---
title: "Redis Creator antirez Runs DeepSeek V4 Flash at 35 Tokens per Second"
subtitle: "antirez uploads custom GGUF quantizations of DeepSeek V4 Flash to Hugging Face, purpose-built to run the 284B MoE model on MacBooks with 128 GB of RAM"
url: "https://alphasignal.ai/news/redis-creator-antirez-runs-deepseek-v4-flash-at-35-tokens-per-second"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-04-26T07:54:05.000Z"
updated: "2026-06-08T20:02:15.203Z"
type: "Model"
topics:
  - "Llms"
  - "Gpus"
subtopics:
  - "Mixture Of Experts"
  - "Small Models"
  - "Quantization"
upvotes: 1230
source_url: "https://huggingface.co/antirez/deepseek-v4-gguf"
comments: 0
truncated: true
---

# Redis Creator antirez Runs DeepSeek V4 Flash at 35 Tokens per Second

> antirez uploads custom GGUF quantizations of DeepSeek V4 Flash to Hugging Face, purpose-built to run the 284B MoE model on MacBooks with 128 GB of RAM

Salvatore Sanfilippo, better known as **antirez** and the creator of Redis, has uploaded a set of hand-crafted GGUF quantizations of DeepSeek V4 Flash to Hugging Face under [antirez/deepseek-v4-gguf](https://huggingface.co/antirez/deepseek-v4-gguf). These are not generic quants produced by an automated pipeline. They are bespoke weight files designed from the ground up to work with his companion inference engine, [DwarfStar 4 (ds4)](https://github.com/antirez/ds4), and to squeeze a 284-billion-parameter Mixture-of-Experts model onto Apple Silicon machines starting at 128 GB of RAM.

## What is actually in the repo

The quants are specific for the DS4 inference engine. They may work with other inference engines, but the MTP model requires a specific loader. The recommendation is to use the Q2 variant on 128 GB Mac machines and Q4 on machines with 256 GB or more RAM, optionally pairing either with an MTP model for speculative decoding.

Two main quantization levels are available:

* **Q2 (\~87 GB):** The Q2 quantized model takes about 76 GB of RAM. Targets MacBook Pro and MacBook Pro M3 Max class machines with 96 or 128 GB.
* **Q4 (\~165 GB):** The Q4 variant requires machines with at least 256 GB and runs comfortably on Mac Studio M3 Ultra with 512 GB.
* **MTP (optional):** A draft model for speculative decoding, providing an experimental speed boost for greedy decoding.

## The quantization design: asymmetric by intent

The key insight behind these GGUFs is not just how aggressively they compress the model, but

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [X post by @alphasignal](https://x.com/0xSero/status/2055773970138362321)
- [antirez/deepseek-v4-gguf · Hugging Face](https://huggingface.co/antirez/deepseek-v4-gguf)
- [GitHub - antirez/ds4: DeepSeek 4 Flash local inference engine for Metal](https://github.com/antirez/ds4)
- [DwarfStar4 (DS4) Roadmap by antirez: DeepSeek V4 Flash on Apple Silicon and CUDA](https://pasqualepillitteri.it/en/news/2253/ds4-antirez-deepseek-v4-flash-inference-engine)
- [ds4 Review: antirez's Pure-C DeepSeek V4 Flash Engine](https://andrew.ooo/posts/ds4-antirez-deepseek-v4-flash-local-inference-review/)
- [DeepSeek V4 Flash Quantized Version Goes Viral](https://pythonlibraries.substack.com/p/deepseek-v4-flash-quantized-version)