Expired50% confidenceFactExact time
The <2GB RAM and <20s performance mentioned in the title reflects performance after Q4 or Q8 quantization, not FP16 full-precision operation.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/11/2026(expired)
Sources
Apple Silicon本地图生3D：20秒生成、2GB内存，Modelr开源应用实测
Related Claims
VerifiedQ4_K_M uses mixed precision, storing some critical weights at 6-bit and the remainder at 4-bit, with an overall average of approximately 4.5-bit.69% similarUnverifiedGiven the same VRAM/storage footprint, a larger model running Q4 will generally outperform a smaller model running Q867% similarUnverifiedA 70B parameter model quantized with Q4_K_M occupies approximately 40GB of storage, and can achieve an inference speed of 10–20 tokens per second on an RTX 4090 with CPU memory offloading.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502822API
curl https://kongchang.com/api/v1/knowledge/claims/502822MCP
get_claim(id=502822)