Supported Models

This page is the model-support entry point. Use the table to find an exact Hugging Face checkpoint and TRTMC profile, then open Model Recipes to browse declared E2E recipes by task and model family.

One table, one source

The table below is rendered from website/data/model-support-matrix.md; the project README links to this page instead of carrying a second copy. The docs build resolves each row's TRTMC family from the exact matching E2E manifest profile. Hugging Face model_type, architectures, and Diffusers pipeline-class values come from the revision-pinned metadata snapshot in website/data/hf-model-metadata.json; the same snapshot renders every family recipe page.

How to read support

• A row applies only to its exact hf_id, checkpoint resolution, TRTMC profile, and build configuration.
• model_type identifies the Hugging Face config family. Architecture names identify the checkpoint class or, for Diffusers checkpoints, the pipeline class. They are not interchangeable with the TRTMC family ID.
• Architecture metadata describes the source checkpoint. The supported TRTMC task/head contract is shown separately and documented on the linked family page. It may intentionally consume only the base graph, such as an encoder that returns hidden states rather than a pretraining head.
• For a manifest without hf_revision, the metadata snapshot SHA pins only the displayed configuration fields; it does not retroactively turn that unpinned recipe into an exact-revision support claim.
• If a retained release row and the current manifest disagree on hf_id, the table shows both values. The GB300 light remains attached to the release-row checkpoint; the current manifest value identifies today's declared CLI recipe.
• Untested fine-tunes from the same family are best-effort compatible; they are not verified supported checkpoints.
• Platform specialization identifies a provider integration for an exact checkpoint or tuple. It does not replace native-runtime platform support.
• Performance color is evidence for the dated release comparison described below, not a blanket compatibility or correctness guarantee.

Release performance snapshot

This is the completed July 29, 2026 release comparison on NVIDIA GB300 at source revision 508613d0bcc7003b123cf5be3d1b3f6e6c6cb667. It covers 105 unique single-process release profiles across 76 families. Distributed and L0 smoke profiles are outside this matrix.

Each light compares TRTMC inference p50 with the row's declared reference under matching workload, output, and timing contracts:

• 🟢 Green: TRTMC is more than 5% faster than the reference.
• 🟡 Yellow: TRTMC is within 5% of the reference.
• 🔴 Red: TRTMC is more than 5% slower than the reference.
• — Not supported: the profile is explicitly unsupported on that platform.

Bundle preparation, model loading, compilation when used, and warmup are excluded from infer-p50 values. Baselines are declared per row and are not uniformly torch.compile. See the revision-matched comparison contract and benchmark documentation for reproduction and interpretation details.

TRTMC family	HF model_type	HF architecture / pipeline class	TRTMC task/head contract	Hugging Face ID (hf_id, CLI input)	Checkpoint ID (TRTMC profile)	GB300 performance	Precision	Quantization	Optimized runtime dispatch
