Unverified50% confidenceBenchmarkExact time
No-padding training reduces Qwen3-0.6B's VRAM usage from 5.45 GB to 1.87 GB, with a per-step speed improvement of 5 to 8 times.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/6/2026
First Seen
Valid until: 1/4/2027
Sources
Related Entities
Related Claims
UnverifiedOn a single RTX PRO 6000, peak VRAM for Qwen3.8-27B-NVFP4 dropped from 72.9 GB to 40.2 GB, and per-step time was reduced by approximately 11%78% similarUnverifiedSwitching to Q2K quantization reduces a single expert from 3.1MB to 1.9MB, and the full weights from 111GB to 75GB.74% similarVerifiedQwen3-series coding models load at approximately 20 GB and can expand to 35–40 GB during inference.73% similarUnverifiedQwen3 27B使用FP16训练每个参数占2字节，直接换算需要约54GB显存73% similarVerifiedThe Qwen2.5:7B model is approximately 4 GB in size after Q4 quantization.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/981700API
curl https://kongchang.com/api/v1/knowledge/claims/981700MCP
get_claim(id=981700)