Search results

A star chart for ds4, if it's of interest

Status: Open.

#655 In antirez/ds4;

· me93-ghb opened on Aug 2, 2026

download_model.sh has no way to fetch the dated GGUF builds

Status: Open.

#654 In antirez/ds4;

· STRML opened on Aug 2, 2026

DeepSeek-V4-Flash-0731: the -imatrix-0731 quants ARE 0731-calibrated, byte-exact reasoning-effort prefixes, and a max_tokens warning

Status: Open.

#653 In antirez/ds4;

· rell666 opened on Aug 2, 2026

ds4-server: expose tensor-parallel roles (field report: 2x M4 Max TB5, TP 22.7 t/s vs server-pipeline ~20)

Status: Open.

#651 In antirez/ds4;

· OmarB97 opened on Aug 2, 2026

Model reaches correct answer internally but won't commit — repeated web_search loops on small reasoning puzzles

Status: Open.

#646 In antirez/ds4;

· altdotboy opened on Aug 1, 2026

[Discussion] Tile indexed-attention scoring independently of the global prefill chunk

Status: Open.

#644 In antirez/ds4;

· apetersson opened on Aug 1, 2026

[Discussion] Reduce 1M Metal peak via comp_mask sizing and phase-separated DSpark residency

Status: Open.

#643 In antirez/ds4;

· apetersson opened on Aug 1, 2026

DSpark preserved-MXFP4 planner support and mixed IQ2/MXFP4 Metal tests

Status: Open.

#642 In antirez/ds4;

· apetersson opened on Aug 1, 2026

ds4 cannot load MXFP4Experts-0731 GGUF: unsupported GGUF type 39 (MXFP4) for routed experts

Status: Open.

#641 In antirez/ds4;

· aaa2015 opened on Aug 1, 2026

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

Status: Open.

#639 In antirez/ds4;

· nexus-cw opened on Aug 1, 2026

Metal wires entire mmap-backed model views on first GPU use (ssd-streaming, mixed-size expert layers)

Status: Open.

#638 In antirez/ds4;

· FabioMalpezzi opened on Aug 1, 2026

SSD streaming: do these measurements from a separate MoE-offload prototype hold on V4 Flash? (eviction key, L+2 router lookahead, MTP union cost)

Status: Open.

#636 In antirez/ds4;

· gabrielekarra opened on Jul 31, 2026