albert	albert	AlbertForMaskedLM Source: config.architectures	encoder_only_nlp	albert/albert-base-v2 Architecture metadata: config.json at 8e2f239c5f8a	albert-base	🟢 Green	FP16	None	No
bert	bert	BertModel Source: config.architectures	encoder_only_nlp	sentence-transformers/all-MiniLM-L6-v2 Architecture metadata: config.json at 1110a243fdf4	all-minilm-l6-v2	🟢 Green	FP16	None	No
mpnet	mpnet	MPNetForMaskedLM Source: config.architectures	encoder_only_nlp	sentence-transformers/all-mpnet-base-v2 Architecture metadata: config.json at e8c3b32edf54	all-mpnet-base-v2	🟢 Green	FP16	None	No
bark	bark	BarkModel Source: config.architectures	text_to_audio	suno/bark Architecture metadata: config.json at 70a8a7d34168	bark-large	🟢 Green	FP32	None	No
bark	bark	BarkModel Source: config.architectures	text_to_audio	suno/bark-small Architecture metadata: config.json at 1dbd7a128513	bark-small	🟢 Green	FP16	None	No
bart	bart	BartModel Source: config.architectures	text_generation_causal	facebook/bart-base Architecture metadata: config.json at aadd2ab0ae0c	bart-base	🟢 Green	FP16	None	No
bert	bert	BertForMaskedLM Source: config.architectures	encoder_only_nlp	google-bert/bert-base-uncased Architecture metadata: config.json at 86b5e0934494	bert-base-uncased	🟢 Green	FP16	None	No
bert	bert	BertModel Source: config.architectures	encoder_only_nlp	BAAI/bge-small-en-v1.5 Architecture metadata: config.json at 5c38ec7c405e	bge-small-en-v1.5	🟢 Green	FP16	None	No
bloom	bloom	BloomForCausalLM Source: config.architectures	text_generation_causal	bigscience/bloom-560m Architecture metadata: config.json at ac2ae5fab2ce	bloom-560m	🟢 Green	FP32	None	No
roberta	camembert	CamembertForMaskedLM Source: config.architectures	encoder_only_nlp	almanach/camembert-base Architecture metadata: config.json at a75967561c78	camembert-base	🟢 Green	FP16	None	No
canary	—	nemo.collections.asr.models.aed_multitask_models.EncDecMultiTaskModel Source: model_config.target	speech_to_text	nvidia/canary-1b-v2 Architecture metadata: canary-1b-v2.nemo:model_config.yaml at 87bc52657add	canary-1b-v2	🟢 Green	FP16	None	No
chronos_bolt	t5	ChronosBoltModelForForecasting Source: config.architectures	neural_operator	amazon/chronos-bolt-tiny Architecture metadata: config.json at a0e552de8349	chronos-bolt-tiny-official	🟢 Green	FP32	None	No
codegen	codegen	CodeGenForCausalLM Source: config.architectures	text_generation_causal	Salesforce/codegen-350M-mono Architecture metadata: config.json at d9107f71cca4	codegen-350m	🟢 Green	FP16	None	No
convbert	convbert	ConvBertModel Source: config.architectures	encoder_only_nlp	YituTech/conv-bert-base Architecture metadata: config.json at 5cb451936b5c	convbert-base	🟢 Green	FP16	None	No
deberta	deberta	— Not declared by checkpoint metadata	encoder_only_nlp	microsoft/deberta-base Architecture metadata: config.json at 0d1b43ccf21b	deberta-base	🟢 Green	FP16	None	No
deepseek_ocr	deepseek_vl_v2	DeepseekOCR2ForCausalLM Source: config.architectures	vision_language_generation	deepseek-ai/DeepSeek-OCR-2 Architecture metadata: config.json at aaa02f381194	deepseek-ocr	🟢 Green	FP16 FP32 layers: 6, 7, 8, 9, 10, 11, 12	None	No
deepseek_v2	deepseek_v2	DeepseekV2ForCausalLM Source: config.architectures	text_generation_causal	deepseek-ai/DeepSeek-V2-Lite Architecture metadata: config.json at 604d5664dddd	deepseek-v2-lite	🟢 Green	FP16	None	No
