Preserve pinned Qwen4Exp FP8 checkpoint with QDQ - #658

Merged

justinchuby merged 22 commits into

mainfrom

justinchuby-integrate-qwen4exp-fp8

Aug 27, 2026

Commits on Aug 27, 2026

Support block-scaled FP8 weights

justinchuby
and
Copilot
committed
2b94ea7
Integrate pinned Qwen4Exp FP8 checkpoint
justinchuby
committed
ede909b
Harden Qwen4Exp FP8 schema validation
justinchuby
committed
80b30b5
Prove FP8 parity and bound serialization memory
justinchuby
committed
4402d59
Fix Qwen4Exp PLE scaling and CI coverage
justinchuby
committed
a4f322b
Resolve Qwen4Exp multimodal restack
justinchuby
committed
02a3f35
Tighten pinned FP8 and memory contracts
justinchuby
committed
6f0b40e
Preserve GGUF indexer preprocessing after restack
justinchuby
committed
05fe63d
Stream packed Qwen4Exp FP8 expert banks
justinchuby
committed
ce0cf15
Align BF16 streamer test with sharded PLE graph
justinchuby
committed
8184a0f
Dispatch Qwen4Exp experts without matrix replication
justinchuby
committed
322b74a
Preserve Qwen4Exp FP8 storage with QDQ
justinchuby
committed
a3ff5b7
Close Qwen4Exp FP8 QDQ review gaps
justinchuby
committed
2999416
Test QDQ cast-overlap package accounting
justinchuby
committed
358cd6e
Finalize Qwen4Exp QDQ review attribution

justinchuby
and
Copilot
committed
c55b8b1
Resolve FP8 QDQ integration on merged main
justinchuby
committed
940f954
Make FP8 fallback and report saves transactional

justinchuby
and
Copilot
committed
797a830
Preserve parent FP8 metadata for text-only Qwen4Exp

justinchuby
and
Copilot
committed
c531a52
Preserve parent FP8 routing and bound Qwen graphs

justinchuby
and
Copilot
committed
26967c2
Bound Qwen4Exp FP8 PLE and MoE execution

justinchuby
and
Copilot
committed
2df2970
Make PLE QDQ metadata and ORT gating exact

justinchuby
and
Copilot
committed
c3e3cdf
Require exact ORT FP8 capability exception

justinchuby
and
Copilot
committed
1e483bf

Navigation Menu

Uh oh!

Preserve pinned Qwen4Exp FP8 checkpoint with QDQ - #658

Preserve pinned Qwen4Exp FP8 checkpoint with QDQ#658
justinchuby merged 22 commits into
mainfrom
justinchuby-integrate-qwen4exp-fp8

Commits on Aug 27, 2026

Support block-scaled FP8 weights

Integrate pinned Qwen4Exp FP8 checkpoint

Harden Qwen4Exp FP8 schema validation

Prove FP8 parity and bound serialization memory

Fix Qwen4Exp PLE scaling and CI coverage

Resolve Qwen4Exp multimodal restack

Tighten pinned FP8 and memory contracts

Preserve GGUF indexer preprocessing after restack

Stream packed Qwen4Exp FP8 expert banks

Align BF16 streamer test with sharded PLE graph

Dispatch Qwen4Exp experts without matrix replication

Preserve Qwen4Exp FP8 storage with QDQ

Close Qwen4Exp FP8 QDQ review gaps

Test QDQ cast-overlap package accounting

Finalize Qwen4Exp QDQ review attribution

Resolve FP8 QDQ integration on merged main

Make FP8 fallback and report saves transactional

Preserve parent FP8 metadata for text-only Qwen4Exp

Preserve parent FP8 routing and bound Qwen graphs

Bound Qwen4Exp FP8 PLE and MoE execution

Make PLE QDQ metadata and ORT gating exact

Require exact ORT FP8 capability exception