Since directly searching via url works, I put together a list of things you can put into the search bar to filter models:
num_parameters=min:X,max:Y
-
X/Y can be replaced with 0, 3B, 6B, 9B, 12B, 24B, 32B, 64B, 128B, 256B, 512B
-
Example: num_parameters=min:12B,max:24B sort=X
-
X can be replaced with trending, likes, downloads, created, modified, alphabetical, most_params, least_params
search=X
-X can be replaced with anything you’re searching for (ie, model name) (No need to add the weird %20 symbols you may see when you normally search, it searches just fine with spaces
-Example: Qwen MTP
-Example: mistral
-Example: gemma
pipeline_tag=X
-(The Multimodal Category): any-to-any, audio-text-to-audio, document-question-answering, visual-document-retrieval, image-text-to-text, image-text-to-image, image-text-to-video, video-text-to-text, visual-question-answering
-(The Natural Language Processing Category): feature-extraction, fill-mask, question-answering, sentence-similarity, summarization, table-question-answering, text-classification, text-generation, text-ranking, token-classification, translation, zero-shot-classification
-(The Computer Vision Category): depth-estimation, image-classification, image-feature-extraction, image-segmentation, image-to-image, image-to-text, image-to-video, keypoint-detection, mask-generation, object-detection, video-classification, text-to-image, text-to-video, unconditional-image-generation, video-to-video, zero-shot-classification, zero-shot-object-detection, text-to-3d, image-to-3d
-(The Audio Category): audio-classification, audio-to-audio, automatic-speech-recognition, text-to-speech
-(The Tabular Category): tabular-classification, tabular-regression
-(The Reinforcement Learning Category): reinforcement-learning
-Example: pipeline_tag=feature-extraction
-Example: pipeline_tag=text-to-image
I haven’t been successful in chaining pipeline tags
library=X
-X can be replaced with: pytorch, tf, jax, safetensors, transformers, peft, tensorboard, gguf, diffusers, onnx, stable-baselines3, sentence-transformers, mlx, ml-agents, keras, tf-keras, joblib, adapter-transformers, transformers.js, setfit, timm, openvino, sample-factory, flair, coreml, nemo, tflite, fastai, espnet, bertopic, spacy, fasttext, rust, sklearn, open_clip, keras-hub, KerasNLP, executorch, asteroid, speechbrain, allennlp, llamafile, fairseq, paddlepaddle, PaddleOCR, stanza, pyannote-audio, optimum_habana, span-marker, optimum_graphcore, paddlenlp, unity-sentis, dduf, univa
-Example: library=paddlenlp
-Example: library=pytorch, diffusers
language=X
-X can be replaced with: en, zh, fr, es, de, etc-- place the language code here
-Example: language=en
-Example: language=en,zh
-Example: language=fr
license=license:mit&other=moe&language=zh&:X
-X can be replaced with: apache-2.0, mit, other, openrail, creativeml-openrail-m, cc-by-nc-4.0, cc-by-4.0, gemma, llama3, openrail++, llama2, llama3.1, llama3.2, cc-by-nc-sa-4.0, afl-3.0, cc-by-sa-4.0, gpl-3.0, bigscience-bloom-rail-1.0, bigscience-openrail-m, artistic-2.0, llama3.3, bigcode-openrail-m, cc, bsd-3-clause, agpl-3.0, cc-by-nc-nd-4.0, cc0-1.0, wtfpl, unilicense, bsl-1.0, bsd-2-clause, bsd, gpl, llama4, c-uda, cc-by-sa-3.0, bsd-3-clause-clear, cc-by-2.0, cc-by-nc-2.0, cdla-permissive-2.0, cc-by-nd-4.0, cc-by-3.0, gpl-2.0, cc-by-nc-3.0, apple-amlr, cc-by-2.5, lgpl-3.0, mpl-2.0, osl-3.0, cc-by-nd-3.0, gfdl, cc-by-nc-sa-3.0, pddl, cc-by-nc-sa-2.0, ecl-2.0, fair-noncommercial-research-license, ms-pl, apple-ascl, deepfloyd-if-license, etalab-2.0, odc-by, epl-2.0, eupl-1.2, cdla-sharing-1.0, grok2-community, cdla-permissive-1.0, lgpl, eupl-1.1, odbl, zlib, epl-1.0, lppl-1.3c, lgpl-2.1, isc, openmdw-1.0, ncsa, lgpl-lr, intel-research, h-research, postgresql, ofl-1.1, open-mdw
-Example: license=license:wtfpl
-Example: license=license:cc-by-nc-nd-4.0
other=X
-X can be replaced with anything you see tagged in a model card; This includes things in the Apps section of Other, the Misc section of Other, as well as tags assigned by model creators
(App tags) -llama.cpp, lmstudio, jan, drawthings, diffusionbee, joyfusion, vllm, ollama, mlx-lm, docker-model-runner, lemonade, sglang, unsloth, pi, hermes-agent
(Misc tags)
-endpoints_compatible, text-generation-inference, model-index, text-embeddings-inference, 4-bit, custom_code, merge, 8-bit, moe, co2_eq_emissions, eval-results
(Examples of unofficial tags)
-conversational, uncensored, abliterated, heretic, mistral, apple-silicon, mtp, etc.
-Example: other=lmstudio
-Example: other=moe
-Example: other=uncensored
-Example: other=llama.cpp, mtp
-Example: other=conversational, mistral, heretic
-Example: other=conversational, mistral, heretic, 4-bit
(Note: not all mixture of experts models are tagged as moe, and some are tagged with alterations of the moe tag)
inference_provider=X
-Where X can be replaced with: groq, novita, cerabras, sambanova, nscale, fal-ai, hyperbolic, together, fireworks-ai, featherless-ai, zai-org, replicate, cohere, scaleway, publicai, ovhcloud, hf-inference, deepinfra, wavespeed
-Example: inference_provider=deepinfra
-Example: inference_provider=deepinfra&inference_provider=cohere
author=X
-X can be replaced with the name of the author
-Example: author=sentence-transformers
-Example: author=google
-Example: author=Qwen
Lets say you want to search with the following criterium:
-It’s a Mixture-Of-Experts (moe)
-It supports Chinese (zh)
-It’s at least 9B parameters and at most 32B parameters
-It can ‘see’ images / read text, and returns text
-And it’s under an mit license
Example: /models?license=license:mit&other=moe&language=zh&num_parameters=min:9B,max:32B&pipeline_tag=image-text-to-text Lets say you want to search with the following criterium:
-It’s made by unsloth
-It’s uses the transformers and gguf library
-It’s imatrix and conversational
-And we want to see the most recently updated model
-And, it has MTP in the name
-Example: /models?search=MTP&author=unsloth&library=transformers, gguf&other=imatrix, conversational&sort=modified