[ { "id": "smollm2-135m-instruct", "name": "SmolLM2-135M-Instruct", "hf_repo": "HuggingFaceTB/SmolLM2-135M-Instruct", "task": "text-generation", "params_m": 134.5, "headline": { "webgpu": { "variant": "q4f16", "bytes": 117691126 }, "wasm": { "variant": "uint8", "bytes": 137147981 } }, "variants": { "bnb4": 175434623, "fp16": 270267067, "fp32": 540345794, "q4": 182068553, "q4f16": 117691126, "q8": 274295848, "uint8": 137147981 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct" } ] }, { "id": "smollm2-360m-instruct", "name": "SmolLM2-360M-Instruct", "hf_repo": "HuggingFaceTB/SmolLM2-360M-Instruct", "task": "text-generation", "params_m": 361.8, "headline": { "webgpu": { "variant": "q4f16", "bytes": 272737275 }, "wasm": { "variant": "uint8", "bytes": 364564671 } }, "variants": { "bnb4": 368284174, "fp16": 724891911, "fp32": 1449582810, "q4": 387943246, "q4f16": 272737275, "q8": 729129229, "uint8": 364564671 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct" } ] }, { "id": "gemma-3-270m-it", "name": "Gemma-3-270M-it", "hf_repo": "onnx-community/gemma-3-270m-it-ONNX", "task": "text-generation", "params_m": 268.1, "headline": { "webgpu": { "variant": "q4f16", "bytes": 272956893 }, "wasm": { "variant": "q8", "bytes": 545274884 } }, "variants": { "fp16": 570135451, "fp32": 1139686504, "q4": 323175770, "q4f16": 272956893, "q8": 545274884 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gemma-3-270m-it-ONNX" }, { "label": "Parameter count source", "url": "https://huggingface.co/google/gemma-3-270m-it" } ] }, { "id": "qwen2.5-0.5b-instruct", "name": "Qwen2.5-0.5B-Instruct", "hf_repo": "onnx-community/Qwen2.5-0.5B-Instruct", "task": "text-generation", "params_m": 494, "headline": { "webgpu": { "variant": "q4f16", "bytes": 483003582 }, "wasm": { "variant": "uint8", "bytes": 512096636 } }, "variants": { "bnb4": 763794004, "fp16": 997354499, "fp32": 1993796793, "q4": 786156820, "q4f16": 483003582, "q8": 1024193114, "uint8": 512096636 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct" }, { "label": "Parameter count source", "url": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct" } ] }, { "id": "llama-3.2-1b-instruct", "name": "Llama-3.2-1B-Instruct", "hf_repo": "onnx-community/Llama-3.2-1B-Instruct", "task": "text-generation", "params_m": 1235.8, "headline": { "webgpu": { "variant": "q4f16", "bytes": 1089755422 }, "wasm": { "variant": "uint8", "bytes": 1236654010 } }, "variants": { "bnb4": 1598832302, "fp16": 2090026952, "fp32": 2082653372, "q4": 1692821112, "q4f16": 1089755422, "q8": 2473307902, "uint8": 1236654010 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Llama-3.2-1B-Instruct" }, { "label": "Parameter count source", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct" } ], "notes": "Every build exceeds 1 GB; q4f16 (about 1.04 GB) is the lightest WebGPU option, and even that is a meaningful download and load-time hit for a browser page." }, { "id": "qwen3-0.6b", "name": "Qwen3-0.6B", "hf_repo": "onnx-community/Qwen3-0.6B-ONNX", "task": "text-generation", "params_m": 751.63, "headline": { "webgpu": { "variant": "q4f16", "bytes": 569789750 }, "wasm": { "variant": "uint8", "bytes": 617687575 } }, "variants": { "bnb4": 891573033, "fp16": 1202829402, "q4": 919096585, "q4f16": 569789750, "q8": 1235375053, "uint8": 617687575 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Qwen3-0.6B-ONNX" } ], "notes": "This repo's fp32 bucket bundles extra onnxruntime-specific sub-builds and is excluded from the sizes shown here; the other variants are clean single-file measurements." }, { "id": "qwen2.5-1.5b-instruct", "name": "Qwen2.5-1.5B-Instruct", "hf_repo": "onnx-community/Qwen2.5-1.5B-Instruct", "task": "text-generation", "params_m": 1543.71, "headline": { "webgpu": { "variant": "q4f16", "bytes": 1221878940 }, "wasm": { "variant": "uint8", "bytes": 1578954395 } }, "variants": { "bnb4": 1705680982, "fp16": 3105275452, "fp32": 6209469491, "q4": 1787566590, "q4f16": 1221878940, "q8": 3157908586, "uint8": 1578954395 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Qwen2.5-1.5B-Instruct" } ], "notes": "This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "deepseek-r1-distill-qwen-1.5b", "name": "DeepSeek-R1-Distill-Qwen-1.5B", "hf_repo": "onnx-community/DeepSeek-R1-Distill-Qwen-1.5B-ONNX", "task": "text-generation", "params_m": 1777.09, "headline": { "webgpu": { "variant": "q4f16", "bytes": 1369397579 }, "wasm": { "variant": "uint8", "bytes": 1845367473 } }, "variants": { "bnb4": 1869990808, "fp16": 3605547692, "fp32": 7175991958, "q4": 1966462264, "q4f16": 1369397579, "q8": 3690734849, "uint8": 1845367473 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/DeepSeek-R1-Distill-Qwen-1.5B-ONNX" } ], "notes": "This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "gemma-3-1b-it", "name": "Gemma-3-1B-it", "hf_repo": "onnx-community/gemma-3-1b-it-ONNX", "task": "text-generation", "params_m": 999.89, "headline": { "webgpu": { "variant": "q4f16", "bytes": 763529245 }, "wasm": { "variant": "uint8", "bytes": 1001482078 } }, "variants": { "bnb4": 1602007661, "fp16": 2033972730, "fp32": 2070575091, "q4": 859454179, "q4f16": 763529245, "q8": 2527433531, "uint8": 1001482078 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gemma-3-1b-it-ONNX" } ] }, { "id": "lfm2.5-350m", "name": "LFM2.5-350M", "hf_repo": "onnx-community/LFM2.5-350M-ONNX", "task": "text-generation", "params_m": 354.48, "headline": { "webgpu": { "variant": "q4f16", "bytes": 255148587 }, "wasm": { "variant": "q8", "bytes": 510117461 } }, "variants": { "fp16": 725490555, "fp32": 1450841916, "q4": 293813394, "q4f16": 255148587, "q8": 510117461 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/LFM2.5-350M-ONNX" } ] }, { "id": "granite-4.0-350m", "name": "Granite-4.0-350M", "hf_repo": "onnx-community/granite-4.0-350m-ONNX-web", "task": "text-generation", "params_m": 352.38, "headline": { "webgpu": { "variant": "q4f16", "bytes": 350484597 }, "wasm": { "variant": "q4f16", "bytes": 350484597 } }, "variants": { "fp16": 709157057, "fp32": 1418109704, "q4": 575912398, "q4f16": 350484597 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/granite-4.0-350m-ONNX-web" } ], "notes": "No int8 (q8/uint8) build is published for this repo; the lightest option is q4f16." }, { "id": "bonsai-1.7b", "name": "Bonsai-1.7B", "hf_repo": "onnx-community/Bonsai-1.7B-ONNX", "task": "text-generation", "params_m": 1720.03, "headline": { "webgpu": { "variant": "q4", "bytes": 1119428094 }, "wasm": { "variant": "q8", "bytes": 2006258923 } }, "variants": { "q4": 1119428094, "q8": 2006258923 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Bonsai-1.7B-ONNX" } ], "notes": "Only q4 and q8 are usable measurements here; the repo's fp32 bucket bundles unrelated experimental 1-bit/2-bit files and is excluded. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "phi-3.5-mini", "name": "Phi-3.5-mini-instruct", "hf_repo": "onnx-community/Phi-3.5-mini-instruct-onnx-web", "task": "text-generation", "params_m": 3821.08, "headline": { "webgpu": { "variant": "q4f16", "bytes": 2317473274 }, "wasm": { "variant": "q4f16", "bytes": 2317473274 } }, "variants": { "q4f16": 2317473274 }, "pipeline_task": "text-generation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Phi-3.5-mini-instruct-onnx-web" } ], "notes": "Only a single q4f16 web build (about 2.3 GB) is published, no fp16 or int8 alternative, which alone makes it a heavy download for a browser page. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "gemma-4-e2b-it", "name": "Gemma-4-E2B-it", "hf_repo": "onnx-community/gemma-4-E2B-it-ONNX", "task": "multimodal", "params_m": 5123.18, "headline": { "webgpu": { "variant": "q4f16", "bytes": 3381966758 }, "wasm": { "variant": "q8", "bytes": 3101630901 } }, "variants": { "fp16": 3827754603, "fp32": 5530277394, "q4": 3930062458, "q4f16": 3381966758, "q8": 3101630901 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gemma-4-E2B-it-ONNX" } ], "notes": "The E2B name is an effective/active-param label (like Gemma 3n), not the real stored parameter count: this checkpoint stores about 5.12B params, not 2B. Any-to-any multimodal generation (text, image and audio in and out) has no stable transformers.js pipeline() task yet; expect to wire the encoders and decoder manually. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "gemma-4-e4b-it", "name": "Gemma-4-E4B-it", "hf_repo": "onnx-community/gemma-4-E4B-it-ONNX", "task": "multimodal", "params_m": 7996.16, "headline": { "webgpu": { "variant": "q4f16", "bytes": 4366542985 }, "wasm": { "variant": "q8", "bytes": 3418704565 } }, "variants": { "fp16": 4363908079, "fp32": 6543747885, "q4": 4240529338, "q4f16": 4366542985, "q8": 3418704565 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gemma-4-E4B-it-ONNX" } ], "notes": "The E4B name is an effective/active-param label (like Gemma 3n), not the real stored parameter count: this checkpoint stores about 8.0B params, not 4B. Any-to-any multimodal generation (text, image and audio in and out) has no stable transformers.js pipeline() task yet; expect to wire the encoders and decoder manually. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "janus-pro-1b", "name": "Janus-Pro-1B", "hf_repo": "onnx-community/Janus-Pro-1B-ONNX", "task": "multimodal", "params_m": 2076.64, "headline": { "webgpu": { "variant": "q4f16", "bytes": 1941567247 }, "wasm": { "variant": "uint8", "bytes": 2269873869 } }, "variants": { "bnb4": 2873668114, "fp16": 4519051913, "fp32": 9037350507, "q4": 2985705893, "q4f16": 1941567247, "q8": 4539747577, "uint8": 2269873869 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Janus-Pro-1B-ONNX" } ], "notes": "Janus-Pro does both image understanding and image generation from one checkpoint; there is no stable transformers.js pipeline() task for that yet, so expect custom wiring. