Skip to content

Enums

Member
Architecture.Mistral
Architecture.Gemma
Architecture.Mixtral
Architecture.Llama
Architecture.Phi2
Architecture.Phi3
Architecture.Qwen2
Architecture.Gemma2
Architecture.Starcoder2
Architecture.Phi3_5MoE
Architecture.DeepseekV2
Architecture.DeepseekV3
Architecture.Qwen3
Architecture.GLM4
Architecture.GLM4Moe
Architecture.GLM4MoeLite
Architecture.Qwen3Moe
Architecture.Flint
Architecture.SmolLm3
Architecture.GraniteMoeHybrid
Architecture.GptOss
Architecture.HunYuanDenseV1
Architecture.HunYuanMoEV1
Architecture.Qwen3Next
Architecture.Lfm2
Architecture.Lfm3
Architecture.Lfm2Moe
Member
EmbeddingArchitecture.EmbeddingGemma
EmbeddingArchitecture.Qwen3Embedding
Member
MultimodalArchitecture.Phi3V
MultimodalArchitecture.Idefics2
MultimodalArchitecture.LLaVANext
MultimodalArchitecture.LLaVA
MultimodalArchitecture.Lfm2Vl
MultimodalArchitecture.Lfm3Vl
MultimodalArchitecture.Lfm3Audio
MultimodalArchitecture.Lfm3Omni
MultimodalArchitecture.Lfm3D1
MultimodalArchitecture.VLlama
MultimodalArchitecture.Qwen2VL
MultimodalArchitecture.Idefics3
MultimodalArchitecture.MiniCpmO
MultimodalArchitecture.Phi4MM
MultimodalArchitecture.Qwen2_5VL
MultimodalArchitecture.Gemma3
MultimodalArchitecture.Mistral3
MultimodalArchitecture.Llama4
MultimodalArchitecture.Gemma3n
MultimodalArchitecture.Qwen3VL
MultimodalArchitecture.Qwen3VLMoE
MultimodalArchitecture.Qwen3_5
MultimodalArchitecture.Qwen3_5Moe
MultimodalArchitecture.Qwen4Exp
MultimodalArchitecture.Voxtral
MultimodalArchitecture.Gemma4
MultimodalArchitecture.MuseGlimmer
MultimodalArchitecture.DiffusionGemma
Member
DiffusionArchitecture.Flux
DiffusionArchitecture.FluxOffloaded
Member
SpeechLoaderType.Dia
Member Description
ModelDType.Auto Pick a type the device supports.
ModelDType.BF16
ModelDType.F16
ModelDType.F32
Member Description
IsqOrganization.Default Quantize every layer that in-situ quantization can quantize.
IsqOrganization.MoQE Quantize only the MoE expert weights and leave the other layers as they are.
Member
ImageGenerationResponseFormat.Url
ImageGenerationResponseFormat.B64Json
Member Description
ToolChoice.NoTools The model can’t call tools.
ToolChoice.Auto The model decides whether to call a tool.
Member
SearchContextSize.Low
SearchContextSize.Medium
SearchContextSize.High
Member
AgentPermission.Auto
AgentPermission.Ask
AgentPermission.Deny
Member
CodeExecutionPermission.Auto
CodeExecutionPermission.Ask
CodeExecutionPermission.Deny
Member Description
NetworkMode.NoNetwork No network access. On Linux, opening a socket fails.
NetworkMode.Loopback Only localhost is reachable.
NetworkMode.Full No network restrictions.
Member
AgentToolSource.BuiltIn
AgentToolSource.User
AgentToolSource.Mcp
AgentToolSource.External
Member
AgentToolKind.CodeExecution
AgentToolKind.WebSearch
AgentToolKind.Shell
AgentToolKind.File
AgentToolKind.Custom
AgentToolKind.External
Member
AgentToolApprovalDecisionKind.Approve
AgentToolApprovalDecisionKind.Deny
Member Description
PagedCacheType.Auto Store the KV cache in the model’s activation type.
PagedCacheType.F8E4M3 Store the KV cache in FP8 (E4M3), half the memory of BF16. Needs F16, BF16, or F32 activations.

Generated from mistralrs-pyo3/mistralrs.pyi.