Skip to content

Supported models

Sonar reads the architectures field from a model’s config.json file. The value must match an architecture in these tables. A compatible Transformers implementation can also use the Transformers backend.

The current registry contains 369 architecture entries in the user-facing categories below.

Text generation

ArchitectureImplementation
AXK1ForCausalLMAXK1:AXK1ForCausalLM
AfmoeForCausalLMafmoe:AfmoeForCausalLM
ApertusForCausalLMapertus:ApertusForCausalLM
ArceeForCausalLMarcee:ArceeForCausalLM
ArcticForCausalLMarctic:ArcticForCausalLM
BailingMoeForCausalLMbailing_moe:BailingMoeForCausalLM
BailingMoeV2ForCausalLMbailing_moe:BailingMoeV2ForCausalLM
BailingMoeV2_5ForCausalLMbailing_moe_linear:BailingMoeV25ForCausalLM
BloomForCausalLMbloom:BloomForCausalLM
ChatGLMForConditionalGenerationchatglm:ChatGLMForCausalLM
ChatGLMModelchatglm:ChatGLMForCausalLM
Cohere2ForCausalLMcommandr:CohereForCausalLM
Cohere2MoeForCausalLMcohere2_moe:Cohere2MoeForCausalLM
CohereForCausalLMcommandr:CohereForCausalLM
CohereMoeForCausalLMcohere_moe:CohereMoeForCausalLM
CwmForCausalLMllama:LlamaForCausalLM
DbrxForCausalLMdbrx:DbrxForCausalLM
DeciLMForCausalLMnemotron_nas:DeciLMForCausalLM
DeepseekForCausalLMdeepseek_v2:DeepseekForCausalLM
DeepseekV2ForCausalLMdeepseek_v2:DeepseekV2ForCausalLM
DeepseekV32ForCausalLMdeepseek_v2:DeepseekV3ForCausalLM
DeepseekV3ForCausalLMdeepseek_v2:DeepseekV3ForCausalLM
DeepseekV4ForCausalLMaphrodite.models.deepseek_v4:DeepseekV4ForCausalLM
Ernie4_5ForCausalLMernie45:Ernie4_5ForCausalLM
Ernie4_5_MoeForCausalLMernie45_moe:Ernie4_5_MoeForCausalLM
Exaone4ForCausalLMexaone4:Exaone4ForCausalLM
ExaoneForCausalLMexaone:ExaoneForCausalLM
ExaoneMoEForCausalLMexaone_moe:ExaoneMoeForCausalLM
Fairseq2LlamaForCausalLMfairseq2_llama:Fairseq2LlamaForCausalLM
FalconForCausalLMfalcon:FalconForCausalLM
FalconH1ForCausalLMfalcon_h1:FalconH1ForCausalLM
FalconMambaForCausalLMmamba:MambaForCausalLM
FlexOlmoForCausalLMflex_olmo:FlexOlmoForCausalLM
GPT2LMHeadModelgpt2:GPT2LMHeadModel
GPTJForCausalLMgpt_j:GPTJForCausalLM
GPTNeoXForCausalLMgpt_neox:GPTNeoXForCausalLM
Gemma2ForCausalLMgemma2:Gemma2ForCausalLM
Gemma3ForCausalLMgemma3:Gemma3ForCausalLM
Gemma3nForCausalLMgemma3n:Gemma3nForCausalLM
Gemma4ForCausalLMgemma4:Gemma4ForCausalLM
GemmaForCausalLMgemma:GemmaForCausalLM
Glm4ForCausalLMglm4:Glm4ForCausalLM
Glm4MoeForCausalLMglm4_moe:Glm4MoeForCausalLM
Glm4MoeLiteForCausalLMglm4_moe_lite:Glm4MoeLiteForCausalLM
GlmForCausalLMglm:GlmForCausalLM
GlmMoeDsaForCausalLMdeepseek_v2:GlmMoeDsaForCausalLM
GptOssForCausalLMgpt_oss:GptOssForCausalLM
GraniteForCausalLMgranite:GraniteForCausalLM
GraniteMoeForCausalLMgranitemoe:GraniteMoeForCausalLM
GraniteMoeHybridForCausalLMgranitemoehybrid:GraniteMoeHybridForCausalLM
GraniteMoeSharedForCausalLMgranitemoeshared:GraniteMoeSharedForCausalLM
GritLMgritlm:GritLM
HCXVisionForCausalLMhyperclovax_vision:HCXVisionForCausalLM
HCXVisionV2ForCausalLMhyperclovax_vision_v2:HCXVisionV2ForCausalLM
HYV3ForCausalLMhy_v3:HYV3ForCausalLM
HrmTextForCausalLMhrm_text:HrmTextForCausalLM
HunYuanDenseV1ForCausalLMhunyuan_v1:HunYuanDenseV1ForCausalLM
HunYuanMoEV1ForCausalLMhunyuan_v1:HunYuanMoEV1ForCausalLM
HyperCLOVAXForCausalLMhyperclovax:HyperCLOVAXForCausalLM
