Search results

[feature] Offload KV cache from GPU (VRAM) to RAM

Status: Open.

#108 In antirez/ds4;

· d3y4n opened on May 12, 2026

ds4 parser vulnerability: user prompt can insert special tokens that are interpreted by the model

Status: Open.

#95 In antirez/ds4;

· pintorem opened on May 12, 2026

Please provide a ModelScope mirror for Chinese users who cannot easily download large models from Hugging Face

Status: Open.

#85 In antirez/ds4;

· cangming009 opened on May 12, 2026

The "for metal" in "about" no longer reflects the reality

Status: Open.

#84 In antirez/ds4;

· kishida opened on May 12, 2026

Feature request: Natural Language Level Speculative Decoding with heterogeneous Draft Models (e.g. Qwen 2B)

help wantedExtra attention is neededExtra attention is needed

speed

wontfixThis will not be worked onThis will not be worked on

Status: Open.

#80 In antirez/ds4;

· vagrillo opened on May 11, 2026

FP4 inference for 4 bit

correctness

Status: Open.

#74 In antirez/ds4;

· antirez opened on May 11, 2026

can you add deepseek q8 ?

Status: Open.

#64 In antirez/ds4;

· jonipupet opened on May 10, 2026

Speculative Decoding via "Draft KV Stitching"

Status: Open.

#63 In antirez/ds4;

· Nitya-003 opened on May 10, 2026

Problems when launching from other programs like llama-swap

Status: Open.

#49 In antirez/ds4;

· links486 opened on May 10, 2026

Context Window Exhaustion Causes Unterminated Tool Calls

Status: Open.

#48 In antirez/ds4;

· yqxu opened on May 10, 2026

Generating GGUF from official DeepSeek releases

Status: Open.

#25 In antirez/ds4;

· MirkoCovizzi opened on May 9, 2026

Question: any interest in OpenAI Responses API support for Codex compatibility?

Status: Open.

#22 In antirez/ds4;

· alessandrobologna opened on May 9, 2026