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "smolvlm-256m-instruct", "name": "SmolVLM-256M-Instruct", "hf_repo": "HuggingFaceTB/SmolVLM-256M-Instruct", "task": "vision-language", "params_m": 256.5, "headline": { "webgpu": { "variant": "q4f16", "bytes": 188843109 }, "wasm": { "variant": "uint8", "bytes": 259855437 } }, "variants": { "bnb4": 249731074, "fp16": 514479822, "fp32": 1028498218, "q4": 263889326, "q4f16": 188843109, "q8": 519710718, "uint8": 259855437 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/HuggingFaceTB/SmolVLM-256M-Instruct" } ], "notes": "No stable transformers.js pipeline() task for image+text chat yet; use AutoModelForVision2Seq and AutoProcessor directly (see the official SmolVLM-WebGPU demo).", "code_snippet": "import { AutoProcessor, AutoModelForVision2Seq, load_image } from \"@huggingface/transformers\";\n\nconst processor = await AutoProcessor.from_pretrained(\"HuggingFaceTB/SmolVLM-256M-Instruct\");\nconst model = await AutoModelForVision2Seq.from_pretrained(\"HuggingFaceTB/SmolVLM-256M-Instruct\", {\n device: \"webgpu\",\n // fp32 on purpose: the q4f16 and fp16 builds generate garbled text over\n // WebGPU today (the official demo ships fp32 for the same reason).\n dtype: \"fp32\", // use \"uint8\" for the WASM build\n});\n\nconst image = await load_image(\"https://your-image-url.jpg\");\nconst messages = [\n { role: \"user\", content: [{ type: \"image\" }, { type: \"text\", text: \"Describe this image.\" }] },\n];\nconst text = processor.apply_chat_template(messages, { add_generation_prompt: true });\nconst inputs = await processor(text, [image]);\n\nconst output = await model.generate({ ...inputs, max_new_tokens: 256 });\nconst result = processor.batch_decode(output, { skip_special_tokens: true });" }, { "id": "florence-2-base-ft", "name": "Florence-2-base-ft", "hf_repo": "onnx-community/Florence-2-base-ft", "task": "vision-language", "params_m": 231.6, "headline": { "webgpu": { "variant": "q4f16", "bytes": 223446634 }, "wasm": { "variant": "uint8", "bytes": 274966162 } }, "variants": { "bnb4": 319312207, "fp16": 544004842, "fp32": 1085912345, "q4": 333249173, "q4f16": 223446634, "q8": 549932264, "uint8": 274966162 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Florence-2-base-ft" }, { "label": "Parameter count source", "url": "https://huggingface.co/microsoft/Florence-2-base-ft" } ], "notes": "Florence-2 uses task-prompt tokens (for example , ) via Florence2ForConditionalGeneration and AutoProcessor, not a pipeline() call.", "code_snippet": "import { Florence2ForConditionalGeneration, AutoProcessor, AutoTokenizer } from \"@huggingface/transformers\";\n\nconst model = await Florence2ForConditionalGeneration.from_pretrained(\"onnx-community/Florence-2-base-ft\", {\n device: \"webgpu\",\n // Per-module dtype map: keep the language-model parts in fp16/fp32 for\n // quality, quantize the (much larger) vision encoder/decoder to q4.\n dtype: {\n embed_tokens: \"fp16\", // or \"fp32\" where WebGPU fp16 (shader-f16) isn't supported\n vision_encoder: \"fp16\", // or \"fp32\" where WebGPU fp16 isn't supported\n encoder_model: \"q4\",\n decoder_model_merged: \"q4\",\n },\n});\nconst processor = await AutoProcessor.from_pretrained(\"onnx-community/Florence-2-base-ft\");\nconst tokenizer = await AutoTokenizer.from_pretrained(\"onnx-community/Florence-2-base-ft\");\n\nconst prompt = \"\"; // task-prompt token, not free text\n// const inputs = await processor(image, prompt);\n// const generated = await model.generate({ ...inputs, max_new_tokens: 100 });\n// const result = processor.batch_decode(generated, { skip_special_tokens: false });" }, { "id": "qwen3.5-0.8b", "name": "Qwen3.5-0.8B", "hf_repo": "onnx-community/Qwen3.5-0.8B-ONNX", "task": "vision-language", "params_m": 873.44, "headline": { "webgpu": { "variant": "q4f16", "bytes": 646838112 }, "wasm": { "variant": "q8", "bytes": 1291007603 } }, "variants": { "fp16": 2220256556, "fp32": 3414640924, "q4": 717652150, "q4f16": 646838112, "q8": 1291007603 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Qwen3.5-0.8B-ONNX" } ], "notes": "Vision-language chat has no stable transformers.js pipeline() task yet; use the model's AutoModel classes directly." }, { "id": "granite-docling-258m", "name": "Granite-Docling-258M", "hf_repo": "onnx-community/granite-docling-258M-ONNX", "task": "vision-language", "params_m": 257.52, "headline": { "webgpu": { "variant": "q4f16", "bytes": 264311910 }, "wasm": { "variant": "q8", "bytes": 472120286 } }, "variants": { "fp16": 632192648, "fp32": 1263876953, "q4": 400028103, "q4f16": 264311910, "q8": 472120286 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/granite-docling-258M-ONNX" } ], "notes": "Document-understanding VLM; no pipeline() task covers it yet. The uint8 and bnb4 builds in this repo are missing files the other variants ship, and are excluded here as incomplete." }, { "id": "florence-2-large-ft", "name": "Florence-2-large-ft", "hf_repo": "onnx-community/Florence-2-large-ft", "task": "vision-language", "params_m": 770.43, "headline": { "webgpu": { "variant": "q4f16", "bytes": 