IQuestCoderForCausalLMllama:LlamaForCausalLM
IQuestLoopCoderForCausalLMiquest_loopcoder:IQuestLoopCoderForCausalLM
InklingForCausalLMaphrodite.models.inkling:InklingForCausalLM
InklingForConditionalGenerationaphrodite.models.inkling:InklingForConditionalGeneration
InternLM2ForCausalLMinternlm2:InternLM2ForCausalLM
InternLM3ForCausalLMllama:LlamaForCausalLM
Jais2ForCausalLMjais2:Jais2ForCausalLM
JambaForCausalLMjamba:JambaForCausalLM
KimiLinearForCausalLMkimi_linear:KimiLinearForCausalLM
LLaMAForCausalLMllama:LlamaForCausalLM
LagunaForCausalLMlaguna:LagunaForCausalLM
Lfm2ForCausalLMlfm2:Lfm2ForCausalLM
Lfm2MoeForCausalLMlfm2_moe:Lfm2MoeForCausalLM
Llama4ForCausalLMllama4:Llama4ForCausalLM
LlamaForCausalLMllama:LlamaForCausalLM
LongcatFlashForCausalLMlongcat_flash:LongcatFlashForCausalLM
LongcatFlashNgramForCausalLMlongcat_flash_ngram:LongcatFlashNgramForCausalLM
MPTForCausalLMmpt:MPTForCausalLM
Mamba2ForCausalLMmamba2:Mamba2ForCausalLM
MambaForCausalLMmamba:MambaForCausalLM
MellumForCausalLMmellum:MellumForCausalLM
MiMoForCausalLMmimo:MiMoForCausalLM
MiMoV2FlashForCausalLMmimo_v2:MiMoV2FlashForCausalLM
MiMoV2ForCausalLMmimo_v2:MiMoV2ForCausalLM
MiniCPM3ForCausalLMminicpm3:MiniCPM3ForCausalLM
MiniCPMForCausalLMminicpm:MiniCPMForCausalLM
MiniMaxM2ForCausalLMminimax_m2:MiniMaxM2ForCausalLM
MiniMaxM3SparseForCausalLMaphrodite.models.minimax_m3:MiniMaxM3SparseForCausalLM
Ministral3ForCausalLMmistral:MistralForCausalLM
MistralForCausalLMmistral:MistralForCausalLM
MistralLarge3ForCausalLMmistral_large_3:MistralLarge3ForCausalLM
MixtralForCausalLMmixtral:MixtralForCausalLM
MptForCausalLMmpt:MPTForCausalLM
NemotronForCausalLMnemotron:NemotronForCausalLM
NemotronHForCausalLMnemotron_h:NemotronHForCausalLM
NemotronHPuzzleForCausalLMnemotron_h:NemotronHForCausalLM
OPTForCausalLMopt:OPTForCausalLM
Olmo3ForCausalLMolmo3:Olmo3ForCausalLM
OlmoHybridForCausalLMolmo_hybrid:OlmoHybridForCausalLM
OlmoeForCausalLMolmoe:OlmoeForCausalLM
OrionForCausalLMorion:OrionForCausalLM
PanguEmbeddedForCausalLMopenpangu:PanguEmbeddedForCausalLM
PanguProMoEV2ForCausalLMopenpangu:PanguProMoEV2ForCausalLM
PanguUltraMoEForCausalLMopenpangu:PanguUltraMoEForCausalLM
Param2MoEForCausalLMparam2moe:Param2MoEForCausalLM
Phi3ForCausalLMphi3:Phi3ForCausalLM
PhiForCausalLMphi:PhiForCausalLM
PhiMoEForCausalLMphimoe:PhiMoEForCausalLM
Plamo3ForCausalLMplamo3:Plamo3ForCausalLM
Qwen2ForCausalLMqwen2:Qwen2ForCausalLM
Qwen2MoeForCausalLMqwen2_moe:Qwen2MoeForCausalLM
Qwen3ForCausalLMqwen3:Qwen3ForCausalLM
Qwen3MoeForCausalLMqwen3_moe:Qwen3MoeForCausalLM
Qwen3NextForCausalLMqwen3_next:Qwen3NextForCausalLM
RWForCausalLMfalcon:FalconForCausalLM
Rnj1ForCausalLMrnj1:Rnj1ForCausalLM
SarvamMLAForCausalLMsarvam:SarvamMLAForCausalLM
SarvamMoEForCausalLMsarvam:SarvamMoEForCausalLM
SeedOssForCausalLMseed_oss:SeedOssForCausalLM
SolarForCausalLMsolar:SolarForCausalLM
StableLMEpochForCausalLMstablelm:StablelmForCausalLM
StableLmForCausalLMstablelm:StablelmForCausalLM
Step1ForCausalLMstep1:Step1ForCausalLM
Step3TextForCausalLMstep3_text:Step3TextForCausalLM
Step3p5ForCausalLMstep3p5:Step3p5ForCausalLM
TeleChat2ForCausalLMtelechat2:TeleChat2ForCausalLM
TeleChat3ForCausalLMllama:LlamaForCausalLM
TeleFLMForCausalLMteleflm:TeleFLMForCausalLM
Zamba2ForCausalLMzamba2:Zamba2ForCausalLM

