ds4
Local LLM inference engine by antirez for DeepSeek V4 and other large open models
HN commenters like it for running DeepSeek V4 Flash and other large open models locally on 128GB Macs, Strix Halo and DGX Spark machines,1234 and several say it is solid enough for daily coding and agent work.567 The most common caveats are slow generation on some setups8910 and fewer features and supported models than llama.cpp.11
Why people recommend it
Large local models · 3
Lets 128GB machines run big sparse models like DeepSeek V4 Flash locally.
“DeepSeek-V4-Flash will fit, served by DwarfStar”
“running via antirez's ds4”
Show 1 more
“allow optimized local inference through open-source tools such as: https://github.com/antirez/ds4”
Daily reliability · 3
Users report it working well and staying solid in daily use.
“I use the q4 quant daily and it works really well.”
“I've been using ds4 for months now and it's already got support for the new vision model”
Show 1 more
“it absolutely is usable”
Model and features · 3
Supports DeepSeek 4, Qwen3.8 Flash Next, a vision model and runtime steering vectors.
“DeepSeek 4 family + more.”
“I've been using ds4 for months now and it's already got support for the new vision model”
Show 1 more
“Antirez's DS4 already supports this”
Usable speed · 3
Commenters report usable token rates on Spark, Strix Halo and M4 Max hardware.
“I started with antirez' DwarfStar[1] on one spark and that (~11-14tok/s generation, ~300-400 tok/s prompt processing) was enough of a taste for me”
“have been extremely excited about Antirez's (Redis author) work on DS4”
Show 1 more
“running via antirez's ds4”
Focused design · 1
Praised for targeting the main use cases instead of every possible setup.
“this is such a great quote about building software in the AI age”
Subscription alternative · 1
Cited as a way to get capable local models without expensive AI subscriptions.
“Antirez/DS4 is an example”
Caveats and dissent
Generation speed · 3
Generation can feel slow, and some models such as GLM 5.2 run poorly.
“running via antirez's ds4”
“have been extremely excited about Antirez's (Redis author) work on DS4”
Show 1 more
“I’ve been using 27B more and more over DS4.”
Versus llama.cpp · 1
One user found llama.cpp better for its web UI, router, APIs and broader model support.
“I didn't see anything that would make me want to stick with it vs llama.cpp.”
Strix Halo stability · 1
One commenter is waiting for it to stabilize on Strix Halo.
“Waiting for this https://github.com/antirez/ds4 to stabilize for strix halo.”
Hardware cost · 2
Needs costly high-memory hardware, and one user says running locally makes no economic sense.
“also use DS4”
“have been extremely excited about Antirez's (Redis author) work on DS4”
Often compared with
1 moved from ds4 to Qwen 3.8
1 moved from ds4 to llama.cpp
Who recommends it in their own words
From the makers not counted in the ranking
“The ds4 q4 quant performs a lot better than other similar sizes that I've seen.”
“I maintain a fork of ds4 as shared libraries”
“DwarfStar work in progress numbers: I see 14 tokens/sec generation”
Threads where it came up
- From the creator of Redis; run LLM locally with ds4 4 mentions
- Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding? 3 mentions
- Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s 2 mentions
- DeepSeek launching v4.1 flash cheaper and more capable than v4 pro 2 mentions
- DeepSeek-v4.1-Flash-1017 1 mention
- Janas: Run LLMs larger than RAM by streaming experts from SSD 1 mention
Also written as DS4, DwarfStar, dwarfstar, Dwarfstar. Data: ds4.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 26
“Try the Qwen 3.8 flash next model on DwarfStar”
“DeepSeek 4 family + more.”
“I use the q4 quant daily and it works really well.”
“this is such a great quote about building software in the AI age”
“The project GitHub page is a much better introduction for the hn crowd.”
“It is pretty nifty.”
“Antirez's DS4 already supports this”
“it seems that this project is much more than PoC and actually both usable and useful for people doing local inference”
“I didn't see anything that would make me want to stick with it vs llama.cpp.”
“I've been using ds4 for months now and it's already got support for the new vision model”
“I’ve been using 27B more and more over DS4.”
“also use DS4”
“why not running DeepSeek V4 on top of https://github.com/antirez/ds4. I think the results could be really good.”
“I wonder how it’s compared to https://github.com/antirez/ds4”
“The DS4 repo is a really interesting place to watch folks who heavily code with agents collaborate together in a way that seems pretty effective.”
“DeepSeek-V4-Flash will fit, served by DwarfStar”
“have been extremely excited about Antirez's (Redis author) work on DS4”
“I use it as my main coding agent - so its running DS4 server on my 128gb mbp”
“Inspired by [https://github.com/antirez/ds4](ds4)”
“I started with antirez' DwarfStar[1] on one spark and that (~11-14tok/s generation, ~300-400 tok/s prompt processing) was enough of a taste for me”
https://github.com/antirez/ds4
“Antirez/DS4 is an example”
“allow optimized local inference through open-source tools such as: https://github.com/antirez/ds4”
“it absolutely is usable”
“Waiting for this https://github.com/antirez/ds4 to stabilize for strix halo.”
“running via antirez's ds4”