Skip to content

Preserve unfavorable trained rank0 derivative evidence #62

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
CompleteDotTech merged 1 commit into from
Oct 3, 2026
Merged

Preserve unfavorable trained rank0 derivative evidence #62

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions docs/CPU_STATE_OPTIMIZER.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,9 @@
# Explicit CPU optimizer state

## Trained rank0 derivative readout (October 2, 2026)

[Unfavorable derivative evidence](evidence/native/rank0-derivatives-failed-20261002.md) records one passing comparison out of five. Both HF comparisons and native shared-prefix/full gradient comparisons fail frozen tolerances. This independent derivative failure establishes no CPU-state optimizer arithmetic result; native acceptance remains incomplete.

`cpu_state_streaming_v1` is an experimental, explicitly selected CUDA AdamW
backend. It requires PyTorch `2.10.0+cu128` and trainable CUDA FP32 parameters,
including FP32 parameters used under BF16 autocast. Default and fused optimizer
Expand Down
4 changes: 4 additions & 0 deletions docs/RELEASE_READINESS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,9 @@
# Release readiness — intentionally incomplete

## Trained rank0 derivative readout (October 2, 2026)

[Unfavorable derivative evidence](evidence/native/rank0-derivatives-failed-20261002.md) records one passing comparison out of five. Both HF comparisons and native shared-prefix/full gradient comparisons fail frozen tolerances. This independent derivative failure establishes no CPU-state optimizer arithmetic result; native acceptance remains incomplete.

## Explicit CPU-state optimizer: partial qualification (September 30, 2026)

The opt-in backend at signed source `b5d8299b768564c09c541e3dfc04b3d28eef2fd6`
Expand Down
9 changes: 9 additions & 0 deletions docs/acceptance-status.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,13 @@
{
"rank0_derivative_failure_2026_10_02": {
"receipt": "docs/evidence/native/rank0-derivatives-failed-20261002.json",
"explanation": "docs/evidence/native/rank0-derivatives-failed-20261002.md",
"status": "completed_numerical_failure",
"passing_comparisons": 1,
"total_comparisons": 5,
"acceptance_claim": false,
"optimizer_arithmetic_claim": false
},
"explicit_CPU_state_optimizer_partial_delivery_2026_09_30": {
"signed_source_commit": "b5d8299b768564c09c541e3dfc04b3d28eef2fd6",
"scope": "partial synthetic tiny-fixture optimizer/trainer delivery; no native/fullweight/context/quality acceptance",
Expand Down
Loading
Loading