Embedding

ArchitectureImplementation
BertForMaskedLMbert:BertForMaskedLM
BertModelbert:BertEmbeddingModel
BertSpladeSparseEmbeddingModelbert:BertSpladeSparseEmbeddingModel
BgeM3EmbeddingModelroberta:BgeM3EmbeddingModel
CLIPModelclip:CLIPEmbeddingModel
ColPaliForRetrievalcolpali:ColPaliModel
CwmForCausalLMllama:LlamaForCausalLM
DeciLMForCausalLMnemotron_nas:DeciLMForCausalLM
Gemma2Modelgemma2:Gemma2ForCausalLM
Gemma3TextModelgemma3:Gemma3Model
GlmForCausalLMglm:GlmForCausalLM
GritLMgritlm:GritLM
GteModelbert_with_rope:SnowflakeGteNewModel
GteNewModelbert_with_rope:GteNewModel
IQuestCoderForCausalLMllama:LlamaForCausalLM
InternLM3ForCausalLMllama:LlamaForCausalLM
JinaEmbeddingsV5Modeljina:JinaEmbeddingsV5Model
LLaMAForCausalLMllama:LlamaForCausalLM
LlamaBidirectionalModelllama:LlamaBidirectionalModel
LlamaForCausalLMllama:LlamaForCausalLM
LlamaModelllama:LlamaForCausalLM
LlamaNemotronVLModelnemotron_vl:LlamaNemotronVLForEmbedding
LlavaNextForConditionalGenerationllava_next:LlavaNextForConditionalGeneration
MistralModelllama:LlamaForCausalLM
ModernBertModelmodernbert:ModernBertModel
NomicBertModelbert_with_rope:NomicBertModel
Phi3ForCausalLMphi3:Phi3ForCausalLM
Phi3VForCausalLMphi3v:Phi3VForCausalLM
PrithviGeoSpatialMAEterratorch:Terratorch
Qwen2ForCausalLMqwen2:Qwen2ForCausalLM
Qwen2Modelqwen2:Qwen2ForCausalLM
Qwen2VLForConditionalGenerationqwen2_vl:Qwen2VLForConditionalGeneration
RobertaForMaskedLMroberta:RobertaEmbeddingModel
RobertaModelroberta:RobertaEmbeddingModel
SiglipModelsiglip:SiglipEmbeddingModel
TeleChat2ForCausalLMtelechat2:TeleChat2ForCausalLM
TeleChat3ForCausalLMllama:LlamaForCausalLM
Terratorchterratorch:Terratorch
VoyageQwen3BidirectionalEmbedModelvoyage:VoyageQwen3BidirectionalEmbedModel
XLMRobertaModelroberta:RobertaEmbeddingModel

