unslothai--unsloth/studio/backend/utils/datasets/model_mappings.py

# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0

"""Model and template mappings for dataset processing.

Maps model names to their chat templates and response markers.
"""

TEMPLATE_TO_MODEL_MAPPER = {
    "phi-3.5": (
        "unsloth/Phi-3.5-mini-instruct-bnb-4bit",
        "unsloth/Phi-3.5-mini-instruct",
        "microsoft/Phi-3.5-mini-instruct",
    ),
    "phi-3": (
        "unsloth/Phi-3-mini-4k-instruct-bnb-4bit",
        "unsloth/Phi-3-mini-4k-instruct",
        "microsoft/Phi-3-mini-4k-instruct",
        "unsloth/Phi-3-medium-4k-instruct-bnb-4bit",
        "unsloth/Phi-3-medium-4k-instruct",
        "microsoft/Phi-3-medium-4k-instruct",
        "unsloth/Phi-3-mini-4k-instruct-v0-bnb-4bit",
        "unsloth/Phi-3-mini-4k-instruct-v0",
    ),
    "phi-4": (
        "unsloth/phi-4-unsloth-bnb-4bit",
        "unsloth/phi-4",
        "microsoft/phi-4",
        "unsloth/phi-4-bnb-4bit",
        "unsloth/phi-4-reasoning-unsloth-bnb-4bit",
        "unsloth/phi-4-reasoning",
        "microsoft/Phi-4-reasoning",
        "unsloth/phi-4-reasoning-bnb-4bit",
        "unsloth/phi-4-reasoning-plus-unsloth-bnb-4bit",
        "unsloth/phi-4-reasoning-plus",
        "microsoft/Phi-4-reasoning-plus",
        "unsloth/phi-4-reasoning-plus-bnb-4bit",
        "unsloth/phi-4-mini-reasoning-unsloth-bnb-4bit",
        "unsloth/phi-4-mini-reasoning",
        "microsoft/Phi-4-mini-reasoning",
        "unsloth/phi-4-mini-reasoning-bnb-4bit",
        "unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit",
        "unsloth/Phi-4-mini-instruct",
        "microsoft/Phi-4-mini-instruct",
        "unsloth/Phi-4-mini-instruct-bnb-4bit",
    ),
    "mistral": (
        "unsloth/mistral-7b-instruct-v0.1-bnb-4bit",
        "unsloth/mistral-7b-instruct-v0.1",
        "mistralai/Mistral-7B-Instruct-v0.1",
        "unsloth/mistral-7b-instruct-v0.2-bnb-4bit",
        "unsloth/mistral-7b-instruct-v0.2",
        "mistralai/Mistral-7B-Instruct-v0.2",
        "unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
        "unsloth/mistral-7b-instruct-v0.3",
        "mistralai/Mistral-7B-Instruct-v0.3",
        "unsloth/Mixtral-8x7B-Instruct-v0.1-unsloth-bnb-4bit",
        "unsloth/Mixtral-8x7B-Instruct-v0.1",
        "mistralai/Mixtral-8x7B-Instruct-v0.1",
        "unsloth/Mixtral-8x7B-Instruct-v0.1-bnb-4bit",
        "unsloth/Mistral-Nemo-Instruct-2407-bnb-4bit",
        "unsloth/Mistral-Nemo-Instruct-2407",
        "mistralai/Mistral-Nemo-Instruct-2407",
        "unsloth/Mistral-Large-Instruct-2407-bnb-4bit",
        "mistralai/Mistral-Large-Instruct-2407",
        "unsloth/Mistral-Small-Instruct-2409-bnb-4bit",
        "unsloth/Mistral-Small-Instruct-2409",
        "mistralai/Mistral-Small-Instruct-2409",
        "unsloth/Mistral-Small-24B-Instruct-2501-unsloth-bnb-4bit",
        "unsloth/Mistral-Small-24B-Instruct-2501",
