5 days ago
-
RFC: device_map form implemented, measured warm start; transformers.patch at PR 1a1652cb5b
-
RFC rewritten for the current state (maintainer directions, hooks); transformers.patch at PR b7477fbb00
-
Update every page to 2026-11-17: details (results, weights, kernels, streaming, cache policy, known issues), RFC, README, results index, transformers.patch at 18abece3ab
6 days ago
-
Upload folder using huggingface_hub
8 days ago
-
Decode 10.0 tok/s: shared-expert overlap, early MPS commits, fused hyper-connection projections; dense ablation
-
Decode on four different prompts (cold cache), cache-policy replay
-
Qwen3.8: 8.7 tok/s decode (fused Gated DeltaNet decode step), current prefill
-
Qwen3.8 only: results, the two Macs, architecture and kernels
9 days ago
-
Final sweep: page cache bypassed vs on; prefill and copy optimizations; kernel bug root cause
-
Add vanilla quantized baselines (MetalConfig, preloaded quantized experts) and BF16 with Hub kernels
-
MoE expert streaming results: OLMoE on M5 (BF16, 8-bit, 4-bit) and earlier cloud runs