deepseek_v2	deepseek_v3	DeepseekV3ForCausalLM Source: config.architectures	text_generation_causal	yujiepan/deepseek-v3-tiny-random Current manifest hf_id: katuni4ka/tiny-random-deepseek-v3 Architecture metadata: config.json at cc35c93ec832	deepseek-v2-tiny	🟢 Green	FP16	None	No
distilbert	distilbert	DistilBertForMaskedLM Source: config.architectures	encoder_only_nlp	distilbert/distilbert-base-uncased Architecture metadata: config.json at 12040accade4	distilbert-base-uncased	🟢 Green	FP16	None	No
gpt2	gpt2	GPT2LMHeadModel Source: config.architectures	text_generation_causal	distilbert/distilgpt2 Architecture metadata: config.json at 2290a62682d0	distilgpt2	🟢 Green	FP16	None	No
dpr	dpr	DPRContextEncoder Source: config.architectures	encoder_only_nlp	facebook/dpr-ctx_encoder-single-nq-base Architecture metadata: config.json at bb21a3c2b165	dpr-ctx-encoder	🟢 Green	FP16	None	No
electra	electra	ElectraForPreTraining Source: config.architectures	encoder_only_nlp	google/electra-base-discriminator Architecture metadata: config.json at 1ae76a97c7e8	electra-base-discriminator	🟢 Green	FP16	None	No
falcon	falcon	FalconForCausalLM Source: config.architectures	text_generation_causal	tiiuae/falcon-rw-1b Architecture metadata: config.json at e4b9872bb803	falcon-rw-1b	🟢 Green	FP16	None	No
llama	llama	LlamaForCausalLM Source: config.architectures	text_generation_causal	tiiuae/Falcon3-1B-Base Architecture metadata: config.json at cb37ef3559b1	falcon3-1b	🟢 Green	FP16	None	No
flux	—	Flux2Pipeline Source: model_index._class_name	diffusion_media_generation	black-forest-labs/FLUX.2-dev Architecture metadata: model_index.json at 26afe3a78bb2	flux-2-dev	🟢 Green	FP16	None	No
flux	—	Flux2Pipeline Source: model_index._class_name	diffusion_media_generation	black-forest-labs/FLUX.2-dev Architecture metadata: model_index.json at 26afe3a78bb2	flux-2-dev-fp8	🟢 Green	FP16	fp8_scales=data/flux2-fp8-scales.json	No
flux	—	FluxPipeline Source: model_index._class_name	diffusion_media_generation	black-forest-labs/FLUX.1-schnell Architecture metadata: model_index.json at 741f7c3ce8b3	flux-schnell	🟢 Green	FP16	None	No
fnet	fnet	FNetForPreTraining Source: config.architectures	encoder_only_nlp	google/fnet-base Architecture metadata: config.json at d89b6fad3cf5	fnet-base	🟢 Green	FP16	None	No
gemma	gemma2	Gemma2ForCausalLM Source: config.architectures	text_generation_causal	google/gemma-2-2b-it Architecture metadata: config.json at 299a8560bedf	gemma-2-2b	🟢 Green	FP16	None	No
glm	glm	GlmForCausalLM Source: config.architectures	text_generation_causal	THUDM/glm-4-9b-hf Architecture metadata: config.json at b44e98fcc8df	glm-4-9b	🟢 Green	FP16	None	No
gpt_neo	gpt_neo	GPTNeoForCausalLM Source: config.architectures	text_generation_causal	EleutherAI/gpt-neo-125m Architecture metadata: config.json at 21def0189f57	gpt-neo-125m	🟢 Green	FP16	None	No
gpt_oss	gpt_oss	GptOssForCausalLM Source: config.architectures	text_generation_causal	openai/gpt-oss-20b Architecture metadata: config.json at 6cee5e81ee83	gpt-oss-20b	🟢 Green	FP16	None	No
gpt2	gpt2	GPT2LMHeadModel Source: config.architectures	text_generation_causal	openai-community/gpt2 Architecture metadata: config.json at 607a30d783df	gpt2-125m	🟢 Green	FP32	None	No
granite	granite	GraniteForCausalLM Source: config.architectures	text_generation_causal	ibm-granite/granite-3.1-2b-base Architecture metadata: config.json at bcf4e33e7deb	granite-3.1-2b	🟢 Green	FP16	None	No