        "mistralai/Mistral-Small-24B-Instruct-2501",
        "unsloth/Mistral-Small-24B-Instruct-2501-bnb-4bit",
        "unsloth/Mistral-Small-3.1-24B-Instruct-2503-unsloth-bnb-4bit",
        "unsloth/Mistral-Small-3.1-24B-Instruct-2503",
        "mistralai/Mistral-Small-3.1-24B-Instruct-2503",
        "unsloth/Mistral-Small-3.1-24B-Instruct-2503-bnb-4bit",
        "unsloth/Mistral-Small-3.2-24B-Instruct-2506-unsloth-bnb-4bit",
        "unsloth/Mistral-Small-3.2-24B-Instruct-2506",
        "mistralai/Mistral-Small-3.2-24B-Instruct-2506",
        "unsloth/Mistral-Small-3.2-24B-Instruct-2506-bnb-4bit",
    ),
    "llama": (
        "meta-llama/Llama-2-13b-chat-hf",
        "unsloth/llama-2-7b-chat-bnb-4bit",
        "unsloth/llama-2-7b-chat",
        "meta-llama/Llama-2-7b-chat-hf",
    ),
    "llama3": (
        "unsloth/llama-3-8b-Instruct-bnb-4bit",
        "unsloth/llama-3-8b-Instruct",
        "meta-llama/Meta-Llama-3-8B-Instruct",
        "unsloth/llama-3-70b-Instruct-bnb-4bit",
        "meta-llama/Meta-Llama-3-70B-Instruct",
    ),
    "llama-3.1": (
        "unsloth/Meta-Llama-3.1-8B-Instruct-unsloth-bnb-4bit",
        "unsloth/Meta-Llama-3.1-8B-Instruct",
        "meta-llama/Meta-Llama-3.1-8B-Instruct",
        "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit",
        "unsloth/Llama-3.1-8B-Instruct-unsloth-bnb-4bit",
        "unsloth/Llama-3.1-8B-Instruct",
        "meta-llama/Llama-3.1-8B-Instruct",
        "unsloth/Llama-3.1-8B-Instruct-bnb-4bit",
        "unsloth/Meta-Llama-3.1-405B-Instruct-bnb-4bit",
        "meta-llama/Meta-Llama-3.1-405B-Instruct",
        "unsloth/Meta-Llama-3.1-70B-Instruct-bnb-4bit",
        "unsloth/Meta-Llama-3.1-70B-Instruct",
        "meta-llama/Meta-Llama-3.1-70B-Instruct",
        "unsloth/Llama-3.1-Storm-8B-bnb-4bit",
        "unsloth/Llama-3.1-Storm-8B",
        "akjindal53244/Llama-3.1-Storm-8B",
        "unsloth/Hermes-3-Llama-3.1-8B-bnb-4bit",
        "unsloth/Hermes-3-Llama-3.1-8B",
        "NousResearch/Hermes-3-Llama-3.1-8B",
        "unsloth/Hermes-3-Llama-3.1-70B-bnb-4bit",
        "unsloth/Hermes-3-Llama-3.1-70B",
        "NousResearch/Hermes-3-Llama-3.1-70B",
        "unsloth/Hermes-3-Llama-3.1-405B-bnb-4bit",
        "NousResearch/Hermes-3-Llama-3.1-405B",
        "unsloth/Llama-3.1-Nemotron-70B-Instruct-bnb-4bit",
        "unsloth/Llama-3.1-Nemotron-70B-Instruct",
        "nvidia/Llama-3.1-Nemotron-70B-Instruct-HF",
        "unsloth/Llama-3.1-Tulu-3-8B-bnb-4bit",
        "unsloth/Llama-3.1-Tulu-3-8B",
        "allenai/Llama-3.1-Tulu-3-8B",
        "unsloth/Llama-3.1-Tulu-3-70B-bnb-4bit",
        "unsloth/Llama-3.1-Tulu-3-70B",
        "allenai/Llama-3.1-Tulu-3-70B",
    ),
    "llama-3.2": (
        "unsloth/Llama-3.2-1B-Instruct-unsloth-bnb-4bit",
        "unsloth/Llama-3.2-1B-Instruct",
        "meta-llama/Llama-3.2-1B-Instruct",
        "unsloth/Llama-3.2-1B-Instruct-bnb-4bit",
