openrknn: document multi-core dispatch blocker for FP16 - #91
Merged
Merged
Conversation
Single-core NPU tasks only cover ~512 of 1024 spatial positions for SmolVLM transformer shards. The vendor runs all 3 NPU cores via kernel-level multi-core dispatch configured during rknn_init. openrknn's OWN mode skips vendor init, so core_mask=0 defaults to single core. Add TODO comment tracking this as the final blocker for correct FP16 output. The DMA address patching is now byte-exact (0 diffs) and the per- channel computation is verified correct (768/768 channel bias match). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Documents the final blocker for correct FP16 transformer output: the vendor's kernel-level multi-core dispatch.
Single-core tasks cover ~512 of 1024 spatial positions. The computation at covered positions is verified correct (768/768 channel bias match, byte-exact DMA addresses). The remaining 50% of positions are zero because multi-core dispatch isn't implemented in OWN mode.
No functional changes — just a TODO comment.
Test plan
🤖 Generated with Claude Code