internlm	internlm2	InternLM2ForCausalLM Source: config.architectures	text_generation_causal	internlm/internlm2-math-plus-1_8b Architecture metadata: config.json at 998dd540b927	internlm2-1.8b	🟢 Green	FP16	None	No
internvl	internvl	InternVLForConditionalGeneration Source: config.architectures	vision_language_generation	OpenGVLab/InternVL3-2B-hf Architecture metadata: config.json at cb57a075cb75	internvl3-2b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
internvl	internvl	InternVLForConditionalGeneration Source: config.architectures	vision_language_generation	OpenGVLab/InternVL3-8B-hf Architecture metadata: config.json at 259a3b64a146	internvl3-8b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
lance	—	Lance Source: config.model_name	vision_language_generation	bytedance-research/Lance Architecture metadata: config.json at 739531575886	lance-3b-x2t-image	🟢 Green	BF16	None	No
locateanything	locateanything	LocateAnythingForConditionalGeneration Source: config.architectures	vision_language_generation	nvidia/LocateAnything-3B Architecture metadata: config.json at c32291ca5e99	locateanything-3b	🟢 Green	FP16	None	No
magpie_tts	decoder_ce	nemo.collections.tts.models.magpietts.MagpieTTSModel Source: model_config.target	text_to_audio	nvidia/magpie_tts_multilingual_357m Revision: 34d7e40da85cabc97f92198889b65cea27bc7fd1 Architecture metadata: magpie_tts_multilingual_357m.nemo:model_config.yaml at 34d7e40da85c	magpie-tts-357m	🟢 Green	FP32	None	No
mamba	mamba	MambaForCausalLM Source: config.architectures	text_generation_causal	state-spaces/mamba-130m-hf Architecture metadata: config.json at 1e76775f628f	mamba-130m	🟢 Green	FP32	None	No
marian	marian	MarianMTModel Source: config.architectures	text_generation_causal	Helsinki-NLP/opus-mt-en-ru Architecture metadata: config.json at bb09c99d1800	marian-en-ru	🟢 Green	FP16	None	No
llama	llama	LlamaForCausalLM Source: config.architectures	text_generation_causal	nvidia/Llama-3.1-Minitron-4B-Depth-Base Architecture metadata: config.json at 47f17b2c73bb	minitron-4b-depth	🟢 Green	FP16	None	No
llama	llama	LlamaForCausalLM Source: config.architectures	text_generation_causal	nvidia/Llama-3.1-Minitron-4B-Width-Base Architecture metadata: config.json at 5205ef7d3620	minitron-4b-width	🟢 Green	FP32 (manifest default)	None	No
mistral	mistral	MistralForCausalLM Source: config.architectures	text_generation_causal	mistralai/Mistral-7B-Instruct-v0.1 Architecture metadata: config.json at ec5deb64f2c6	mistral-7b	🟢 Green	FP16	None	No
mixtral	mixtral	MixtralForCausalLM Source: config.architectures	text_generation_causal	ggml-org/stories15M_MOE Architecture metadata: config.json at b6dd73749746	mixtral-stories-15m	🟢 Green	FP16 FP32 layers: 4	None	No
modernbert	modernbert	ModernBertForMaskedLM Source: config.architectures	encoder_only_nlp	answerdotai/ModernBERT-base Architecture metadata: config.json at 8949b909ec90	modernbert-base	🟢 Green	FP32	None	No
nemotron_speech_streaming	nemotron3_5_asr	Nemotron3_5AsrForRNNT Source: config.architectures	speech_to_text	nvidia/nemotron-3.5-asr-streaming-0.6b Architecture metadata: config.json at 1c8deaecc64b	nemotron-3.5-asr-streaming-0.6b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