        "unsloth/Llama-3.2-3B-Instruct-unsloth-bnb-4bit",
        "unsloth/Llama-3.2-3B-Instruct",
        "meta-llama/Llama-3.2-3B-Instruct",
        "unsloth/Llama-3.2-3B-Instruct-bnb-4bit",
        "unsloth/Llama-3.2-11B-Vision-Instruct-unsloth-bnb-4bit",
        "unsloth/Llama-3.2-11B-Vision-Instruct",
        "meta-llama/Llama-3.2-11B-Vision-Instruct",
        "unsloth/Llama-3.2-11B-Vision-Instruct-bnb-4bit",
        "unsloth/Llama-3.2-90B-Vision-Instruct-bnb-4bit",
        "unsloth/Llama-3.2-90B-Vision-Instruct",
        "meta-llama/Llama-3.2-90B-Vision-Instruct",
    ),
    "llama-3.3": (
        "unsloth/Llama-3.3-70B-Instruct-bnb-4bit",
        "unsloth/Llama-3.3-70B-Instruct",
        "meta-llama/Llama-3.3-70B-Instruct",
    ),
    "gemma": (
        "unsloth/gemma-7b-it-bnb-4bit",
        "unsloth/gemma-7b-it",
        "google/gemma-7b-it",
        "google/gemma-2b-it",
        "unsloth/gemma-1.1-2b-it-bnb-4bit",
        "unsloth/gemma-1.1-2b-it",
        "google/gemma-1.1-2b-it",
        "unsloth/gemma-1.1-7b-it-bnb-4bit",
        "unsloth/gemma-1.1-7b-it",
        "google/gemma-1.1-7b-it",
    ),
    "gemma2": (
        "unsloth/gemma-2-9b-it-bnb-4bit",
        "unsloth/gemma-2-9b-it",
        "google/gemma-2-9b-it",
        "unsloth/gemma-2-27b-it-bnb-4bit",
        "unsloth/gemma-2-27b-it",
        "google/gemma-2-27b-it",
        "unsloth/gemma-2-2b-it-bnb-4bit",
        "unsloth/gemma-2-2b-it",
        "google/gemma-2-2b-it",
    ),
    "gemma-3": (
        "unsloth/gemma-3-1b-it-unsloth-bnb-4bit",
        "unsloth/gemma-3-1b-it",
        "google/gemma-3-1b-it",
        "unsloth/gemma-3-1b-it-bnb-4bit",
        "unsloth/gemma-3-4b-it-unsloth-bnb-4bit",
        "unsloth/gemma-3-4b-it",
        "google/gemma-3-4b-it",
        "unsloth/gemma-3-4b-it-bnb-4bit",
        "unsloth/gemma-3-12b-it-unsloth-bnb-4bit",
        "unsloth/gemma-3-12b-it",
        "google/gemma-3-12b-it",
        "unsloth/gemma-3-12b-it-bnb-4bit",
        "unsloth/gemma-3-27b-it-unsloth-bnb-4bit",
        "unsloth/gemma-3-27b-it",
        "google/gemma-3-27b-it",
        "unsloth/gemma-3-27b-it-bnb-4bit",
        "unsloth/gemma-3-270m-it-unsloth-bnb-4bit",
        "unsloth/gemma-3-270m-it",
        "google/gemma-3-270m-it",
        "unsloth/gemma-3-270m-it-bnb-4bit",
        "unsloth/gemma-3-270m-unsloth-bnb-4bit",
        "unsloth/medgemma-4b-it-unsloth-bnb-4bit",
        "unsloth/medgemma-4b-it",
        "google/medgemma-4b-it",
        "unsloth/medgemma-4b-it-bnb-4bit",
        "unsloth/medgemma-27b-text-it-unsloth-bnb-4bit",
        "unsloth/medgemma-27b-text-it",
        "google/medgemma-27b-text-it",
        "unsloth/medgemma-27b-text-it-bnb-4bit",
    ),
    "gemma3n": (
        "unsloth/gemma-3n-E4B-it-unsloth-bnb-4bit",
        "unsloth/gemma-3n-E4B-it",
        "google/gemma-3n-E4B-it",
        "unsloth/gemma-3n-E4B-it-unsloth-bnb-4bit",