582232182 }, "wasm": { "variant": "uint8", "bytes": 829497938 } }, "variants": { "bnb4": 744202849, "fp16": 1648741163, "fp32": 3294625906, "q4": 790572038, "q4f16": 582232182, "q8": 1658995841, "uint8": 829497938 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Florence-2-large-ft" } ], "notes": "Same task-prompt-token usage as Florence-2-base (see that model); no pipeline() call, use Florence2ForConditionalGeneration and AutoProcessor directly.", "code_snippet": "import { Florence2ForConditionalGeneration, AutoProcessor, AutoTokenizer } from \"@huggingface/transformers\";\n\nconst model = await Florence2ForConditionalGeneration.from_pretrained(\"onnx-community/Florence-2-large-ft\", {\n device: \"webgpu\",\n // Per-module dtype map: keep the language-model parts in fp16/fp32 for\n // quality, quantize the (much larger) vision encoder/decoder to q4.\n dtype: {\n embed_tokens: \"fp16\", // or \"fp32\" where WebGPU fp16 (shader-f16) isn't supported\n vision_encoder: \"fp16\", // or \"fp32\" where WebGPU fp16 isn't supported\n encoder_model: \"q4\",\n decoder_model_merged: \"q4\",\n },\n});\nconst processor = await AutoProcessor.from_pretrained(\"onnx-community/Florence-2-large-ft\");\nconst tokenizer = await AutoTokenizer.from_pretrained(\"onnx-community/Florence-2-large-ft\");\n\nconst prompt = \"\"; // task-prompt token, not free text\n// const inputs = await processor(image, prompt);\n// const generated = await model.generate({ ...inputs, max_new_tokens: 100 });\n// const result = processor.batch_decode(generated, { skip_special_tokens: false });" }, { "id": "whisper-tiny", "name": "Whisper Tiny", "hf_repo": "onnx-community/whisper-tiny", "task": "speech-to-text", "params_m": 37.8, "headline": { "webgpu": { "variant": "fp16", "bytes": 76113088 }, "wasm": { "variant": "uint8", "bytes": 40844261 } }, "variants": { "bnb4": 94702865, "fp16": 76113088, "fp32": 151458819, "q4": 95734369, "q8": 81688449, "uint8": 40844261 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/whisper-tiny" }, { "label": "Parameter count source", "url": "https://huggingface.co/openai/whisper-tiny" } ] }, { "id": "whisper-base", "name": "Whisper Base", "hf_repo": "onnx-community/whisper-base", "task": "speech-to-text", "params_m": 72.6, "headline": { "webgpu": { "variant": "fp16", "bytes": 146060430 }, "wasm": { "variant": "uint8", "bytes": 76894657 } }, "variants": { "bnb4": 139623558, "fp16": 146060430, "fp32": 290989606, "q4": 142374870, "q8": 153789241, "uint8": 76894657 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/whisper-base" }, { "label": "Parameter count source", "url": "https://huggingface.co/openai/whisper-base" } ] }, { "id": "whisper-small", "name": "Whisper Small", "hf_repo": "onnx-community/whisper-small", "task": "speech-to-text", "params_m": 241.7, "headline": { "webgpu": { "variant": "fp16", "bytes": 485190832 }, "wasm": { "variant": "uint8", "bytes": 249077066 } }, "variants": { "bnb4": 286947383, "fp16": 485190832, "fp32": 968150171, "q4": 299331431, "q8": 498153977, "uint8": 249077066 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/whisper-small" }, { "label": "Parameter count source", "url": "https://huggingface.co/openai/whisper-small" } ] }, { "id": "moonshine-base", "name": "Moonshine Base", "hf_repo": "onnx-community/moonshine-base-ONNX", "task": "speech-to-text", "params_m": 61.5, "headline": { "webgpu": { "variant": "q4f16", "bytes": 101727600 }, "wasm": { "variant": "uint8", "bytes": 62916138 } }, "variants": { "bnb4": 94770106, "fp16": 201203834, "fp32": 247030126, "q4": 97537620, "q4f16": 101727600, "q8": 125927995, "uint8": 62916138 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/moonshine-base-ONNX" }, { "label": "Parameter count source", "url": "https://huggingface.co/UsefulSensors/moonshine-base" } ] }, { "id": "whisper-large-v3-turbo", "name": "Whisper Large v3 Turbo", "hf_repo": "onnx-community/whisper-large-v3-turbo", "task": "speech-to-text", "params_m": 808.88, "headline": { "webgpu": { "variant": "q4f16", "bytes": 563479095 }, "wasm": { "variant": "uint8", "bytes": 1084758905 } }, "variants": { "bnb4": 713216977, "fp16": 1618569942, "fp32": 3236346357, "q4": 759089997, "q4f16": 563479095, "q8": 2169517721, "uint8": 1084758905 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/whisper-large-v3-turbo" } ] }, { "id": "moonshine-tiny", "name": "Moonshine Tiny", "hf_repo": "onnx-community/moonshine-tiny-ONNX", "task": "speech-to-text", "params_m": 27.09, "headline": { "webgpu": { "variant": "q4f16", "bytes": 56024215 }, "wasm": { "variant": "uint8", "bytes": 28109400 } }, "variants": { "bnb4": 54388824, "fp16": 91770581, "fp32": 109109881, "q4": 55382958, "q4f16": 56024215, "q8": 56290294, "uint8": 28109400 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/moonshine-tiny-ONNX" } ] }, { "id": "voxtral-mini-4b-realtime", "name": "Voxtral Mini 4B Realtime", "hf_repo": "onnx-community/Voxtral-Mini-4B-Realtime-2602-ONNX", "task": "speech-to-text", "params_m": 4429.68, "headline": { "webgpu": { "variant": "q4f16", "bytes": 2835604336 }, "wasm": { "variant": "q8", "bytes": 3724575097 } }, "variants": { "fp16": 4871297853, "fp32": 5753690980, "q4": 2926531867, "q4f16": 2835604336, "q8": 3724575097 }, "pipeline_task": "automatic-speech-recognition", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Voxtral-Mini-4B-Realtime-2602-ONNX" } ], "notes": "Built for realtime streaming transcription; the standard automatic-speech-recognition pipeline() call works for one-shot audio, but streaming needs custom chunking beyond the default pipeline. