Publication summary: Huihui Qwen3.8 27B NInfer NVFP4 scout

This derivative copy does not replace the dated finding or native evidence.

Canonical facts

• Model identity: lyf/Qwen3.8-27B-Huihui-Abliterated-NInfer-NVFP4@181446902fc777c479749e98cf2abf2250263a8d.
• Runtime identity: NInfer a99407c63fc5bbd25d9fb597cbb8ab352bdb01ef; exact CUDA 13.1.2 image digest in configuration.
• Local setup: RTX 5090, isolated Windows/WSL2 direct lane, NVFP4 no-spec, 8K/C1.
• Headline result: strict tools failed 0/3 at default and low-effort diagnostics; default diff also failed 0/3.
• Capability result: corrected --vision image-only corpus passed 12/12; earlier 0/12 vision_disabled is retained.
• Important caveat: no matched speed, footprint, or context comparison; immutable runtime recreation is unproven. Managed restoration is verified.
• Decision: rejected, no-promotion, promoted=false.
• Artifact set: manifest and evidence index.

X / short post

RTX 5090 Huihui Qwen3.8 NInfer 8K profile: corrected vision passed 12/12, but strict tools were 0/3 at both tested settings. Exact tested profile rejected; no promotion or performance claim. Evidence: docs/findings/2026-09-19-qwen38-huihui-ninfer.md

Reddit

RTX 5090 Huihui Qwen3.8 NInfer scout: vision passes, strict tools fail

Pinned Huihui Qwen3.8 NInfer NVFP4 at 8K/C1 passed a corrected 12/12 image-only subset, while strict tools failed 0/3 under both tested settings. This is a bounded local scout, not a speed, footprint, or model-ranking claim. The exact tested profile is rejected with no promotion.

Screenshot alt text

RTX 5090 Huihui Qwen3.8 NInfer NVFP4 8K/C1 scout: corrected vision passed 12 of 12 images, while strict tools failed all three attempts at both settings; no promotion.

Claim ledger

Public claim	Conditions	Evidence
Strict tools failed 0/3 at both settings	8K/C1, three attempts each	default; low
Vision-enabled image subset passed 12/12	six cases, two attempts each	artifact
No performance or footprint claim	no matched capacity artifacts	coverage