# m5 pro

Published articles for m5 pro.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Running a 20B Video Diffusion Transformer on Consumer Hardware: A Deployment and Measurement Report for MiniMax H3 on Apple Silicon

DevFeed: [Running a 20B Video Diffusion Transformer on Consumer Hardware: A Deployment and Measurement Report for MiniMax H3 on Apple Silicon](<https://devfeed.tech/articles/running-a-20b-video-diffusion-transformer-on-consumer-hardware-a-deployment-and-measurement-report-for-minimax-h3-on-apple-silicon-113781.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.00843>)

Author: Luke Shi, Steven Shi

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [diffusion-transformers](<https://devfeed.tech/topics/diffusion-transformers.md>), [MLX](<https://devfeed.tech/topics/mlx.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [ai evaluation](<https://devfeed.tech/topics/ai-evaluation.md>)

Tags: [17](<https://devfeed.tech/tags/17.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [apple-silicon](<https://devfeed.tech/tags/apple-silicon.md>), [clip](<https://devfeed.tech/tags/clip.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [consistency](<https://devfeed.tech/tags/consistency.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [data](<https://devfeed.tech/tags/data.md>), [dependency](<https://devfeed.tech/tags/dependency.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [generation](<https://devfeed.tech/tags/generation.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [human-review](<https://devfeed.tech/tags/human-review.md>), [innovation](<https://devfeed.tech/tags/innovation.md>), [latency](<https://devfeed.tech/tags/latency.md>), [m5-pro](<https://devfeed.tech/tags/m5-pro.md>), [m5-pro-chip](<https://devfeed.tech/tags/m5-pro-chip.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [minimax](<https://devfeed.tech/tags/minimax.md>), [video](<https://devfeed.tech/tags/video.md>)

### AI overview

The report demonstrates a local text-to-video pipeline for MiniMax H3, a 20.38B-parameter model, on an Apple Mac mini with an M5 Pro and 64 GiB of unified memory. The authors used the official bf16-to-Q8 quantization path on-device, without cloud services, and describe two deployment blockers and their fixes. A 15.08-second 576x1024 clip took 4,838.49 seconds to render, with 27.467 GiB peak MLX memory. In unattended use, the pipeline completed 17 clips over 26.49 hours with no render failures. Blind ratings of those clips averaged 3.12/5 for structural integrity, 3.18 for temporal consistency, and 2.94 for visible artifacts. These are the authors' measurements; the supplied excerpt does not include the deployment steps or results for all four resolution tiers.

### Source excerpt

arXiv:2611.00843v1 Announce Type: cross Abstract: We report the end-to-end deployment and measurement of MiniMax H3, a 20.38-billion-parameter video diffusion transformer (DiT) - the count measured from the safetensors headers of the Q8 checkpoint as deployed locally (Section 3.2) - running entirely on consumer Apple Silicon hardware: an Apple Mac mini with M5 Pro chip and 64 GiB of unified memory. Using the official bf16-to-Q8 (8-bit, group size 64) quantization path executed locally on the machine, we produced a working text-to-video pipeline with no cloud dependency. We document the full deployment procedure, including two non-trivial blockers and their fixes, and report measured figures for build time, peak memory, and per-clip render latency across four resolution tiers (Draft 384x640 through Native 768x1344). A 15.08-second 576x1024 clip rendered in 4838.49 seconds with a measured MLX peak of 27.467 GiB. We further operated the system as an unattended production pipeline, completing 17 clips in 26.49 hours with zero render failures. A blind human review of the 17 production clips (generation prompts withheld) averaged 3.12, 3.18, and 2.94 out of 5 for structural integrity, temporal consistency, and visible artifacts. All figures in this report are our own measurements except where explicitly labeled as third-party data. This is a systems/empirical report; we make no claim of model innovation.

## Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint

DevFeed: [Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint](<https://devfeed.tech/articles/bonsai-2-27b-near-lossless-compression-in-a-9x-smaller-footprint-60219.md>)

Original publisher: [Read original article](<https://simonwillison.net/2026/Sep/17/hn-49747390/>)

Author: Simon Willison

Published: 2026-09-17T22:13:05Z

Content type: opinion

Language: en

Sources: [Simon Willison's Weblog](<https://devfeed.tech/sources/simon-willison-s-weblog.md>)

Topics: [Bonsai](<https://devfeed.tech/topics/bonsai-rx.md>), [llama.cpp](<https://devfeed.tech/topics/llama-cpp.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Compression](<https://devfeed.tech/topics/compression.md>), [ggml](<https://devfeed.tech/topics/ggml.md>), [macOS](<https://devfeed.tech/topics/macos.md>), [API](<https://devfeed.tech/topics/api.md>), [Server](<https://devfeed.tech/topics/server.md>)

