Enum LM.TemplateFormat

Namespace

LMKit.Model

Assembly

LM-Kit.NET.dll

Defines the possible formats for model chat templates, indicating specific structures and encoding standards used for chatbot interactions.

public enum LM.TemplateFormat

Fields

Unknown = -1

The format is not recognized. This value is used as a default when the format cannot be determined.

LLama2 = 0

Format associated with LLama2 models, typically involving certain structural or functional characteristics unique to this version.

Phi2 = 1

Format designed for Phi2 models, supporting features and specifications tailored to these models.

Phi3 = 2

Format used by Phi3 models, optimized for advanced capabilities provided by this iteration.

ChatML = 3

ChatML format, which may use XML-based markup language specifically designed for chatbot scripts.

Zephyr = 4

Zephyr format, potentially signifying a model or template with specific optimizations or enhancements.

Vicuna = 5

Vicuna format, indicating a newer or possibly experimental format tailored for specific functionalities.

LLama3 = 6

Represents the format used by LLama3 models, indicating evolution or updates from earlier LLama versions.

CommandR = 7

CommandR format, which might denote a command-driven or interaction-specific scripting format for chatbots.

Gemma = 8

Gemma format, which may denote a format developed for enhancing user experience or specific interactive capabilities in chatbots.

Alpaca = 9

Format associated with Alpaca models, indicating specific structural or functional characteristics unique to this version.

DeepSeek = 10

DeepSeek template format.

Granite = 11

IBM Granite format.

Phi4 = 12

Phi template format v4.

MistralV7 = 13

Prompt format used by some Mistral models (V7 variant).

MistralV1 = 14

Mistral template format (V1 variant).

MistralV3Tekken = 15

Mistral template format (V3 Tekken variant).

MistralV3 = 16

Mistral template format (V3 variant).

LLama2Sys = 17

LLama2 template format with system message support (using <<SYS>>).

LLama2SysBOS = 18

LLama2 template format with BOS token inserted inside the history.

LLama2SysStrip = 19

LLama2 template format with stripped message content.

Falcon3 = 20

Falcon3 template format.

GLMEDGE = 21

GLMEDGE template format.

Monarch = 22

Monarch template format.

Orion = 23

Orion template format for OrionStarAI/Orion-14B-Chat.

OpenChat = 24

OpenChat template format.

VicunaOrca = 25

Vicuna-Orca template format, a variant of Vicuna with additional system instructions.

ChatGML3 = 26

ChatGML format (version 3).

ChatGML4 = 27

ChatGML format (version 4).

MiniCPM = 28

MiniCPM template format, identified by localized tokens.

DeepSeek2 = 29

DeepSeek template format v2.

DeepSeek3 = 30

DeepSeek template format v3.

Exaone3 = 31

Exaone template format.

RWKVWorld = 32

RWKV World template format.

GigaChat = 33

GigaChat template format.

Megrez = 34

Megrez template format.

Yandex = 35

Yandex template format, used by Yandex-developed chatbot models.

Bailing = 36

Bailing template format, signifying specialized structure for Bailing models.

Llama4 = 37

Format representing LLama4 models, indicating the next generation of LLama templates.

SmolVLM = 38

SmolVLM template format, optimized for very lightweight VLM-based chat experiences.

OpenAiMOE = 39

OpenAI MOE format.

Nemotron3 = 40

Nemotron3 template format.

PaddleOCR = 41

PaddleOCR VL template format, used by PaddlePaddle vision-language OCR models.

ChatGLM3 = 42

ChatGLM3 template format, used by ChatGLM3-6B and similar models from Zhipu AI.

ChatGLM4 = 43

ChatGLM4 template format, used by GLM-4 series models from Zhipu AI.

Gemma4 = 44

Gemma4 template format, used by Gemma 4 models with turn-based markers and native tool calling.

MistralV7Tekken = 45

Mistral V7 Tekken variant template format.

DeepSeekOCR = 46

DeepSeek OCR template format, raw content with no role formatting.

Exaone4 = 47

Exaone 4 template format with bracketed role markers and tool support.

ExaoneMOE = 48

Exaone MoE template format with angle-bracket role markers and tool declarations.

Granite4 = 49

IBM Granite 4.0 template format with start/end role markers and tool call support.

BailingThink = 50

Bailing template format with thinking/reasoning support.

Bailing2 = 51

Bailing v2 template format with role_end markers.

Dots1 = 52

Dots1 template format with system/userprompt/response markers.

HunyuanMOE = 53

Hunyuan MoE template format with extra token markers.

HunyuanDense = 54

Hunyuan Dense template format with fullwidth Unicode role markers.

KimiK2 = 55

Kimi K2 template format with im_system/im_user/im_assistant markers.

SeedOSS = 56

Seed OSS template format with seed:bos/seed:eos markers.

Grok2 = 57

Grok 2 template format with plain-text role labels and separator tokens.

PanguEmbed = 58

Pangu Embed template format with Chinese role markers and unused tokens.

SolarOpen = 59

Solar Open template format with begin/content/end markers.