MLX-LM M3 Pro benchmarks for Qwen3-4B-Instruct-2507-4bit (mlx-lm PR #1569)

Fields

id

sources/bench-apple-mlx-lm-m3-pro-qwen3-4b

URL

https://github.com/ml-explore/mlx-lm/pull/1569

Publisher

community / ml-explore

Kind

forum

Accessed

2026-10-03

notes

GitHub PR #1569 in ml-explore/mlx-lm adding M3 Pro benchmarks for Qwen3-4B-Instruct-2507-4bit. Hardware: MacBook Pro 14" M3 Pro, 18GB unified memory. Results: Prompt 615.28 tok/s, Generation 44.62 tok/s, Peak memory 3.30 GB. Verified 2026-10-10: Read PR 'Add M3 Pro benchmarks for Qwen3-4B-Instruct-2507' (https://github.com/ml-explore/mlx-lm/pull/1569); confirms exact benchmark figures (prompt: 615.28 tok/s, generation: 44.62 tok/s, peak memory: 3.30 GB on M3 Pro 18GB).

Notes

GitHub PR #1569 in ml-explore/mlx-lm adding M3 Pro benchmarks for Qwen3-4B-Instruct-2507-4bit. Hardware: MacBook Pro 14" M3 Pro, 18GB unified memory. Results: Prompt 615.28 tok/s, Generation 44.62 tok/s, Peak memory 3.30 GB. Verified 2026-10-10: Read PR 'Add M3 Pro benchmarks for Qwen3-4B-Instruct-2507' (https://github.com/ml-explore/mlx-lm/pull/1569); confirms exact benchmark figures (prompt: 615.28 tok/s, generation: 44.62 tok/s, peak memory: 3.30 GB on M3 Pro 18GB).

Cross-references

This record references no other records.