Repository navigation
feat(WO79): fp16->fp32 graph-IO normalizer + normalized artifact digests - #327
Merged
Merged
Conversation
added 3 commits
October 21, 2026 13:45
Wraps fp16 graph outputs in Cast(to=FLOAT), feeds fp32 decoder inputs through an internal Cast(to=FLOAT16), keeps the raw fp16 tensors internal (one fp32 output per original output), and recomputes the metadata.yaml member digests. Verified: the normalized fas artifact decodes the frozen fixtures 12/12 byte-identically through the reference path.
Both fp16 artifacts re-normalized (Cast-to-fp32 graph IO; weights untouched; frozen fixtures reproduced 12/12 byte-identically through the reference path). Also fixes the urd entry, whose recorded digest matched neither the old nor the new asset.
…digest + normalized size
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Implements the owner-approved WO79 fix: tools/normalize_fp16_io.py wraps fp16 graph outputs in Cast(to=FLOAT) and feeds fp32 decoder inputs through an internal cast -- weights stay fp16 (size unchanged), all graph IO becomes fp32, every runtime executes the reference math. Both fp16 artifacts normalized and verified: frozen fixtures reproduce 12/12 byte-identically through the reference path (fas and urd). models.yaml digests/sizes updated (also fixes the urd entry, which carried a doubled-digest concatenation matching neither asset). Published index-v12 yaml + sidecar already repaired server-side with the same digests.