Tags: [api](<https://devfeed.tech/tags/api.md>), [bonsai-2-27b](<https://devfeed.tech/tags/bonsai-2-27b.md>), [compression](<https://devfeed.tech/tags/compression.md>), [ggml](<https://devfeed.tech/tags/ggml.md>), [llama](<https://devfeed.tech/tags/llama.md>), [llama-cpp](<https://devfeed.tech/tags/llama-cpp.md>), [llama-cpp-30](<https://devfeed.tech/tags/llama-cpp-30.md>), [llm](<https://devfeed.tech/tags/llm.md>), [m5-pro](<https://devfeed.tech/tags/m5-pro.md>), [macos](<https://devfeed.tech/tags/macos.md>), [server](<https://devfeed.tech/tags/server.md>)

### AI overview

A technical comment explains how to run the Bonsai 2 27B GGUF model using Prism's llama.cpp fork and its llama-server web UI or API. The author reports observed local inference speeds on an M5 Pro and notes an unresolved startup warning.

### Source excerpt

My comment on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint -- Hacker News. If you want to try out out the GGUFs from https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf#th... be aware that you need Prism's llama.cpp fork to get them to work, from https://github.com/PrismML-Eng/llama.cpp/releases/tag/prism-... This should work: cd /tmp # Get the Prism macOS runtime curl -fL https://github.com/PrismML-Eng/llama.cpp/releases/download/prism-b10685-7dffb15/llama-prism-b10685-7dffb15-bin-macos-arm64.tar.gz -o bonsai-runtime.tar.gz tar -xzf bonsai-runtime.tar.gz # Get the ~5.95 GB GGUF model: curl -fL https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf/resolve/main/Ternary-Bonsai-2-27B-PTQ1_0.gguf -o Ternary-Bonsai-2-27B-PTQ1_0.gguf # Run the server, I used port 8331 ./llama-prism-b10685-7dffb15/llama-server \ -m Ternary-Bonsai-2-27B-PTQ1_0.gguf \ --port 8331 -ngl 99 -fa on -c 32768 Then open http://localhost:8331 for the (very good) baked in llama-server web UI... or run a prompt via the API like this: uvx llm openai endpoint http://127.0.0.1:8331/v1 \ --model bonsai-2-27b --responses hi That's running at ~20 token/second for me on an M5 Pro (after a server restart I got 44 token/second, not sure why), but I'm pretty sure something isn't working right, on startup the server said "ggml_metal_device_init: - the tensor API is not supported in this environment - disabling". Tags: llama-cpp

## Ollama is now powered by MLX on Apple Silicon in preview

DevFeed: [Ollama is now powered by MLX on Apple Silicon in preview](<https://devfeed.tech/articles/ollama-is-now-powered-by-mlx-on-apple-silicon-in-preview-83388.md>)

Original publisher: [Read original article](<https://ollama.com/blog/mlx>)

Published: 2026-03-30T00:00:00Z

Content type: release

Language: en

Sources: [Ollama](<https://devfeed.tech/sources/ollama.md>)

Topics: [MLX](<https://devfeed.tech/topics/mlx.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>), [AI tools for developers](<https://devfeed.tech/topics/ai-tools-for-developers.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Language models](<https://devfeed.tech/topics/language-models.md>), [GPU](<https://devfeed.tech/topics/gpu.md>)

Tags: [apple](<https://devfeed.tech/tags/apple.md>), [apple-silicon](<https://devfeed.tech/tags/apple-silicon.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [ggml](<https://devfeed.tech/tags/ggml.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llama-cpp](<https://devfeed.tech/tags/llama-cpp.md>), [m5-max](<https://devfeed.tech/tags/m5-max.md>), [m5-pro](<https://devfeed.tech/tags/m5-pro.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mlx](<https://devfeed.tech/tags/mlx.md>), [ollama](<https://devfeed.tech/tags/ollama.md>)

### AI overview

Ollama previews a version powered by Apple's MLX framework for Apple silicon, with reported performance gains and updated cache reuse and checkpointing for coding and agentic tasks. It also adds NVFP4 model support and provides commands to try a Qwen3.5 coding model on a Mac with more than 32 GB of unified memory.

### Source excerpt

Today, we're previewing the fastest way to run Ollama on Apple silicon, powered by MLX, Apple's machine learning framework.