eagle_vlm	llama_nemotron_vl	LlamaNemotronVLModel Source: config.architectures	embedding	nvidia/llama-nemotron-embed-vl-1b-v2 Architecture metadata: config.json at 582e3bf72aee	nemotron-embed-vl-1b-v2	🟢 Green	FP16	None	No
nemotron_h	nemotron_h	NemotronHForCausalLM Source: config.architectures	text_generation_causal	nvidia/NVIDIA-Nemotron-Nano-9B-v2 Architecture metadata: config.json at 6533e8de2c68	nemotron-h-nano-9b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
nemotron	nemotron	NemotronForCausalLM Source: config.architectures	text_generation_causal	nvidia/Nemotron-4-Mini-Hindi-4B-Base Architecture metadata: config.json at 05c6ef6d2dda	nemotron-hindi-4b	🟢 Green	FP16	None	No
nemotron_labs_diffusion	nemotron_labs_diffusion	NemotronLabsDiffusionModel Source: config.architectures	text_generation_causal	nvidia/Nemotron-Labs-Diffusion-8B Architecture metadata: config.json at 16c67f0560b9	nemotron-labs-diffusion-8b	🟢 Green	BF16	None	No
nemotron	nemotron	NemotronForCausalLM Source: config.architectures	text_generation_causal	nvidia/Nemotron-Mini-4B-Instruct Architecture metadata: config.json at 791833e92ebd	nemotron-mini-4b	🟢 Green	FP16	None	No
llama	llama	LlamaForCausalLM Source: config.architectures	text_generation_causal	nvidia/Llama-3.1-Nemotron-Nano-4B-v1.1 Architecture metadata: config.json at d552708a9d57	nemotron-nano-4b	🟢 Green	FP16	None	No
eagle_vlm	llama_nemotron_vl_rerank	LlamaNemotronVLForSequenceClassification Source: config.architectures	reranking	nvidia/llama-nemotron-rerank-vl-1b-v2 Architecture metadata: config.json at 9c20c4aedf9e	nemotron-rerank-vl-1b-v2	🟢 Green	FP16	None	No
nemotron_speech_streaming	nemotron_asr_streaming	NemotronAsrStreamingForRNNT Source: config.architectures	speech_to_text	nvidia/nemotron-speech-streaming-en-0.6b Architecture metadata: config.json at ebe59e5a8171	nemotron-speech-streaming-en-0.6b	🟢 Green	FP16	None	No
m2m_100	m2m_100	M2M100ForConditionalGeneration Source: config.architectures	text_generation_causal	facebook/nllb-200-distilled-600M Architecture metadata: config.json at f8d333a098d1	nllb-200-distilled-600m	🟢 Green	FP16	None	No
olmo	olmo	OlmoForCausalLM Source: config.architectures	text_generation_causal	allenai/OLMo-1B-hf Architecture metadata: config.json at aee7752d9c08	olmo-1b	🟢 Green	FP16	None	No
olmo2	olmo2	Olmo2ForCausalLM Source: config.architectures	text_generation_causal	allenai/OLMo-2-0425-1B Architecture metadata: config.json at a1847dff3500	olmo2-1b	🟢 Green	FP16	None	No
opt	opt	OPTForCausalLM Source: config.architectures	text_generation_causal	facebook/opt-125m Architecture metadata: config.json at 27dcfa74d334	opt-125m	🟢 Green	FP16	None	No
bert	bert	BertModel Source: config.architectures	encoder_only_nlp	sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 Architecture metadata: config.json at e8f8c211226b	paraphrase-multilingual-minilm-l12-v2	🟢 Green	FP16	None	No
patchtsmixer	patchtsmixer	PatchTSMixerForPrediction Source: config.architectures	neural_operator	ibm-granite/granite-timeseries-patchtsmixer Architecture metadata: config.json at 90dc5a88d45f	patchtsmixer-granite-official	🟢 Green	FP16	None	No
patchtst	patchtst	PatchTSTForRegression Source: config.architectures	neural_operator	ibm-research/patchtst-etth1-regression-distribution Architecture metadata: config.json at d5ef731ed80f	patchtst-etth1-regression-distribution	🟢 Green	FP16 FP32 layers: 5	None	No
