Add LFM2 (hybrid short-conv) target support#28
Open
nathanrchn wants to merge 2 commits into
Open
Conversation
- Lfm2TargetOps in engine/target_lfm2.py mirrors target_qwen_gdn structure - ShortConvRollbackCache for conv-state rollback during speculation - patched ShortConv.__call__ records pre-conv tape when armed - forward path uses inner.fa_idx/conv_idx dual masks and embedding_norm - DFlashDraftModelArgs.from_dict reads rope_parameters (LFM2 config layout) - Lfm2TargetOps registered in TARGET_BACKENDS Verified on LiquidAI/LFM2.5-1.2B-Instruct + nathanrchn/LFM2.5-1.2B-Instruct-DFlash on Apple M4 Max with the README protocol (AIME prompt, 3 repeats, 60s cooldown): | Tokens | Baseline | DFlash | Speedup | Acceptance | |--------|----------------|----------------|---------|------------| | 1024 | 141.94 tok/s | 338.98 tok/s | 2.39x | 86.82% | | 2048 | 141.03 tok/s | 209.59 tok/s | 1.49x | 78.76% | The bundled draft was trained on 2k sequences; speedup degrades past that horizon as draft acceptance falls. Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
Two rows within the draft model's 2k training horizon. Footnote calls out the hardware difference (M4 Max vs M5 Max in the rest of the table) and the omitted 4k/8k rows. Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Verified on LiquidAI/LFM2.5-1.2B-Instruct + nathanrchn/LFM2.5-1.2B-Instruct-DFlash on Apple M4 Max with the README protocol (AIME prompt, 3 repeats, 60s cooldown):
The bundled draft was trained on 2k sequences; speedup degrades past that horizon as draft acceptance falls.