Benjamin Marie(@bnjmn_marie):"Full" NVFP4 Qwen3.8 27B: https://t.co/ucgA6Mg1yd Did you try this? I can't trust the eval. LiveCodeBench results for the BF16 baseline are 10~15 points below what they should be. Same for AIME. Evaluated with a low context length (32K), so in a condition where seeing quantization degradation is unlikely Just wondering whether I should spend one or two days on an RTX Pro 6000 to confirm, but at native context length.

Benjamin Marie

@bnjmn_marie

Independent AI researcher (LLM, NLP). My blog, The Kaitchup - AI on a Budget:

参加 June 2019

221 フォロー中 6.9K ファン

Benjamin Marie@bnjmn_marie

2026.09.09 19:56

"Full" NVFP4 Qwen3.8 27B: Did you try this? I can't trust the eval. LiveCodeBench results for the BF16 baseline are 10~15 points below what they should be. Same for AIME. Evaluated with a low context length (32K), so in a condition where seeing quantization degradation is unlikely Just wondering whether I should spend one or two days on an RTX Pro 6000 to confirm, but at native context length.