Late interaction

ArchitectureImplementation
ColBERTJinaRobertaModelcolbert:ColBERTJinaRobertaModel
ColBERTLfm2Modelcolbert:ColBERTLfm2Model
ColBERTModernBertModelcolbert:ColBERTModernBertModel
ColModernVBertForRetrievalcolmodernvbert:ColModernVBertForRetrieval
ColPaliForRetrievalcolpali:ColPaliModel
ColQwen3colqwen3:ColQwen3Model
ColQwen3_5colqwen3_5:ColQwen3_5Model
HF_ColBERTcolbert:ColBERTModel
JinaForRankingjina:JinaForRanking
OpsColQwen3Modelcolqwen3:ColQwen3Model
Qwen3VLNemotronEmbedModelcolqwen3:ColQwen3Model

Reward

ArchitectureImplementation
InternLM2ForRewardModelinternlm2:InternLM2ForRewardModel
Qwen2ForProcessRewardModelqwen2_rm:Qwen2ForProcessRewardModel
Qwen2ForRewardModelqwen2_rm:Qwen2ForRewardModel

Token classification

ArchitectureImplementation
BertForTokenClassificationbert:BertForTokenClassification
ModernBertForTokenClassificationmodernbert:ModernBertForTokenClassification
OpenAIPrivacyFilterForTokenClassificationopenai_privacy_filter:OpenAIPrivacyFilterForTokenClassification
Qwen3ASRForcedAlignerForTokenClassificationqwen3_asr_forced_aligner:Qwen3ASRForcedAlignerForTokenClassification
RobertaForTokenClassificationroberta:RobertaForTokenClassification
XLMRobertaForTokenClassificationroberta:RobertaForTokenClassification

Sequence classification

ArchitectureImplementation
BertForSequenceClassificationbert:BertForSequenceClassification
GPT2ForSequenceClassificationgpt2:GPT2ForSequenceClassification
GteNewForSequenceClassificationbert_with_rope:GteNewForSequenceClassification
JambaForSequenceClassificationjamba:JambaForSequenceClassification
JinaVLForRankingjina_vl:JinaVLForSequenceClassification
LlamaBidirectionalForSequenceClassificationllama:LlamaBidirectionalForSequenceClassification
LlamaNemotronVLForSequenceClassificationnemotron_vl:LlamaNemotronVLForSequenceClassification
ModernBertForSequenceClassificationmodernbert:ModernBertForSequenceClassification
RobertaForSequenceClassificationroberta:RobertaForSequenceClassification
XLMRobertaForSequenceClassificationroberta:RobertaForSequenceClassification

