Run open LLMs locally on Apple Silicon — fast, private, Apache-2.0. A Mac app or a one-line install.
San Francisco
Joined February 2021
- Tweets137
- Following9
- Followers9.2K
- Likes212
Rapid-MLX · local AI on your Mac retweeted
🚀 Trio-Spark DI (9B) just hit the Top 10 on the Decision Index for models ≤9B!
Built for bulletproof agent reliability in the real world, not benchmark inflation:
✅ 0 errors on 110k+ requests
✅ 97.04% independent match rate
✅ Zero text generation overhead
Turn action selection into a single, perfectly calibrated forward pass. Live now as a gated 9B GGUF (CC BY-NC 4.0).
📊 Board: huggingface.co/spaces/multim…
📦 Weights: huggingface.co/MachineFi/Tri…
Trio-Spark v1.3 is live ⚡
The SOTA System-1 decision model for autonomous agents.
Most agent pipelines burn compute generating text just to pick an action. Spark turns action selection into a single forward pass: zero generated tokens, perfectly calibrated confidence.
🏆 SOTA Performance Across 13 Benchmarks:
• 90.9% on the ten-task panel (+7.8% over v1.0, overtaking Jev 1.13 and level with Clef)
• #1 among all 7 models on PubMedQA, RTE, TweetEval, and SNLI
• Calibration error crushed to 0.019 (vs 0.234 in v1.0) — true confidence you can actually threshold
Upgrade is drop-in: same API, identical limits, text & multimodal vision support. Just point to "model": "trio-spark-v1.3".
⚡ Get 100 free decisions: platform.machinefi.com/spark
📖 Full benchmark report & transparency note:
machinefi.com/blog/trio-spar…
Rapid-MLX · local AI on your Mac retweeted
TensorFold popping up everywhere 🤯
🚀 Rapid-MLX v0.16.0 is here! Turn your Mac into an absolute local AI powerhouse.
Introducing TensorFold profiles with massive speedups: 🔥 DeepSeek V4 Flash: up to 2.1× throughput
🧠 Bonsai 2 27B: +42%
✨ Gemma 4 26B: +4%
Upgrade and get more tokens: github.com/raullenchai/Rapid…
ALT Rapid-MLX v0.16.0 throughput benchmarks showing up to 2.1x speedup on DeepSeek V4 Flash
Rapid-MLX · local AI on your Mac retweeted
Bring Your Own Model is cool, e.g.,
rapid-mlx serve TheCluster/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-MLX-mxfp4 \
--host 127.0.0.1 \
--port 18766 \
--max-num-seqs 1 \
--context-length 32768 \
--no-thinking
Bring Your Own Model (BYOM) is now seamless. 🧩
Want to run an external community checkpoint? Just point the server directly at a Hugging Face repo:
`rapid-mlx serve prism-ml/Ternary-Bonsai-2-27B-mlx-2bit --port 18861`
We handle the preflight checks, caching, and diagnostics automatically so you can get straight to a working, OpenAI-compatible API.
Rapid-MLX · local AI on your Mac retweeted
Rapid-MLX releases Qwen3.8-Flash-Next-4bit, an MLX quantization for running Qwen3.8-Flash-Next locally on Apple Silicon.
🤖 modelscope.ai/models/rapid-m…
⚡ Native MTP speeds up generation by 36–42%, reaching 34.85 tok/s in the measured Mac Studio setup.
🧠 Preserves the upstream model’s ~6B active parameters per token and native 262,144-token context.
✅ Passes 45/45 checks covering reasoning, coding, tool calls, structured output, and long-context recall.
💻 Around 105GB to download. 192GB unified memory is recommended.
📜 Experimental and text-only. Vision is not enabled. Qwen Community License 1.0.
Rapid-MLX · local AI on your Mac retweeted
rapid-mlx 0.12.15 is out 🐆
⚡️ We just shipped lossless MTP speculative decoding for Qwen3.8-27B, 1.25x faster! Just one flag: rapid-mlx serve qwen3.8-27b-mixed-3.5bpw --speculative-config '{"method":"mtp"}'
Other major updates:
🛠️ Coding agents fix: If you use Claude Code, Aider, etc., upgrade now. v0.12.5–0.12.14 could silently shorten what your agent wrote to files. Fixed. (Worth a re-check if your agent edited anything important recently).
🚀 Blazing downloads: Grab a model at 80–90 MB/s. A 9B is ready to chat in ~2 mins, with auto-rerouting if it slows down.
🎙️ Better voice: Whisper no longer makes up words when you pause or the room is quiet.
🛡️ Safer testing: Model downloads can't run code on your machine anymore, and every file is integrity-checked.
Full changelog: rapidmlx.com/changelog/0.12.…
