Use FVP NPU output as the conlarge_xl golden - #11
Merged
Merged
Conversation
Replace the conlarge_xl sidecars with an int16 input generated from seed 22 and the Ethos-U85 output captured on the Corstone-320 FVP via the evaluation kit's inference_compare use case (Vela 5.2.0, AmbiqHP_HBLRAM, Dedicated_Sram_1MB, 3 MB arena). The NPU output differs from host TFLite in 204 of 76875 elements; per-frame argmax is unchanged. The command stream is byte-identical, but 12 int16 lookup-table entries in the weight blob move by one LSB relative to the previous build, which came from a different host. The golden was captured with these weights, so the npz and weights.txt are regenerated with it.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Same treatment as #10 did for efficientnet.
Weights note: the command stream is byte-identical to main, but 12 int16 lookup-table entries in the weight blob differ by one LSB from the previous build, which came from a different host. Vela is deterministic on the machine used here. The golden was captured with these weights, so conlarge_xl_vela.npz and src/conlarge_xl_weights.txt are regenerated and must be loaded together with the new input and golden.
conlarge_xl_weights.h stays untracked, as on main.