patchtst	patchtst	PatchTSTForPrediction Source: config.architectures	neural_operator	ibm-granite/granite-timeseries-patchtst Architecture metadata: config.json at 7fe295d8bc8f	patchtst-granite-official	🟢 Green	FP16	None	No
personaplex	personaplex	— Not declared by checkpoint metadata	speech_to_speech	nvidia/personaplex-7b-v1 Architecture metadata: config.json at fdaf4090a61c	personaplex-7b	🟢 Green	FP16 FP32 layers: 0, 1	None	No
phi_moe	phimoe	PhiMoEForCausalLM Source: config.architectures	text_generation_causal	microsoft/Phi-tiny-MoE-instruct Architecture metadata: config.json at 2fe50e88d0e2	phi-moe	🟢 Green	FP16	None	No
phi	phi3	Phi3ForCausalLM Source: config.architectures	text_generation_causal	microsoft/Phi-3-mini-4k-instruct Architecture metadata: config.json at f39ac1d28e92	phi3-mini	🟢 Green	FP16	None	No
phi4_multimodal	phi4mm	Phi4MMForCausalLM Source: config.architectures	vision_language_generation	microsoft/Phi-4-multimodal-instruct Architecture metadata: config.json at 93f923e1a772	phi4-multimodal	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
pixart	—	PixArtSigmaPipeline Source: model_index._class_name	diffusion_media_generation	PixArt-alpha/PixArt-Sigma-XL-2-1024-MS Architecture metadata: model_index.json at e102b3591cc8	pixart-sigma-1024	🟢 Green	FP16 FP32 layers: 0	None	No
gpt_neox	gpt_neox	GPTNeoXForCausalLM Source: config.architectures	text_generation_causal	EleutherAI/pythia-70m Architecture metadata: config.json at a39f36b100fe	pythia-70m	🟢 Green	FP32	None	No
qwen_image	—	QwenImagePipeline Source: model_index._class_name	diffusion_media_generation	Qwen/Qwen-Image Architecture metadata: model_index.json at 75e0b4be04f6	qwen-image	🟢 Green	BF16	None	No
qwen_image	—	QwenImagePipeline Source: model_index._class_name	diffusion_media_generation	Qwen/Qwen-Image-2512 Architecture metadata: model_index.json at 25468b98e327	qwen-image-2512	🟢 Green	BF16	None	No
qwen_image	—	QwenImageEditPlusPipeline Source: model_index._class_name	diffusion_media_generation	Qwen/Qwen-Image-Edit-2511 Architecture metadata: model_index.json at 6f3ccc0b56e4	qwen-image-edit-2511	🟢 Green	BF16	None	No
qwen_vl	qwen2_5_vl	Qwen2_5_VLForConditionalGeneration Source: config.architectures	vision_language_generation	Qwen/Qwen2.5-VL-3B-Instruct Architecture metadata: config.json at 66285546d2b8	qwen25vl-3b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
qwen	qwen3	Qwen3ForCausalLM Source: config.architectures	text_generation_causal	Qwen/Qwen3-0.6B Architecture metadata: config.json at c1899de289a0	qwen3-0.6b-fp16	🟢 Green	FP16	None	Yes TensorRT Edge-LLM Qualified TRTMC dispatch target: Linux x86_64, NVIDIA A100 80GB PCIe (SM80), FP16 Additional dispatch targets: Coming soon
qwen	qwen3	Qwen3ForCausalLM Source: config.architectures	text_generation_causal	Qwen/Qwen3-0.6B Architecture metadata: config.json at c1899de289a0	qwen3-0.6b-fp8	🟢 Green	BF16	format=fp8 scale_source=modelopt calibration_samples=64	Yes TensorRT Edge-LLM Qualified TRTMC dispatch target: Linux x86_64, NVIDIA A100 80GB PCIe (SM80), FP16 Additional dispatch targets: Coming soon
qwen	qwen3	Qwen3ForCausalLM Source: config.architectures	text_generation_causal	Qwen/Qwen3-0.6B Architecture metadata: config.json at c1899de289a0	qwen3-0.6b-topp	🟢 Green	FP16	None	Yes TensorRT Edge-LLM Qualified TRTMC dispatch target: Linux x86_64, NVIDIA A100 80GB PCIe (SM80), FP16 Additional dispatch targets: Coming soon
