tonythethompson/Qwen2.5-0.5B-Instruct-ONNX

Qwen2.5-0.5B-Instruct — ONNX / Transformers.js

Qwen/Qwen2.5-0.5B-Instruct with ONNX weights structured for use with Transformers.js. It is not a newly trained model.

ONNX weights are in the onnx/ subfolder, following the standard Transformers.js layout. Intended for in-browser or cross-platform text processing where a small, fast instruction model is needed without a Python dependency.

Source

Field	Value
Upstream model	Qwen/Qwen2.5-0.5B-Instruct
Upstream source revision	7ae557604adf67be50417f59c2c2f167def9a775
Export tool/script	Hugging Face Optimum ONNX export (Transformers.js layout)
Quantization recipe	Optimum default FP32 plus optional onnx/model_q4.onnx / model_q8.onnx variants if present

Files

ONNX weights are stored in the onnx/ subfolder per Transformers.js convention. Full file inventory not documented here; inspect the Files tab for the current file list.

Intended Use

A lightweight (~500M parameter) instruction model for text-processing tasks where minimal latency and a small memory footprint are priorities. Suitable for in-browser inference via Transformers.js or CPU-only environments.

Runtime Notes

• Runtime: Transformers.js (@huggingface/transformers).
• ONNX weights also usable with ONNX Runtime directly from the onnx/ subfolder.

Precision and Packaging

Export tooling, precision, and quantization are recorded in the Source table above. This packaging mirror does not publish independent parity benchmarks; validate on your target execution provider before production use.

Limitations

• 0.5B parameters; instruction following and text quality are substantially weaker than larger models.
• No repository-specific quality evaluation is documented here.
• Multilingual capability varies by language; Qwen2.5 covers primarily Chinese and English well with moderate support for others.

License

Apache 2.0 — same as Qwen/Qwen2.5-0.5B-Instruct. This packaging repo adds no new license terms.