--- license: apache-2.0 base_model: Qwen/Qwen2.5-0.5B-Instruct base_model_relation: quantized pipeline_tag: text-generation library_name: transformers.js language: - multilingual tags: - transformers.js - onnx - qwen2 - text-generation - conversational --- # Qwen2.5-0.5B-Instruct — ONNX / Transformers.js [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) with ONNX weights structured for use with [Transformers.js](https://huggingface.co/docs/transformers.js/index). 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](https://huggingface.co/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](https://huggingface.co/tonythethompson/Qwen2.5-0.5B-Instruct/tree/main) 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](https://huggingface.co/docs/transformers.js/index) (`@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](https://choosealicense.com/licenses/apache-2.0/) — same as [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct). This packaging repo adds no new license terms.
