Jon Barker <[email protected]>
Support etp==tp when epp==0 and enforce torch ckpt-format when epp>1 See merge request ADLR/megatron-lm!2260
Co-authored-by: Shanmugam Ramasamy <[email protected]>
QKNorm to work with TENorm See merge request ADLR/megatron-lm!2347
Support RMSNorm when TE and Apex are not installed See merge request ADLR/megatron-lm!2015
Clarifications for batch x pipeline parallel logic See merge request ADLR/megatron-lm!2343
…or TE cuDNN FusedAttention Co-authored-by: yaoyu-33 <[email protected]>
Add attention bias arg in MCore transformer for TE cuDNN FusedAttention See merge request ADLR/megatron-lm!2293
chore: Add mypy optionally See merge request ADLR/megatron-lm!2360
ci: JET improvements See merge request ADLR/megatron-lm!2365
Co-authored-by: Huy Vu2 <[email protected]>
… 'main' update golden values for nightly test See merge request ADLR/megatron-lm!2364
ci: Update golden values of nightlies See merge request ADLR/megatron-lm!2511
…r newly added requests
Make generate function only return results for newly added requests See merge request ADLR/megatron-lm!2370
ci: Use torchrun See merge request ADLR/megatron-lm!2507
chore: Fix local generator script See merge request ADLR/megatron-lm!2519
Co-authored-by: William Dykas <[email protected]> Co-authored-by: Mcore Bot <[email protected]> Co-authored-by: William Dykas <[email protected]> Co-authored-by: root <[email protected]>
Fix log probs output for inference See merge request ADLR/megatron-lm!2430
…ective=True Co-authored-by: Oliver Koenig <[email protected]>
Add tests for MoE models with average_in_collective=True See merge request ADLR/megatron-lm!2489
ci: Allow running nemo-ci See merge request ADLR/megatron-lm!2509
ci: Fail-fast on unit tests See merge request ADLR/megatron-lm!2520
remove tensorstore pin See merge request ADLR/megatron-lm!2516
ci: nemo-ci inputs See merge request ADLR/megatron-lm!2522
…ng to T5 Co-authored-by: Huy Vu2 <[email protected]>
Adding (bias-based) relative position embedding to T5 See merge request ADLR/megatron-lm!2428
Co-authored-by: Jimmy Zhang <[email protected]>
Inference CUDA graphs (MCore version) See merge request ADLR/megatron-lm!2429
…ayers Co-authored-by: Jon Barker <[email protected]> Co-authored-by: Jon Barker <[email protected]> Co-authored-by: Jon Barker <[email protected]>
Fix bug when loading pp>1 model with frozen layers See merge request ADLR/megatron-lm!2523
…token-drop and padding
Make MoE token dispatcher cuda graph-able if token-drop and padding See merge request ADLR/megatron-lm!2426
ci: Implement `frozen-ckpt` tests See merge request ADLR/megatron-lm!2514
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.