Multimodal

ArchitectureImplementation
AriaForConditionalGenerationaria:AriaForConditionalGeneration
AudioFlamingo3ForConditionalGenerationaudioflamingo3:AudioFlamingo3ForConditionalGeneration
BagelForConditionalGenerationbagel:BagelForConditionalGeneration
BeeForConditionalGenerationbee:BeeForConditionalGeneration
Blip2ForConditionalGenerationblip2:Blip2ForConditionalGeneration
ChameleonForConditionalGenerationchameleon:ChameleonForConditionalGeneration
Cheerscheers:CheersForConditionalGeneration
CheersForConditionalGenerationcheers:CheersForConditionalGeneration
Cohere2VisionForConditionalGenerationcohere2_vision:Cohere2VisionForConditionalGeneration
CohereAsrForConditionalGenerationcohere_asr:CohereAsrForConditionalGeneration
Cosmos3EdgeForConditionalGenerationcosmos3_edge:Cosmos3EdgeForConditionalGeneration
Cosmos3ForConditionalGenerationcosmos3:Cosmos3ForConditionalGeneration
DeepseekOCR2ForCausalLMdeepseek_ocr2:DeepseekOCR2ForCausalLM
DeepseekOCRForCausalLMdeepseek_ocr:DeepseekOCRForCausalLM
DeepseekVLV2ForCausalLMdeepseek_vl2:DeepseekVLV2ForCausalLM
DiffusionGemmaForBlockDiffusiondiffusion_gemma:DiffusionGemmaForConditionalGeneration
DotsOCRForCausalLMdots_ocr:DotsOCRForCausalLM
Eagle2_5_VLForConditionalGenerationeagle2_5_vl:Eagle2_5_VLForConditionalGeneration
Ernie4_5_VLMoeForConditionalGenerationernie45_vl:Ernie4_5_VLMoeForConditionalGeneration
Exaone4_5_ForConditionalGenerationexaone4_5:Exaone4_5_ForConditionalGeneration
FireRedASR2ForConditionalGenerationfireredasr2:FireRedASR2ForConditionalGeneration
FireRedLIDForConditionalGenerationfireredlid:FireRedLIDForConditionalGeneration
FunASRForConditionalGenerationfunasr:FunASRForConditionalGeneration
FunAudioChatForConditionalGenerationfunaudiochat:FunAudioChatForConditionalGeneration
GLM4VForCausalLMglm4v:GLM4VForCausalLM
Gemma3ForConditionalGenerationgemma3_mm:Gemma3ForConditionalGeneration
Gemma3nForConditionalGenerationgemma3n_mm:Gemma3nForConditionalGeneration
Gemma4ForConditionalGenerationgemma4_mm:Gemma4ForConditionalGeneration
Gemma4UnifiedForConditionalGenerationgemma4_unified:Gemma4UnifiedForConditionalGeneration
Glm4vForConditionalGenerationglm4_1v:Glm4vForConditionalGeneration
Glm4vMoeForConditionalGenerationglm4_1v:Glm4vMoeForConditionalGeneration
GlmAsrForConditionalGenerationglmasr:GlmAsrForConditionalGeneration
GlmOcrForConditionalGenerationglm_ocr:GlmOcrForConditionalGeneration
Granite4VisionForConditionalGenerationgranite4_vision:Granite4VisionForConditionalGeneration
GraniteSpeechForConditionalGenerationgranite_speech:GraniteSpeechForConditionalGeneration
GraniteSpeechPlusForConditionalGenerationgranite_speech_plus:GraniteSpeechPlusForConditionalGeneration
