Unverified50% confidenceEventExact time
In the reproduction experiment (torch 2.14.0, macOS 26.5.2, M5 Pro 48 GB, 4 GiB tensor), right after moving to CPU both phys_footprint and the MPS driver count returned to near zero, and only after the CPU read through the copy page by page did the footprint return to 4.2 GiB, with the results of three runs completely identical.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/13/2026
First Seen
Valid until: 1/11/2027
Sources
Related Entities
Related Claims
UnverifiedOn an Apple Silicon Mac, after moving a PyTorch tensor on the MPS device to the CPU via .to("cpu"), the process's phys_footprint immediately drops to near zero, but the tensor data remains fully in memory.72% similarUnverifiedThe Q8 main model occupies 162 GB, preventing the drafter from fitting into unified memory, rendering DSpark acceleration ineffective; after moving three layers of MoE experts back to the CPU, output only improved from 22.87 to 23.49 tokens/second (2.7% faster), while prefill dropped from 95.38 to 76.65.66% similarUnverifiedThe FP4-based model (svdq-fp4_r32) built on Nunchaku has a size of 6.6GB, fits entirely within 8GB of VRAM without CPU offloading, and reduces the per-step time to 3.4–3.8 seconds.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1015596API
curl https://kongchang.com/api/v1/knowledge/claims/1015596MCP
get_claim(id=1015596)