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today." }, { "id": "pyannote-segmentation-3.0", "name": "pyannote Speaker Segmentation 3.0", "hf_repo": "onnx-community/pyannote-segmentation-3.0", "task": "speaker-diarization", "params_m": null, "headline": { "webgpu": { "variant": "q4f16", "bytes": 2929425 }, "wasm": { "variant": "uint8", "bytes": 1542308 } }, "variants": { "bnb4": 5815342, "fp16": 3000918, "fp32": 5986908, "q4": 5818448, "q4f16": 2929425, "q8": 3084612, "uint8": 1542308 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/pyannote-segmentation-3.0" } ], "notes": "Frame-wise speaker segmentation, not transcription. transformers.js has no dedicated diarization pipeline() task; use AutoModel and decode the frame-level speaker-activity output yourself." }, { "id": "kokoro-82m", "name": "Kokoro-82M", "hf_repo": "onnx-community/Kokoro-82M-v1.0-ONNX", "task": "text-to-speech", "params_m": 82, "headline": { "webgpu": { "variant": "q4f16", "bytes": 154586422 }, "wasm": { "variant": "q8", "bytes": 92361116 } }, "variants": { "fp16": 163234740, "fp32": 325532232, "q4": 305215966, "q4f16": 154586422, "q8": 92361116, "uint8": 177464632 }, "pipeline_task": "text-to-speech", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX" }, { "label": "Parameter count source", "url": "https://huggingface.co/hexgrad/Kokoro-82M" } ], "notes": "Sizes are not monotonic by quant here: q8 (about 88 MB) is smaller than q4f16 (about 147 MB) and is the community-standard build most demos ship. In practice Kokoro is usually driven through the dedicated kokoro-js package rather than a generic pipeline() call, and kokoro-js recommends the fp32 build on WebGPU: the q4f16 build produces audibly degraded audio there (verified in Chrome on Apple silicon, 2026-08-01)." }, { "id": "supertonic-tts-2", "name": "Supertonic TTS 2", "hf_repo": "onnx-community/Supertonic-TTS-2-ONNX", "task": "text-to-speech", "params_m": null, "headline": { "webgpu": { "variant": "fp32", "bytes": 262872737 }, "wasm": { "variant": "fp32", "bytes": 262872737 } }, "variants": { "fp32": 262872737 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Supertonic-TTS-2-ONNX" } ], "notes": "Ships as three separate ONNX parts (text encoder, latent denoiser, voice decoder) in fp32 only, no quantized build published yet. There is no pipeline() task that wires a 3-stage TTS graph together; expect custom ONNX Runtime code." }, { "id": "all-minilm-l6-v2", "name": "all-MiniLM-L6-v2", "hf_repo": "Xenova/all-MiniLM-L6-v2", "task": "embeddings", "params_m": 22.7, "headline": { "webgpu": { "variant": "q4f16", "bytes": 30018257 }, "wasm": { "variant": "uint8", "bytes": 22845806 } }, "variants": { "bnb4": 53915448, "fp16": 45297825, "fp32": 90387606, "q4": 54578772, "q4f16": 30018257, "q8": 45944740, "uint8": 22845806 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/Xenova/all-MiniLM-L6-v2" }, { "label": "Parameter count source", "url": "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2" } ] }, { "id": "bge-small-en-v1.5", "name": "bge-small-en-v1.5", "hf_repo": "Xenova/bge-small-en-v1.5", "task": "embeddings", "params_m": 33.4, "headline": { "webgpu": { "variant": "q4f16", "bytes": 36190171 }, "wasm": { "variant": "uint8", "bytes": 33760859 } }, "variants": { "bnb4": 60147542, "fp16": 66749212, "fp32": 133093490, "q4": 61474190, "q4f16": 36190171, "q8": 67775257, "uint8": 33760859 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/Xenova/bge-small-en-v1.5" }, { "label": "Parameter count source", "url": "https://huggingface.co/BAAI/bge-small-en-v1.5" } ] }, { "id": "embeddinggemma-300m", "name": "EmbeddingGemma-300M", "hf_repo": "onnx-community/embeddinggemma-300m-ONNX", "task": "embeddings", "params_m": 302.9, "headline": { "webgpu": { "variant": "q4f16", "bytes": 176115397 }, "wasm": { "variant": "q8", "bytes": 309458498 } }, "variants": { "fp16": 618089375, "fp32": 1235001020, "q4": 197245082, "q4f16": 176115397, "q8": 309458498 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/embeddinggemma-300m-ONNX" }, { "label": "Parameter count source", "url": "https://huggingface.co/google/embeddinggemma-300m" } ] }, { "id": "gte-multilingual-base", "name": "GTE Multilingual Base", "hf_repo": "onnx-community/gte-multilingual-base", "task": "embeddings", "params_m": 305, "headline": { "webgpu": { "variant": "q4f16", "bytes": 465204574 }, "wasm": { "variant": "uint8", "bytes": 340318797 } }, "variants": { "bnb4": 866226340, "fp16": 627988827, "fp32": 1255502649, "q4": 873303844, "q4f16": 465204574, "q8": 680637594, "uint8": 340318797 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gte-multilingual-base" } ], "notes": "Param count is the model card's stated figure (no safetensors metadata to read directly), so treat it as approximate." }, { "id": "qwen3-embedding-0.6b", "name": "Qwen3-Embedding-0.6B", "hf_repo": "onnx-community/Qwen3-Embedding-0.6B-ONNX", "task": "embeddings", "params_m": 595.78, "headline": { "webgpu": { "variant": "q4f16", "bytes": 567458583 }, "wasm": { "variant": "uint8", "bytes": 613527631 } }, "variants": { "bnb4": 886593402, "fp16": 1200510818, "fp32": 2400598343, "q4": 914121462, "q4f16": 567458583, "q8": 1227055170, "uint8": 613527631 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/Qwen3-Embedding-0.6B-ONNX" } ] }, { "id": "bge-m3", "name": "BGE-M3", "hf_repo": "onnx-community/bge-m3-ONNX", "task": "embeddings", "params_m": 569, "headline": { "webgpu": { "variant": "q4f16", "bytes": 699926120 }, "wasm": { "variant": "uint8", "bytes": 568479466 } }, "variants": { "bnb4": 1248100641, "fp32": 2267319617, "q4": 1248100641, "q4f16": 699926120, "q8": 1136958790, "uint8": 568479466 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/bge-m3-ONNX" } ], "notes": "Param count is BAAI's stated figure (no safetensors metadata to read directly), so treat it as approximate. The bnb4 and q4 builds in this repo are byte-identical (apparently the same file uploaded under both names); treat them as one option. The fp16 bucket is a near-empty placeholder file and is excluded here." }, { "id": "voyage-4-nano", "name": "voyage-4-nano", "hf_repo": "onnx-community/voyage-4-nano-ONNX", "task": "embeddings", "params_m": 346.45, "headline": { "webgpu": { "variant": "q4f16", "bytes": 211073759 }, "wasm": { "variant": "q8", "bytes": 421893932 } }, "variants": { "fp16": 703605360, "fp32": 1406994406, "q4": 243267303, "q4f16": 211073759, "q8": 421893932 }, "pipeline_task": "feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/voyage-4-nano-ONNX" } ] }, { "id": "bge-reranker-v2-m3", "name": "BGE Reranker v2 M3", "hf_repo": "onnx-community/bge-reranker-v2-m3-ONNX", "task": "reranker", "params_m": 567.76, "headline": { "webgpu": { "variant": "q4f16", "bytes": 702120377 }, "wasm": { "variant": "uint8", "bytes": 570727094 } }, "variants": { "bnb4": 1233652957, "fp16": 1136209678, "fp32": 2271745547, "q4": 1252526149, "q4f16": 702120377, "q8": 1141454188, "uint8": 570727094 }, "pipeline_task": "text-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/bge-reranker-v2-m3-ONNX" } ], "notes": "Rerankers score a (query, passage) pair, not a single string. Confirm the model card's expected input shape before assuming the generic text-classification pipeline() call handles pairs; AutoModelForSequenceClassification with a tokenizer pair encoding is the safer default." }, { "id": "clip-vit-base-patch32", "name": "CLIP ViT-B/32", "hf_repo": "Xenova/clip-vit-base-patch32", "task": "vision", "params_m": 151.3, "headline": { "webgpu": { "variant": "q4f16", "bytes": 251617632 }, "wasm": { "variant": "uint8", "bytes": 305458290 } }, "variants": { "bnb4": 363373339, "fp16": 606935621, "fp32": 1211543291, "q4": 378788443, "q4f16": 251617632, "q8": 460036878, "uint8": 305458290 }, "pipeline_task": "zero-shot-image-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/Xenova/clip-vit-base-patch32" }, { "label": "Parameter count source", "url": "https://huggingface.co/openai/clip-vit-base-patch32" } ] }, { "id": "rmbg-1.4", "name": "RMBG-1.4", "hf_repo": "briaai/RMBG-1.4", "task": "vision", "params_m": 44.1, "headline": { "webgpu": { "variant": "fp16", "bytes": 88217533 }, "wasm": { "variant": "q8", "bytes": 44403226 } }, "variants": { "fp16": 88217533, "fp32": 176153355, "q8": 44403226 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/briaai/RMBG-1.4" } ], "notes": "The repo's config declares a custom architecture, so there is no working pipeline() call for this model; use the AutoModel flow shown in the snippet (the same one the official remove-background demo ships).", "code_snippet": "import { AutoModel, AutoProcessor, RawImage } from \"@huggingface/transformers\";\n\n// RMBG-1.4's config uses a custom architecture string, so pipeline() rejects\n// it; load it as a custom model with an explicit processor config instead\n// (this mirrors the official remove-background-web demo).\nconst model = await AutoModel.from_pretrained(\"briaai/RMBG-1.4\", {\n config: { model_type: \"custom\" },\n device: \"webgpu\", // or \"wasm\"\n dtype: \"fp16\", // use \"q8\" for the WASM build\n});\nconst processor = await AutoProcessor.from_pretrained(\"briaai/RMBG-1.4\", {\n config: {\n do_normalize: true, do_pad: false, do_rescale: true, do_resize: true,\n image_mean: [0.5, 0.5, 0.5], image_std: [1, 1, 1],\n feature_extractor_type: \"ImageFeatureExtractor\",\n resample: 2, rescale_factor: 1 / 255, size: { width: 1024, height: 1024 },\n },\n});\n\nconst image = await RawImage.fromURL(\"https://your-image-url.jpg\");\nconst { pixel_values } = await processor(image);\nconst { output } = await model({ input: pixel_values });\n// output[0] is the alpha mask; resize it to the source and composite.