Qwen3.8-27B NVFP4 quants compared: lm_head precision makes or breaks MTP speedups
danielhanchen · x · 2026-09-14
- Daniel Han of Unsloth compared multiple NVFP4 quantizations of Qwen3.8-27B: accuracy is close across variants; the real differences are memory and speed.
- For VRAM-limited setups, minima-ai/mnmaqwen3.827bnvfp4 is a good pick but lacks MTP. NVIDIA's variant includes MTP, but in long-context coding tests MTP-4 is only 2x faster than no MTP and still 2.5–3x slower than Unsloth on an RTX Pro 6000.
- Likely culprit: NVIDIA quantizes lmhead to NVFP4 while Unsloth keeps it FP8. Since MTP shares the target model's lmhead, this can hurt prediction quality and acceptance rate.
- Verdict: Unsloth NVFP4 is the pick; long-horizon agentic coding accuracy tests are ongoing.
More from Infra
- Your AI agent would cost 260% more without prompt caching, Arena data shows — arena · 2026-10-30
- Nebius runs auction-driven GPU pricing, lifting rates ~20% amid tight supply to 2028 — BenBajarin · 2026-10-30
- macOS update silently re-adds eMudhra root certs users had removed — steipete · 2026-10-30
- Bittensor's Terza Inference routes one API across miners' models on subnet 103 — const_reborn · 2026-10-30
- SGLang Summit 2026 sells out sponsor booths and Full Passes ahead of Nov 12–13 event — ying11231 · 2026-10-30
- OpenRouter rebuilds provider routing: 46% fewer capacity errors, 16.9% faster TTFT — steipete · 2026-10-30