H2OVLChatModelh2ovl:H2OVLChatModel
HfMoondreammoondream3:Moondream3ForCausalLM
HunYuanVLForConditionalGenerationhunyuan_vision:HunYuanVLForConditionalGeneration
Idefics3ForConditionalGenerationidefics3:Idefics3ForConditionalGeneration
InternS1ForConditionalGenerationinterns1:InternS1ForConditionalGeneration
InternS1ProForConditionalGenerationinterns1_pro:InternS1ProForConditionalGeneration
InternS2PreviewForConditionalGenerationinterns2_preview:InternS2PreviewForConditionalGeneration
InternVLChatModelinternvl:InternVLChatModel
InternVLForConditionalGenerationinterns1:InternS1ForConditionalGeneration
IsaacForConditionalGenerationisaac:IsaacForConditionalGeneration
KananaVForConditionalGenerationkanana_v:KananaVForConditionalGeneration
KeyeForConditionalGenerationkeye:KeyeForConditionalGeneration
KeyeVL1_5ForConditionalGenerationkeye_vl1_5:KeyeVL1_5ForConditionalGeneration
KimiK25ForConditionalGenerationkimi_k25:KimiK25ForConditionalGeneration
KimiVLForConditionalGenerationkimi_vl:KimiVLForConditionalGeneration
Lfm2VlForConditionalGenerationlfm2_vl:Lfm2VLForConditionalGeneration
LightOnOCRForConditionalGenerationlightonocr:LightOnOCRForConditionalGeneration
Llama4ForConditionalGenerationmllama4:Llama4ForConditionalGeneration
Llama_Nemotron_Nano_VLnemotron_vl:LlamaNemotronVLChatModel
LlavaForConditionalGenerationllava:LlavaForConditionalGeneration
LlavaNextForConditionalGenerationllava_next:LlavaNextForConditionalGeneration
LlavaNextVideoForConditionalGenerationllava_next_video:LlavaNextVideoForConditionalGeneration
LlavaOnevision2ForConditionalGenerationllava_onevision2:LlavaOnevision2ForConditionalGeneration
LlavaOnevisionForConditionalGenerationllava_onevision:LlavaOnevisionForConditionalGeneration
MiDashengLMModelmidashenglm:MiDashengLMModel
MiMoV2OmniForCausalLMmimo_v2_omni:MiMoV2OmniForCausalLM
MiniCPMOminicpmo:MiniCPMO
MiniCPMVminicpmv:MiniCPMV
MiniCPMV4_6ForConditionalGenerationminicpmv4_6:MiniCPMV4_6ForConditionalGeneration
MiniMaxM3SparseForConditionalGenerationaphrodite.models.minimax_m3:MiniMaxM3SparseForConditionalGeneration
Mistral3ForConditionalGenerationmistral3:Mistral3ForConditionalGeneration
Molmo2ForConditionalGenerationmolmo2:Molmo2ForConditionalGeneration
MolmoForCausalLMmolmo:MolmoForCausalLM
Moondream3ForCausalLMmoondream3:Moondream3ForCausalLM
MoonshotKimiaForCausalLMkimi_audio:KimiAudioForConditionalGeneration
MossAudioModelmoss_audio:MossAudioModel
MossTranscribeDiarizeForConditionalGenerationmoss_transcribe_diarize:MossTranscribeDiarizeForConditionalGeneration
NVLM_Dnvlm_d:NVLM_D_Model
NemotronH_Nano_Omni_Reasoning_V3nano_nemotron_vl:NemotronH_Nano_VL_V2
