lfm25vl-opencal

LFM2.5-VL 450M - OpenCal fine-tune

On-device vision-language model for OpenCal's meal-logging PWA. Fine-tuned from the public onnx-community/LFM2.5-VL-450M-ONNX base on meal-photo + meal-text extraction; exported to ONNX (fp16 + q4 variants) for transformers.js / WebGPU on mobile + desktop.

Layout

• config.json, generation_config.json, tokenizer.json, tokenizer_config.json, preprocessor_config.json, processor_config.json, chat_template.jinja
• onnx/ - embed_tokens, vision_encoder, decoder_model_merged (each in _fp16 and _q4 variants, with external _data weight files)

Load (transformers.js)

import { AutoProcessor, AutoModelForImageTextToText } from "@huggingface/transformers";
const processor = await AutoProcessor.from_pretrained("opencal/lfm25vl-opencal");
const model = await AutoModelForImageTextToText.from_pretrained("opencal/lfm25vl-opencal", { device: "webgpu" });