All issues

DSpark: per-accepted-token replay cancels ~100% of the speculative saving (M3 Ultra measurements, net_saved = -3.98 s)

#731

· justinz0920-png opened

on Aug 7, 2026

M3 Ultra 512 Testing results with DSv4F 0731

#726

· trueimage opened

on Aug 6, 2026

DSpark proposals systematically wrong under --ssd-streaming (DeepSeek-V4-Flash-0731, Metal): enabling patch observed, miss_first=100%

#724

· johnpippett opened

on Aug 6, 2026

CUDA loader: Q4-class GGUF load transient is ~1.8× mapped tensor size on unified memory (GB10) — freezes 128 GB DGX Spark

#721

· snhwang opened

on Aug 6, 2026

Show and tell: DSBrain (macOS menu bar for ds4-server)

#719

· derkan opened

on Aug 6, 2026

Laguna server batch prefill fails at >=7k tokens: routed MoE intermediate quantize launch invalid argument

#713

· nexus-cw opened

on Aug 5, 2026

ds4f-q2-q4 cannot be served resident on 128 GB DGX Spark (GB10) — global OOM during startup span preparation

#705

· adamlawi opened

on Aug 5, 2026

Extend live-prefix rewind (#668) beyond GLM: DeepSeek V4 Flash on Metal still cold-prefills exact-prefix retries

#698

· Snail3D opened

on Aug 5, 2026

Segfault (exit 139) when --imatrix-out is combined with --role coordinator

#697

· datanerdie opened

on Aug 5, 2026

DSpark scheduler break-even model appears not to account for replay cost (Metal, net -1327ms despite 70% accept rate)

#695

· datanerdie opened

on Aug 5, 2026

--dist-replay-check is a silent no-op on the CLI coordinator path

#694

· datanerdie opened

on Aug 5, 2026

gguf-tools splicer cannot read the MXFP4 GGUF this repo publishes (unsupported GGML tensor type 39)

#693

· datanerdie opened

on Aug 5, 2026