NemotronH_Nano_VL_V2nano_nemotron_vl:NemotronH_Nano_VL_V2
NemotronH_Super_Omni_Reasoning_V3nano_nemotron_vl:NemotronH_Nano_VL_V2
NemotronParseForConditionalGenerationnemotron_parse:NemotronParseForConditionalGeneration
OpenCUAForConditionalGenerationopencua:OpenCUAForConditionalGeneration
OpenPanguVLForConditionalGenerationopenpangu_vl:OpenPanguVLForConditionalGeneration
OpenVLAForActionPredictionopenvla:OpenVLAForActionPrediction
Ovisovis:Ovis
Ovis2_5ovis2_5:Ovis2_5
Ovis2_6ForCausalLMovis2_5:Ovis2_5
Ovis2_6_MoeForCausalLMovis2_5:Ovis2_5
PaddleOCRVLForConditionalGenerationpaddleocr_vl:PaddleOCRVLForConditionalGeneration
PaliGemmaForConditionalGenerationpaligemma:PaliGemmaForConditionalGeneration
Phi3VForCausalLMphi3v:Phi3VForCausalLM
Phi4ForCausalLMVphi4siglip:Phi4ForCausalLMV
Phi4MMForCausalLMphi4mm:Phi4MMForCausalLM
PixtralForConditionalGenerationpixtral:PixtralForConditionalGeneration
QianfanOCRForConditionalGenerationqianfan_ocr:QianfanOCRForConditionalGeneration
Qwen2AudioForConditionalGenerationqwen2_audio:Qwen2AudioForConditionalGeneration
Qwen2VLForConditionalGenerationqwen2_vl:Qwen2VLForConditionalGeneration
Qwen2_5OmniForConditionalGenerationqwen2_5_omni_thinker:Qwen2_5OmniThinkerForConditionalGeneration
Qwen2_5OmniModelqwen2_5_omni_thinker:Qwen2_5OmniThinkerForConditionalGeneration
Qwen2_5_VLForConditionalGenerationqwen2_5_vl:Qwen2_5_VLForConditionalGeneration
Qwen3ASRForConditionalGenerationqwen3_asr:Qwen3ASRForConditionalGeneration
Qwen3ASRRealtimeGenerationqwen3_asr_realtime:Qwen3ASRRealtimeGeneration
Qwen3OmniMoeForConditionalGenerationqwen3_omni_moe_thinker:Qwen3OmniMoeThinkerForConditionalGeneration
Qwen3VLForConditionalGenerationqwen3_vl:Qwen3VLForConditionalGeneration
Qwen3VLMoeForConditionalGenerationqwen3_vl_moe:Qwen3VLMoeForConditionalGeneration
Qwen3_5ForConditionalGenerationqwen3_5:Qwen3_5ForConditionalGeneration
Qwen3_5MoeForConditionalGenerationqwen3_5:Qwen3_5MoeForConditionalGeneration
RForConditionalGenerationrvl:RForConditionalGeneration
SkyworkR1VChatModelskyworkr1v:SkyworkR1VChatModel
SmolVLMForConditionalGenerationsmolvlm:SmolVLMForConditionalGeneration
Step3VLForConditionalGenerationstep3_vl:Step3VLForConditionalGeneration
Step3p7ForConditionalGenerationstep3p7:Step3p7ForConditionalGeneration
StepVLForConditionalGenerationstep_vl:StepVLForConditionalGeneration
UltravoxModelultravox:UltravoxModel
UnlimitedOCRForCausalLMunlimited_ocr:UnlimitedOCRForCausalLM
VoxtralForConditionalGenerationvoxtral:VoxtralForConditionalGeneration
VoxtralRealtimeGenerationvoxtral_realtime:VoxtralRealtimeGeneration
WhisperForConditionalGenerationwhisper:WhisperForConditionalGeneration