qwen	qwen3	Qwen3ForCausalLM Source: config.architectures	text_generation_causal	Qwen/Qwen3-4B-Instruct-2507 Architecture metadata: config.json at cdbee75f17c0	qwen3-4b-instruct-2507	🟢 Green	FP16	None	Yes TensorRT Edge-LLM Qualified TRTMC dispatch target: Linux x86_64, NVIDIA A100 80GB PCIe (SM80), FP16 Additional dispatch targets: Coming soon
qwen_moe	qwen3_moe	Qwen3MoeForCausalLM Source: config.architectures	text_generation_causal	Qwen/Qwen3-30B-A3B Architecture metadata: config.json at ad44e777bcd1	qwen3-moe-30b-a3b	🟢 Green	FP16	None	No
qwen_moe	qwen3_moe	Qwen3MoeForCausalLM Source: config.architectures	text_generation_causal	amd-quark/tiny-random-qwen3_moe Architecture metadata: config.json at ca2aaa9a82e9	qwen3-moe-tiny-random	🟢 Green	FP16	None	No
qwen3_omni	qwen3_omni_moe	Qwen3OmniMoeForConditionalGeneration Source: config.architectures	omni_multimodal	Qwen/Qwen3-Omni-30B-A3B-Instruct Architecture metadata: config.json at 26291f793822	qwen3-omni-30b-a3b-instruct	🔴 Red	BF16	None	No
qwen_vl	qwen3_vl	Qwen3VLForConditionalGeneration Source: config.architectures	vision_language_generation	Qwen/Qwen3-VL-2B-Instruct Architecture metadata: config.json at 89644892e4d8	qwen3-vl-2b	🟢 Green	FP16 FP32 layers: 0, 1, 2	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
qwen3_5	qwen3_5	Qwen3_5ForConditionalGeneration Source: config.architectures	text_generation_causal	Qwen/Qwen3.5-9B Architecture metadata: config.json at c20223623576	qwen35-9b	🟢 Green	FP16	None	Coming soon TensorRT Edge-LLM Qualified TRTMC dispatch target: Coming soon
mistral	mistral	MistralForCausalLM Source: config.architectures	text_generation_causal	nvidia/Riva-Translate-4B-Instruct-v1.1 Architecture metadata: config.json at 1eb944ce9f27	riva-translate-4b	🟢 Green	FP16	None	No
roberta	roberta	RobertaForMaskedLM Source: config.architectures	encoder_only_nlp	FacebookAI/roberta-base Architecture metadata: config.json at e2da8e2f811d	roberta-base	🟢 Green	FP16	None	No
roberta	roberta	RobertaForMaskedLM Source: config.architectures	encoder_only_nlp	FacebookAI/roberta-large Architecture metadata: config.json at 722cf37b1afa	roberta-large	🟢 Green	FP16	None	No
rwkv	rwkv	RwkvForCausalLM Source: config.architectures	text_generation_causal	RWKV/rwkv-4-169m-pile Architecture metadata: config.json at 46bdc280eb97	rwkv-169m	🟢 Green	FP32	None	No
sam	sam	SamModel Source: config.architectures	prompted_segmentation	facebook/sam-vit-base Architecture metadata: config.json at 70c1a07f894e	sam-vit-base	🟢 Green	FP16	None	No
sam3	sam3_video	Sam3VideoModel Source: config.architectures	prompted_segmentation	facebook/sam3 Architecture metadata: config.json at 3c879f39826c	sam3	🟢 Green	FP32	None	No
sana_wm	—	SanaMSVideoCamCtrl_1600M_P1_D20 Source: config.model.model	diffusion_media_generation	Efficient-Large-Model/SANA-WM_bidirectional Architecture metadata: config.yaml at e96271d77398	sana-wm-bidirectional	🟡 Yellow	BF16	None	No
segformer	segformer	SegformerForSemanticSegmentation Source: config.architectures	segmentation	nvidia/segformer-b0-finetuned-ade-512-512 Architecture metadata: config.json at 489d5cd81a0b	segformer-b0-ade	🟢 Green	FP16	None	No
stablelm	stablelm	StableLmForCausalLM Source: config.architectures	text_generation_causal	stabilityai/stablelm-2-1_6b Architecture metadata: config.json at f499ead74c53	stablelm2-1.6b	🟢 Green	FP16	None	No
