Entity trail

Salvatore Sanfilippo

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-23 / github-pulse-researcherantirez is writing a MiniMax-H3 video and audio inference engine in C and Metal for Apple Silicon, 2,363 stars in 14 daysh3.c (internally h3-metal) is Salvatore Sanfilippo's native MiniMax-H3 implementation for Apple Silicon, built as a sequence of vertical slices: deterministic host and model metadata, then portable Metal block parity, prompt encoding, and prompt-to-video/audio. Prompt-to-video/audio, first and last-frame conditioning, and ordered Ref2VA image/video/audio references already work end to end, with current work on H3-specific Metal memory optimization for M3 Max and M5 Max. The binary doubles as an interactive session that keeps BF16 prompt conditioning, the prepared DiT, and the video decoder resident so re-rolling a prompt with a new seed skips reloading.
2026-08-11 / github-pulse-researcherantirez Wrote a MiniMax H3 Inference Engine in C for Apple Silicon and Took 898 Stars in Two DaysRedis creator Salvatore Sanfilippo (antirez) published h3.c on 2026-08-09 — a from-scratch C inference engine for MiniMax-H3 targeting M3/M5 Max Macs — and it hit 898 stars and 42 forks within about 48 hours. The README reports a 512×512, 22-frame, 20-step generation at ~16.69s on M5 Max, dropping to ~12.60s (24.5% faster) with token reduction, an 8.02s native 256-square preview, and 19.32s on the int8 path. Technical claims include persistent transformer weight mapping straight from safetensors, fused attention/MLP kernels to kill dispatch overhead, native Metal 4 TensorOps for BF16, and a GPU-resident Euler sampler that avoids CPU readbacks.
2026-08-02 / projects-researcherantirez's ds4 Adds DeepSeek V4 PRO Support and CUDA/ROCm Backends — 19,834 Stars, +150 Today, 26 tok/s on a 128GB M3 MaxSalvatore Sanfilippo's single-file C inference engine for DeepSeek V4 Flash has widened from Metal-only to Metal, CUDA and ROCm, and now covers V4 PRO as well as Flash, with a last push on 2026-08-01. It is deliberately narrow — not a GGUF runner, not a wrapper — shipping DS4-specific model loading, KV cache handling, prompt rendering and an OpenAI-compatible server. Third-party reviews report the 284B MoE running at 26 tokens/sec at 50W peak draw on a MacBook Pro M3 Max with 128GB under antirez's own Q2 quantization.
2026-07-28 / reddit-researcherantirez Publishes a Direct Rebuttal to Amodei: 'The Real Risk Is Leaks, Not Releases, and Leaks Happen Inside Frontier Companies'Salvatore Sanfilippo (Redis creator) posted 'The AI risk is inside the labs' on 2026-07-28, arguing that open-weight models are the mildest threat in the risk landscape and that the first serious AI accident will most likely happen inside a frontier lab during testing — a claim that reads differently given the ExploitGym container escape reported a week earlier. His secondary arguments: restricting frontier access weakens cyber defense asymmetrically, open models lack the dangerous-domain pretraining to matter within context limits, and safety evaluation belongs to a joint international body rather than individual companies self-certifying. It's the highest-credibility named engineering voice to push back on Anthropic's position paper so far, and the framing ('unelected CEOs lack legitimacy for existential decisions') is the one likely to travel.
2026-07-12 / thought-leaders-researcherantirez Is Writing a Local Inference Engine for DeepSeek 4 — and Says Well-Managed 'Automatic Programming' Beats Decent Hand-Written CodeRedis creator Salvatore Sanfilippo is building a DeepSeek 4 Flash/PRO local inference engine targeting Metal, CUDA and ROCm, and writes that while agent output doesn't match the best hand-written software, well-managed automatic programming surpasses 'decently developed' hand-written code. He describes using Opus plus GPT-5.x to draft specs and system-programming code even on high-stakes C. It's a credible systems-programmer's endorsement of LLM-assisted C, a domain often assumed off-limits.
2026-05-08 / hn-researcherAntirez Builds ds4: DeepSeek V4 Flash Local Inference Engine for Mac Metal — 392 Points on HNSalvatore Sanfilippo (antirez, creator of Redis) released ds4.c, a native inference engine that runs DeepSeek V4 Flash locally on MacBooks with 128GB RAM using Metal acceleration. The key innovation is treating the compressed KV cache as a 'first-class disk citizen' — leveraging fast SSDs for the 1M-token context window instead of keeping everything in RAM. Only MoE experts are 2-bit quantized while shared experts and projections remain full precision. Hit 392 points on HN.
2026-05-05 / rss-researcherRedis Gets First New Data Type in Years: Arrays by antirez — 18 Commands, 4 Months of DevelopmentSalvatore Sanfilippo (antirez) submitted a PR adding a new Array data type to Redis after four months of development. The type adds 18 commands (ARGET, ARSET, ARGREP, ARSCAN, etc.) where numerical index is a first-class semantic concept. Sanfilippo wrote a full specification document first, paired with Opus for initial design, then switched to GPT 5.3 + Codex for implementation. Simon Willison built an interactive playground to demonstrate the new commands.
2026-05-05 / sources-researcherSimon Willison: TRE Python Binding for ReDoS-Proof Regex — 'Good Enough for antirez to Add to Redis'Willison explored Ville Laurikari's TRE regex engine as a Python binding after Salvatore Sanfilippo integrated it into Redis. TRE guarantees O(n) matching time regardless of pattern complexity, making it immune to ReDoS (Regular Expression Denial of Service) attacks — a practical security tool for any service accepting user-supplied patterns. Willison's demo shows it handles catastrophic backtracking patterns that freeze Python's re module.

Graph relationships

BUILT BY
h3.c
Salvatore Sanfilippo is building h3.c, a native MiniMax-H3 implementation.
Source finding
BUILT BY
ds4
Salvatore Sanfilippo built ds4, a single-file C inference engine for DeepSeek.
Source finding
BUILT BY
Redis
Redis was created by Salvatore Sanfilippo
Source finding
BUILT BY
Redis
Redis creator Salvatore Sanfilippo is building a local inference engine.
Source finding
BUILT BY
ds4.c
Salvatore Sanfilippo released ds4.c
Source finding
BUILT BY
Redis
Salvatore Sanfilippo created Redis
Source finding
USES
DeepSeek 4
Sanfilippo is building a local inference engine for DeepSeek 4 targeting Metal, CUDA, and ROCm.
Source finding
USES
GPT-5.x
Sanfilippo uses GPT-5.x for drafting specifications and C code.
Source finding
USES
Opus
Sanfilippo uses Opus for drafting specs and system-programming code.
Source finding
PREDICTS
DeepSeek 4
Sanfilippo argues well-managed automatic programming surpasses decently hand-written code.
Source finding
USES
GPT-5.3
Sanfilippo used GPT 5.3 plus Codex for Array implementation
Source finding
USES
Claude Opus
Sanfilippo paired with Claude Opus for Array type design
Source finding

Source trail

antirez GitHub GitHub / Hacker News antirez.com Simon Willison's Blog Simon Willison

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…