        "unsloth/gemma-3n-E2B-it-unsloth-bnb-4bit",
        "unsloth/gemma-3n-E2B-it",
        "google/gemma-3n-E2B-it",
        "unsloth/gemma-3n-E2B-it-unsloth-bnb-4bit",
    ),
    "gemma-4": (
        "unsloth/gemma-4-E2B-it",
        "google/gemma-4-E2B-it",
        "unsloth/gemma-4-E4B-it",
        "google/gemma-4-E4B-it",
        "unsloth/gemma-4-E2B-it-unsloth-bnb-4bit",
        "unsloth/gemma-4-E4B-it-unsloth-bnb-4bit",
    ),
    "gemma-4-thinking": (
        "unsloth/gemma-4-26B-A4B-it",
        "google/gemma-4-26B-A4B-it",
        "unsloth/gemma-4-31B-it",
        "unsloth/gemma-4-31B-it-unsloth-bnb-4bit",
        "google/gemma-4-31B-it",
    ),
    "qwen2.5": (
        "unsloth/Qwen2.5-0.5B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-0.5B-Instruct",
        "Qwen/Qwen2.5-0.5B-Instruct",
        "unsloth/Qwen2.5-0.5B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-1.5B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-1.5B-Instruct",
        "Qwen/Qwen2.5-1.5B-Instruct",
        "unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-3B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-3B-Instruct",
        "Qwen/Qwen2.5-3B-Instruct",
        "unsloth/Qwen2.5-3B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-7B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-7B-Instruct",
        "Qwen/Qwen2.5-7B-Instruct",
        "unsloth/Qwen2.5-7B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-14B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-14B-Instruct",
        "Qwen/Qwen2.5-14B-Instruct",
        "unsloth/Qwen2.5-14B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-32B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-32B-Instruct",
        "Qwen/Qwen2.5-32B-Instruct",
        "unsloth/Qwen2.5-72B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-72B-Instruct",
        "Qwen/Qwen2.5-72B-Instruct",
        "unsloth/Qwen2.5-0.5B-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-Math-1.5B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Math-1.5B-Instruct",
        "Qwen/Qwen2.5-Math-1.5B-Instruct",
        "unsloth/Qwen2.5-Math-7B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Math-7B-Instruct",
        "Qwen/Qwen2.5-Math-7B-Instruct",
        "unsloth/Qwen2.5-Math-72B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Math-72B-Instruct",
        "Qwen/Qwen2.5-Math-72B-Instruct",
        "unsloth/Qwen2.5-Coder-0.5B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-0.5B-Instruct",
        "Qwen/Qwen2.5-Coder-0.5B-Instruct",
        "unsloth/Qwen2.5-Coder-1.5B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-1.5B-Instruct",
        "Qwen/Qwen2.5-Coder-1.5B-Instruct",
        "unsloth/Qwen2.5-Coder-3B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-3B-Instruct",
        "Qwen/Qwen2.5-Coder-3B-Instruct",
        "unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-7B-Instruct",
