Convergence record (from issue 5 / PR #6, per the 08:51 directive): the online-FP8 collapse isolated here and the offline int8-wo collapse from issue 5 are one kernel family's defect — Marlin…
Round two (amendment response) — the decisive experiment is a YES: W8A8 int8-dynamic delivers the capacity route on one 3090.
W8A8-dynamic artifact (cutlass int8 scaled-MM, no Marlin):…
e37c5aed40
probe(evidence): round two - W8A8 int8-dynamic passes the structure gate at BF16-parity PSNR; verdict flips to YES; upstream Marlin report draft staged
704b8e29ae
probe(round two): W8A8 int8-dynamic scheme — kernel isolation, raw-op probe, artifact builder support
Amendment to the 08-07 directive, given PR #6's verdict: the isolation experiment already ran and eliminated the directive's stated hypotheses — NaN occurs with synthetic int8 weights (not…
Implemented in PR: anodyine/vllm-omni#6
Verdict in brief: the int8 weight-only compressed-tensors artifact delivers the 3-4 reference capacity and speed (4-ref…
bf65a41ceb
probe(evidence): int8-wo capacity matrix, structure gate, PSNR, Marlin 8-bit NaN isolation (issue 5)
Directive after the int8-wo result (2026-10-08, owner-approved): worth one more bounded round — pivot from quantizer variants to kernel-path isolation.
The convergence is the finding: online-FP8…
321e8bb47f
fix(quant): load offline-quantized compressed-tensors checkpoints directly on CPU under offload
2f8363c5fc
fix(quant): resolve hyphenated registry methods and disk-dict from_config in quant factory
Correction from issue #2's probe (PR #4, commit 4f5f99d8): reinspecting this issue's FP8 artifacts shows both qwen_image_21_fp8_all_rtx3090.png and qwen_image_21_fp8_imgmlp_rtx3090.png are…