HF's merve shares eval tip: 9B bf16 needs 17.9GB VRAM, use Gemma on screenshots
mervenoyann · x · 2026-09-07
Hugging Face's mervenoyann answered a question about eval compute requirements: it depends on Qwen3.5's size — a 9B model in bf16 needs 17.9GB VRAM.
Her shared trick: instead of evaluating glb files directly, pass screenshots of the renders as images to a similar or smaller model (like Gemma) to judge generation quality, lowering the compute barrier.
Related event: Hugging Face staff share low-compute tips for model evaluation(2 posts)→
More from Models
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27