Speculative decoding

ArchitectureImplementation
BailingMoeV25MTPModelbailing_moe_mtp:BailingMoeV25MTPModel
DFlashDraftModelqwen3_dflash:DFlashQwen3ForCausalLM
DFlashLagunaForCausalLMlaguna_dflash:DFlashLagunaForCausalLM
DSparkDraftModelaphrodite.models.deepseek_v4:DSparkDeepseekV4ForCausalLM
DeepSeekMTPModeldeepseek_mtp:DeepSeekMTP
DeepSeekV4MTPModelaphrodite.models.deepseek_v4:DeepSeekV4MTP
Eagle3DeepseekV2ForCausalLMdeepseek_eagle3:Eagle3DeepseekV2ForCausalLM
Eagle3DeepseekV3ForCausalLMdeepseek_eagle3:Eagle3DeepseekV2ForCausalLM
Eagle3LlamaForCausalLMllama_eagle3:Eagle3LlamaForCausalLM
Eagle3MiniMaxM2ForCausalLMllama_eagle3:Eagle3LlamaForCausalLM
Eagle3Qwen2_5vlForCausalLMllama_eagle3:Eagle3LlamaForCausalLM
Eagle3Qwen3ForCausalLMqwen3_eagle3:Eagle3Qwen3ForCausalLM
Eagle3Qwen3vlForCausalLMllama_eagle3:Eagle3LlamaForCausalLM
EagleCohereForCausalLMcohere_eagle:EagleCohereForCausalLM
EagleDeepSeekMTPModeldeepseek_eagle:EagleDeepseekV3ForCausalLM
EagleLlama4ForCausalLMllama4_eagle:EagleLlama4ForCausalLM
EagleLlamaForCausalLMllama_eagle:EagleLlamaForCausalLM
EagleMiniCPMForCausalLMminicpm_eagle:EagleMiniCPMForCausalLM
EagleMistralForCausalLMmistral_eagle:EagleMistralForCausalLM
EagleMistralLarge3ForCausalLMmistral_large_3_eagle:EagleMistralLarge3ForCausalLM
ErnieMTPModelernie_mtp:ErnieMTP
Exaone4_5_MTPexaone4_5_mtp:Exaone4_5_MTP
ExaoneMoeMTPexaone_moe_mtp:ExaoneMoeMTP
ExtractHiddenStatesModelextract_hidden_states:ExtractHiddenStatesModel
Gemma4DSparkModelgemma4_dspark:Gemma4DSparkForCausalLM
Gemma4MTPModelgemma4_mtp:Gemma4MTP
Glm4MoeLiteMTPModelglm4_moe_lite_mtp:Glm4MoeLiteMTP
Glm4MoeMTPModelglm4_moe_mtp:Glm4MoeMTP
GlmOcrMTPModelglm_ocr_mtp:GlmOcrMTP
HYV3MTPModelhy_v3_mtp:HYV3MTP
InklingMTPModelaphrodite.models.inkling:InklingMTP
LlamaForCausalLMEagle3llama_eagle3:Eagle3LlamaForCausalLM
LongCatFlashMTPModellongcat_flash_mtp:LongCatFlashMTP
MedusaModelmedusa:Medusa
MiMoMTPModelmimo_mtp:MiMoMTP
MiMoV2MTPModelmimo_v2_mtp:MiMoV2MTP
MiMoV2OmniMTPModelmimo_v2_mtp:MiMoV2OmniMTP
MiniMaxM3MTPaphrodite.models.minimax_m3:MiniMaxM3MTP
NemotronHMTPModelnemotron_h_mtp:NemotronHMTP
OpenPanguMTPModelopenpangu_mtp:OpenPanguMTP
PEagleDraftModelllama_eagle3:Eagle3LlamaForCausalLM
PeagleLlamaForCausalLMllama_eagle3:Eagle3LlamaForCausalLM
PeagleQwen3ForCausalLMqwen3_eagle3:Eagle3Qwen3ForCausalLM
Qwen3DSparkModelqwen3_dspark:Qwen3DSparkForCausalLM
Qwen3NextMTPqwen3_next_mtp:Qwen3NextMTP
Qwen3_5MTPqwen3_5_mtp:Qwen3_5MTP
Qwen3_5MoeMTPqwen3_5_mtp:Qwen3_5MoeMTP
Step3p5MTPstep3p5_mtp:Step3p5MTP

Transformers backend

ArchitectureImplementation
Emu3ForConditionalGenerationtransformers:TransformersMultiModalForCausalLM
GPTBigCodeForCausalLMtransformers:TransformersForCausalLM
Olmo2ForCausalLMtransformers:TransformersForCausalLM
OlmoForCausalLMtransformers:TransformersForCausalLM
SmolLM3ForCausalLMtransformers:TransformersForCausalLM
Starcoder2ForCausalLMtransformers:TransformersForCausalLM
VaultGemmaForCausalLMtransformers:TransformersForCausalLM
VibeVoiceAsrForConditionalGenerationtransformers:TransformersMultiModalForCausalLM