This directory contains example scripts demonstrating MLX-Node capabilities.

lm.ts runs token generation through mlx-node and reports tokens/second.

# Build the project (from project root)
yarn install && yarn build

Convert a Qwen model to MLX float32 format:

# Using mlx-lm
python -m mlx_lm.convert \
    --hf-path Qwen/Qwen2.5-0.5B-Instruct \
    --mlx-path .cache/models/qwen3-0.6b-mlx-f32 \
    --dtype float32

node examples/lm.ts
# Or with oxnode
npx oxnode examples/lm.ts

lm.ts:

• Uses the model at .cache/models/qwen3-0.6b-mlx-f32
• Generates with temperature=0.7, topP=0.9
• Displays tokens/second for performance

Expected output format:

Generated (42 tokens, 850ms, 49.41 tokens/s):
[generated text...]

• First run may be slower due to model loading and Metal shader compilation
• Subsequent runs typically show consistent performance
• tokens/s metric excludes model loading time, only measures generation
• Uses Apple Metal GPU acceleration