\nconst mask = await RawImage.fromTensor(output[0].mul(255).to(\"uint8\"))\n .resize(image.width, image.height);" }, { "id": "modnet", "name": "MODNet", "hf_repo": "Xenova/modnet", "task": "vision", "params_m": 6.5, "headline": { "webgpu": { "variant": "q4f16", "bytes": 11801931 }, "wasm": { "variant": "uint8", "bytes": 6627048 } }, "variants": { "bnb4": 23080899, "fp16": 12984781, "fp32": 25888640, "q4": 23132083, "q4f16": 11801931, "q8": 6632188, "uint8": 6627048 }, "pipeline_task": "background-removal", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/Xenova/modnet" }, { "label": "Parameter count source", "url": "https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/public/modnet-photographic-portrait-matting/README.md" } ] }, { "id": "depth-anything-v2-small", "name": "Depth Anything V2 Small", "hf_repo": "onnx-community/depth-anything-v2-small", "task": "vision", "params_m": 24.8, "headline": { "webgpu": { "variant": "q4f16", "bytes": 19126267 }, "wasm": { "variant": "uint8", "bytes": 27258801 } }, "variants": { "bnb4": 26077648, "fp16": 49642442, "fp32": 99060839, "q4": 27404416, "q4f16": 19126267, "q8": 54517602, "uint8": 27258801 }, "pipeline_task": "depth-estimation", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/depth-anything-v2-small" }, { "label": "Parameter count source", "url": "https://huggingface.co/depth-anything/Depth-Anything-V2-Small-hf" } ] }, { "id": "slimsam-77", "name": "SlimSAM-77", "hf_repo": "Xenova/slimsam-77-uniform", "task": "vision", "params_m": 9.7, "headline": { "webgpu": { "variant": "fp16", "bytes": 20720775 }, "wasm": { "variant": "q8", "bytes": 13785975 } }, "variants": { "fp16": 20720775, "fp32": 39833906, "q8": 13785975 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/Xenova/slimsam-77-uniform" }, { "label": "Parameter count source", "url": "https://huggingface.co/Zigeng/SlimSAM-uniform-77" } ], "notes": "transformers.js has no mask-generation pipeline() task; SAM-family models are driven with the raw SamModel and AutoProcessor classes plus a click or box prompt.", "code_snippet": "import { SamModel, AutoProcessor, RawImage } from \"@huggingface/transformers\";\n\nconst model = await SamModel.from_pretrained(\"Xenova/slimsam-77-uniform\", {\n device: \"webgpu\",\n dtype: \"fp16\", // use \"q8\" for the WASM build\n});\nconst processor = await AutoProcessor.from_pretrained(\"Xenova/slimsam-77-uniform\");\n\nconst image = await RawImage.read(\"https://your-image-url.jpg\");\nconst input_points = [[[340, 250]]]; // one (x, y) click point on the object to mask\nconst inputs = await processor(image, { input_points });\nconst outputs = await model(inputs);\n\nconst masks = await processor.post_process_masks(\n outputs.pred_masks,\n inputs.original_sizes,\n inputs.reshaped_input_sizes,\n);" }, { "id": "siglip2-base-patch16-224", "name": "SigLIP2 Base Patch16-224", "hf_repo": "onnx-community/siglip2-base-patch16-224-ONNX", "task": "vision", "params_m": 375.19, "headline": { "webgpu": { "variant": "q4f16", "bytes": 994828997 }, "wasm": { "variant": "uint8", "bytes": 755991743 } }, "variants": { "bnb4": 1790152569, "fp16": 1501811944, "fp32": 3002563235, "q4": 1812195067, "q4f16": 994828997, "q8": 1511983486, "uint8": 755991743 }, "pipeline_task": "zero-shot-image-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/siglip2-base-patch16-224-ONNX" } ] }, { "id": "dinov3-vits16", "name": "DINOv3 ViT-S/16", "hf_repo": "onnx-community/dinov3-vits16-pretrain-lvd1689m-ONNX", "task": "vision", "params_m": 21.6, "headline": { "webgpu": { "variant": "q4", "bytes": 14836561 }, "wasm": { "variant": "q8", "bytes": 21945937 } }, "variants": { "fp32": 86485745, "q4": 14836561, "q8": 21945937 }, "pipeline_task": "image-feature-extraction", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/dinov3-vits16-pretrain-lvd1689m-ONNX" } ], "notes": "This repo has no fp16 build (only fp32/q4/q8); q4 is the smallest option and stands in as the WebGPU headline here." }, { "id": "grounding-dino-tiny", "name": "Grounding DINO Tiny", "hf_repo": "onnx-community/grounding-dino-tiny-ONNX", "task": "vision", "params_m": 172.28, "headline": { "webgpu": { "variant": "q4f16", "bytes": 151069879 }, "wasm": { "variant": "uint8", "bytes": 203824675 } }, "variants": { "bnb4": 218129274, "fp16": 360393267, "fp32": 718761381, "q4": 227229745, "q4f16": 151069879, "q8": 407649156, "uint8": 203824675 }, "pipeline_task": "zero-shot-object-detection", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/grounding-dino-tiny-ONNX" } ] }, { "id": "sam2.1-hiera-tiny", "name": "SAM 2.1 Hiera Tiny", "hf_repo": "onnx-community/sam2.1-hiera-tiny-ONNX", "task": "vision", "params_m": 