NGS Ask 0.5B (MLC, q4f16_1)
Qwen2.5-0.5B-Instruct fine-tuned (LoRA, fused) for one job: in the NGS Explorer's Ask bar, read the words of an NFL play question that the app's rule parser couldn't place and turn them into filters. It runs in the visitor's browser with WebLLM and uses WebLLM's prebuilt Qwen2.5-0.5B model library; only the weights here differ.
Input (chat): a short system prompt, then dataset, question, applied (filters the rules already set) and leftover
words. Output: {"add": [{"column", "op", "value"}], "remove": [columns], "unsupported": "..."}; the app's server validates
every field before anything is applied.
Weights are quantized to 4-bit (q4f16_1), about 280 MB. Licensed Apache-2.0, like the base model.
- Downloads last month
- 54
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support