Rare find

DwarfStar. Run DeepSeek AI models on your own computer without sending data to the cloud.

github.com/antirez/ds4

Problem

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Audience

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Market

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Pricing

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Vaya's read on this project

Problem, audience, market, and the verdict — sign in to see it.

Updates

July 2026

Fix batched server session recovery race Jul 23
README: update the AI assisted models list. Jul 20
README: remove title now that there is the logo. Jul 20
Add DwarfStar logo and credits Jul 20
Optimize Metal prefill and decode kernels Jul 20
Make release builds warning-free Jul 20
Stabilize CUDA and Metal session batching Jul 18
Add ROCm GLM 5.2 support Jul 18
Unify distributed inference CLI and documentation Jul 18
Add CUDA tensor parallelism and session batching Jul 18
Add native Metal session batching Jul 18
Add GLM 5.2 inference and quality fixtures Jul 18
Add two-machine Metal tensor parallelism Jul 18
Add DSpark speculative decoding Jul 18
Fix PRO streaming and sampling correctness Jul 18

June 2026

Merge pull request #319 from rinaldofesta/fix/eval-grader-false-negat… Jun 17
Fix server const warning Jun 16
Harden server JSON request parsing Jun 16
Fix ROCm MTP startup residency Jun 16
Use safer ROCm prefill default Jun 16
Guard Metal tensor frees Jun 16
Fix distributed SSD streaming layer slices Jun 15
Fix CPU build after streaming refactor Jun 15
Refactor streaming expert cache API Jun 14
Fix ROCm mixed streaming expert fallback Jun 14
Add CUDA and ROCm SSD streaming Jun 14
Add release QA checklist Jun 14
Polish mixed streaming expert handling Jun 14
streaming: serve mixed-precision routed experts (per-layer boosted qu… Jun 14
Fix ROCm distributed slice mapping Jun 11