[sergqwer]

Under STRATA_PF_FUSED=1, the shipped IQ2_XS pack gives a wrong answer after a long prompt. The prompt path writes past its MoE buffers.

Cause (src/prefill/prefill.cpp):

• fused_layout() switches a streamed chunk (T >= stream_all_min()) to the fused path's smaller MoE buffers when any layer can run the fused kernels. That is the fused_ring() test.
• In moe_bufs(..., fused=true), GU / H / Xq / Hq then hold MMQ rows for only stream_all_min() - 1 tokens.
• The path is chosen per layer, though (fused_l / fused_nat). A layer the native kernels do not cover runs MMQ or the FP16 path over the chunk's T x K rows, and runs on into the next buffers.
• The IQ2_XS pack is such a pack. Its 45 IQ2_XXS / IQ2_S layers are covered; its three IQ1_M layers are not, and take the FP16 path. Nothing is reported, and the MoE output is silently corrupted.

Fix: the smaller layout only when every layer takes the fused path: each layer is in the MMQ plan and, for a native pack, fused::native_supported().

• A mixed pack keeps MMQ-sized buffers. Its covered layers still run the fused kernels in them: the log still says "prompt experts on the fused int8 kernels".
• A pack whose every layer is covered, and the Q2_0 pack, keep the smaller layout. Nothing changes for them.
• Without STRATA_PF_FUSED=1 nothing changes at all.

Measured on an RTX 5090: IQ2_XS (ISTA), 262K, --expert-cache 14900, first-token logits. The reference is the same engine without the flag (MMQ):

prompt	v0.1.37, STRATA_PF_FUSED=1	this PR, STRATA_PF_FUSED=1
2K (1 chunk)	KL 0.0039, max abs diff 2.23	KL 0.0014, max abs diff 0.36
32K (4 chunks of 8192)	KL 3.79, a different top token, max abs diff 9.56	KL 0.0036, same top token, max abs diff 0.69
prompt time, 32K	5,676 ms	5,134 ms

The remaining KL is the fused kernels' own rounding against MMQ, which the release notes put at MMQ's level. #519 runs STRATA_PF_FUSED=1 on an IQ3_XXS pack in production: if that pack mixes formats by layer, its long prompts were affected too.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VZy1yKaDDiA8a7svdwaHio

[sergqwer]

0.1.38's 220e0e8 fixes this in fused_ring(): the native kernels must cover every layer, the same condition this PR put in fused_layout().

Measured on v0.1.38 (RTX 5090, the IQ2_XS pack with three IQ1_M layers, fixed --expert-cache 14900), first-token logits with STRATA_PF_FUSED=1 against the default MMQ path:

prompt	0.1.37	0.1.38
2K	(not measured)	KL 0.0014, same top token
32K	KL 3.79, different top token	KL 0.0036, same top token

0.1.38's 0.0036 is the same as this branch gave. I'm closing this as superseded.