Repository Issue Activity (beta)

#736 OpenVino support

#735 ds4-server failed to run with 4xA100/80G (prefill failed)

#734 CUDA --ssd-streaming: prefill dies with illegal memory access at exactly 128 prompt tokens

#732 CUDA --ssd-streaming: MMQ prefill tier OOMs on a 12 GB card (main b030961)

#733 DSpark net-negative on M3 Ultra + MXFP4 despite 82.8% accept rate — independent confirmation of #695 replay/verify overhead

#629 CUDA SSD streaming: host-RAM tier for routed experts (working branch, 1.65-3.1x decode)

#521 gfx1201 (RDNA4, R9700) build fails: __builtin_amdgcn_wmma_f32_16x16x16_f16_w32 not supported

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

#639 --ssd-streaming-cache-experts is silently inert on CUDA — expert cache never populated, hit rate 0

#726 M3 Ultra 512 Testing results with DSv4F 0731

#658 DSpark speculative decode breaks --temp 0 greedy identity (batch-verify vs single-token KV numerics)

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

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

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

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

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

#496 GLM 5.2 error: `ds4: required tensor is missing: token_embd.weight`

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

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

#685 ds4-server reasoning_content deltas fragmented mid-token at arbitrary byte offsets (OpenAI-compatible streaming)

#660 New 0731 iq2 thinks forever!

#444 Disk KV eviction preferentially removes short-prefix cache files causing disk cache miss & full pre-fill when Disk KV Store is full

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

#691 KV cache reuse breaks for OpenAI tools clients that don't replay reasoning_content

#585 CUDA OOM regression on DGX Spark (GB10)