MPS in PyTorch 2.15
We contribute across PyTorch, and one of our focus areas is the MPS backend, where the goal is to make your Mac a first-class place to run and train PyTorch models.
ProofVendor claim only
Key points
- A quick word about who we are: we're the PyTorch team at Hugging Face.
- So we run real models from the Hugging Face Hub across language, image and video, find where MPS falls short and fix it upstream in PyTorch, more and more often by replacing MPSGraph with dedicated Metal kernels.
- FP8 comes to the Mac: store float8e4m3fn values in half the space of BF16 and use scaled matrix multiplication and all other ops which are supported on CUDA/CPU!
- Much faster local LLMs. Mixture-of-experts models decode 5–6.5× faster than in 2.14: on an M3 Ultra with torch.compile, Qwen3.6-35B-A3B goes from 9.4 to 61.3 tokens/s and LFM2.5-8B-A1B from 23.6 to 123.6 tokens/s.
Sources (1)
- [1]MPS in PyTorch 2.15Hugging Face Blog · Oct 20, 12:00 AM
We contribute across PyTorch, and one of our focus areas is the MPS backend, where the goal is to make your Mac a first-class place to run and train PyTorch models.
A quick word about who we are: we're the PyTorch team at Hugging Face.
Extractive summary: sentences quoted from the sources.
Hype check
Independently evaluated6 independent looks since
- Evaluation · Oct 20, 2026Test-time Adaptation for Point Cloud Denoising under Unseen Noise
- Evaluation · Oct 20, 2026Runtime Control for AI Agents: Closing the Evaluation--Execution Gap in Operational Workflows
- Evaluation · Oct 20, 2026The More the Better? A Comparison of Single- and Multi-Agent LLM Setups on a Financial Review Task
- Evaluation · Oct 20, 2026When Health Assistants Agree Too Readily: A Benchmark of Time-Critical Self-Diagnosis
- Evaluation · Oct 20, 2026Act on Your Behalf, Beyond Your Authorization: Benchmarking LLM Agent Authorization Alignment
- Evaluation · Oct 20, 2026AnomalyJev: Towards Evaluating Image Anomaly Explanations via Claim-Level Verification
Before this
- Oct 19, 2026Amazon SageMaker JumpStart: 2026 year-to-date in review
- Oct 19, 2026We built our own inference engine so a 32B model could keep 262K context inside a 16 GB GPU budget. Here's an uncut run.
- Oct 19, 2026sgl-project/sglang v0.5.22
- Oct 19, 2026TorchNeuron: running HF models on AWS Trainium is easier than ever
- Oct 18, 2026GPT-2 KV cache from scratch: correctness, latency, and memory measurements [P]
- Oct 18, 2026vllm-project/vllm v0.32.0