38.96, "headline": { "webgpu": { "variant": "q4f16", "bytes": 33667410 }, "wasm": { "variant": "uint8", "bytes": 61965721 } }, "variants": { "bnb4": 49551078, "fp16": 78004244, "fp32": 155610424, "q4": 51481493, "q4f16": 33667410, "q8": 123932099, "uint8": 61965721 }, "pipeline_task": null, "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/sam2.1-hiera-tiny-ONNX" } ], "notes": "Same as SlimSAM: transformers.js has no mask-generation pipeline() task. SAM 2 is a distinct architecture from SAM 1, so it uses the Sam2Model and Sam2Processor classes (not SamModel/SamProcessor) with a point or box prompt.", "code_snippet": "import { Sam2Model, Sam2Processor, RawImage } from \"@huggingface/transformers\";\n\nconst model = await Sam2Model.from_pretrained(\"onnx-community/sam2.1-hiera-tiny-ONNX\", {\n device: \"webgpu\",\n dtype: \"q4f16\", // use \"uint8\" for the WASM build\n});\nconst processor = await Sam2Processor.from_pretrained(\"onnx-community/sam2.1-hiera-tiny-ONNX\");\n\nconst image = await RawImage.read(\"https://your-image-url.jpg\");\nconst input_points = [[[340, 250]]]; // one (x, y) click point on the object to mask\nconst inputs = await processor(image, { input_points });\nconst outputs = await model(inputs);\n\nconst masks = await processor.post_process_masks(\n outputs.pred_masks,\n inputs.original_sizes,\n inputs.reshaped_input_sizes,\n);" }, { "id": "birefnet-lite", "name": "BiRefNet Lite", "hf_repo": "onnx-community/BiRefNet_lite-ONNX", "task": "vision", "params_m": 44.36, "headline": { "webgpu": { "variant": "fp16", "bytes": 114538221 }, "wasm": { "variant": "fp16", "bytes": 114538221 } }, "variants": { "fp16": 114538221, "fp32": 224005088 }, "pipeline_task": "background-removal", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/BiRefNet_lite-ONNX" } ], "notes": "Only fp16 and fp32 builds are published for this repo, no quantized variant yet." }, { "id": "ben2", "name": "BEN2", "hf_repo": "onnx-community/BEN2-ONNX", "task": "vision", "params_m": 94.63, "headline": { "webgpu": { "variant": "fp16", "bytes": 219121675 }, "wasm": { "variant": "fp16", "bytes": 219121675 } }, "variants": { "fp16": 219121675 }, "pipeline_task": "background-removal", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/BEN2-ONNX" } ], "notes": "Only a single fp16 build is published for this repo." }, { "id": "ormbg", "name": "ORMBG", "hf_repo": "onnx-community/ormbg-ONNX", "task": "vision", "params_m": null, "headline": { "webgpu": { "variant": "q4f16", "bytes": 88117949 }, "wasm": { "variant": "uint8", "bytes": 44315205 } }, "variants": { "bnb4": 176116038, "fp16": 88117930, "fp32": 176116019, "q4": 176116038, "q4f16": 88117949, "q8": 88630341, "uint8": 44315205 }, "pipeline_task": "background-removal", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/ormbg-ONNX" } ] }, { "id": "punctuate-all", "name": "Punctuate All", "hf_repo": "onnx-community/punctuate-all-ONNX", "task": "utility", "params_m": 278.89, "headline": { "webgpu": { "variant": "q4f16", "bytes": 433149933 }, "wasm": { "variant": "uint8", "bytes": 278658098 } }, "variants": { "bnb4": 818186877, "fp16": 555239208, "fp32": 1110154243, "q4": 823495046, "q4f16": 433149933, "q8": 557316196, "uint8": 278658098 }, "pipeline_task": "token-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/punctuate-all-ONNX" } ], "notes": "Param count is read from its xlm-roberta-base backbone (the token-classification head adds negligibly more); treat it as approximate." }, { "id": "gliner-small-v2.1", "name": "GLiNER Small v2.1", "hf_repo": "onnx-community/gliner_small-v2.1", "task": "utility", "params_m": 166, "headline": { "webgpu": { "variant": "q4f16", "bytes": 245226118 }, "wasm": { "variant": "uint8", "bytes": 183403759 } }, "variants": { "bnb4": 463106905, "fp16": 306253040, "fp32": 611293061, "q4": 463106905, "q4f16": 245226118, "q8": 366807468, "uint8": 183403759 }, "pipeline_task": "token-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/gliner_small-v2.1" } ], "notes": "Param count is the model card's stated figure, so treat it as approximate. GLiNER uses a span-based architecture rather than plain per-token labels; some setups need the dedicated GLiNER inference code rather than the generic token-classification pipeline()." }, { "id": "piiranha-v1", "name": "Piiranha v1 (PII Detection)", "hf_repo": "onnx-community/piiranha-v1-detect-personal-information-ONNX", "task": "utility", "params_m": 278.23, "headline": { "webgpu": { "variant": "q4f16", "bytes": 453136858 }, "wasm": { "variant": "uint8", "bytes": 317144829 } }, "variants": { "bnb4": 857781721, "fp16": 575239383, "fp32": 1149780766, "q4": 863090465, "q4f16": 453136858, "q8": 634289658, "uint8": 317144829 }, "pipeline_task": "token-classification", "sources": [ { "label": "HuggingFace (sizes measured from HF API 2026-08-01)", "url": "https://huggingface.co/onnx-community/piiranha-v1-detect-personal-information-ONNX" } ] } ]