starcoder2	starcoder2	Starcoder2ForCausalLM Source: config.architectures	text_generation_causal	bigcode/starcoder2-3b Architecture metadata: config.json at 733247c55e3f	starcoder2-3b	🟢 Green	FP16	None	No
t5	t5	T5ForConditionalGeneration Source: config.architectures	text_generation_causal	google-t5/t5-small Architecture metadata: config.json at df1b051c4962	t5-small	🟢 Green	FP16	None	No
timesfm	timesfm	TimesFmModelForPrediction Source: config.architectures	neural_operator	google/timesfm-2.0-500m-pytorch Architecture metadata: config.json at dc2443792ce5	timesfm-2.0-500m-official	🟢 Green	FP32	None	No
timm_vit	—	vit_base_patch16_224 Source: config.architecture	image_classification	timm/vit_base_patch16_224.augreg_in21k_ft_in1k Architecture metadata: config.json at 2ec9fb3d7bb6	timm-vit-base-p16-224-augreg-in21k-ft-in1k	🟢 Green	FP16	None	No
llama	llama	LlamaForCausalLM Source: config.architectures	text_generation_causal	TinyLlama/TinyLlama-1.1B-Chat-v1.0 Architecture metadata: config.json at fe8a4ea1ffed	tinyllama-1.1b	🟢 Green	FP16	None	No
wan_t2v	—	WanPipeline Source: model_index._class_name	diffusion_media_generation	Wan-AI/Wan2.1-T2V-1.3B-Diffusers Architecture metadata: model_index.json at 0fad780a534b	wan21-t2v-1.3b	🟢 Green	FP16 FP32 layers: 24	None	No
wan2_2_ti2v	ti2v	WanModel Source: config._class_name	diffusion_media_generation	Wan-AI/Wan2.2-TI2V-5B Revision: 921dbaf3f1674a56f47e83fb80a34bac8a8f203e Architecture metadata: config.json at 921dbaf3f167	wan22-ti2v-5b	🟢 Green	BF16	None	No
whisper	whisper	WhisperForConditionalGeneration Source: config.architectures	speech_to_text	openai/whisper-large-v3-turbo Architecture metadata: config.json at 41f01f3fe87f	whisper-large-v3-turbo	🟢 Green	FP32	None	No
whisper	whisper	WhisperForConditionalGeneration Source: config.architectures	speech_to_text	openai/whisper-tiny Architecture metadata: config.json at 169d4a4341b3	whisper-tiny-fp16	🟢 Green	FP16 FP32 layers: 0	None	No
xglm	xglm	XGLMForCausalLM Source: config.architectures	text_generation_causal	facebook/xglm-564M Architecture metadata: config.json at f3059f01b98c	xglm-564m	🟢 Green	FP16	None	No
roberta	xlm-roberta	XLMRobertaForMaskedLM Source: config.architectures	encoder_only_nlp	FacebookAI/xlm-roberta-base Architecture metadata: config.json at e73636d4f797	xlm-roberta-base	🟢 Green	FP16	None	No
xlnet	xlnet	XLNetLMHeadModel Source: config.architectures	encoder_only_nlp	xlnet/xlnet-base-cased Architecture metadata: config.json at ceaa69c7bc5e	xlnet-base	🟢 Green	FP16	None	No
z_image	—	ZImagePipeline Source: model_index._class_name	diffusion_media_generation	Tongyi-MAI/Z-Image-Turbo Architecture metadata: model_index.json at f332072aa78b	z-image-turbo	🔴 Red	FP16 FP32 layers: 2, 3, 4, 7, 8	None	No

Declared recipes beyond the release snapshot

The support table is a retained release snapshot. The current checkout may contain additional E2E manifests, L0 replacements, and distributed variants. Those declarations are available through Model Recipes, organized as Hugging Face task → model family → exact manifest recipe.

Platform specializations will roll out in phases aligned with model coverage available in TensorRT Edge-LLM and TensorRT-Model-Connect releases. Each batch must document the exact qualified model, target, precision, and configuration tuple.