        "Qwen/Qwen2.5-Coder-7B-Instruct",
        "unsloth/Qwen2.5-Coder-14B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-14B-Instruct",
        "Qwen/Qwen2.5-Coder-14B-Instruct",
        "unsloth/Qwen2.5-Coder-32B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-Coder-32B-Instruct",
        "Qwen/Qwen2.5-Coder-32B-Instruct",
        "unsloth/Qwen2.5-VL-3B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-VL-3B-Instruct",
        "Qwen/Qwen2.5-VL-3B-Instruct",
        "unsloth/Qwen2.5-VL-3B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-VL-7B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-VL-7B-Instruct",
        "Qwen/Qwen2.5-VL-7B-Instruct",
        "unsloth/Qwen2.5-VL-7B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-VL-32B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-VL-32B-Instruct",
        "Qwen/Qwen2.5-VL-32B-Instruct",
        "unsloth/Qwen2.5-VL-32B-Instruct-bnb-4bit",
        "unsloth/Qwen2.5-VL-72B-Instruct-unsloth-bnb-4bit",
        "unsloth/Qwen2.5-VL-72B-Instruct",
        "Qwen/Qwen2.5-VL-72B-Instruct",
        "unsloth/Qwen2.5-VL-72B-Instruct-bnb-4bit",
        "unsloth/OpenThinker-7B-unsloth-bnb-4bit",
        "unsloth/OpenThinker-7B",
        "open-thoughts/OpenThinker-7B",
        "unsloth/OpenThinker-7B-bnb-4bit",
    ),
    "qwen3": (
        "unsloth/Qwen3-0.6B-unsloth-bnb-4bit",
        "unsloth/Qwen3-0.6B",
        "Qwen/Qwen3-0.6B",
        "unsloth/Qwen3-0.6B-bnb-4bit",
        "unsloth/Qwen3-1.7B-unsloth-bnb-4bit",
        "unsloth/Qwen3-1.7B",
        "Qwen/Qwen3-1.7B",
        "unsloth/Qwen3-1.7B-bnb-4bit",
        "unsloth/Qwen3-4B-unsloth-bnb-4bit",
        "unsloth/Qwen3-4B",
        "Qwen/Qwen3-4B",
        "unsloth/Qwen3-4B-bnb-4bit",
        "unsloth/Qwen3-8B-unsloth-bnb-4bit",
        "unsloth/Qwen3-8B",
        "Qwen/Qwen3-8B",
        "unsloth/Qwen3-8B-bnb-4bit",
        "unsloth/Qwen3-14B-unsloth-bnb-4bit",
        "unsloth/Qwen3-14B",
        "Qwen/Qwen3-14B",
        "unsloth/Qwen3-14B-bnb-4bit",
        "unsloth/Qwen3-32B-unsloth-bnb-4bit",
        "unsloth/Qwen3-32B",
        "Qwen/Qwen3-32B",
        "unsloth/Qwen3-32B-bnb-4bit",
        "unsloth/Qwen3-30B-A3B-unsloth-bnb-4bit",
        "unsloth/Qwen3-30B-A3B",
        "Qwen/Qwen3-30B-A3B",
        "unsloth/Qwen3-30B-A3B-bnb-4bit",
    ),
    "qwen3-instruct": (
        "unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit",
        "unsloth/Qwen3-4B-Instruct-2507",
        "Qwen/Qwen3-4B-Instruct-2507",
        "unsloth/Qwen3-4B-Instruct-2507-bnb-4bit",
        "unsloth/Qwen3-30B-A3B-Instruct-2507",
        "Qwen/Qwen3-30B-A3B-Instruct-2507",
        "unsloth/Qwen3-Coder-30B-A3B-Instruct",
        "Qwen/Qwen3-Coder-30B-A3B-Instruct",
        "unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit",
        "unsloth/Qwen3-4B-Instruct-2507",
        "Qwen/Qwen3-4B-Instruct-2507",
        "unsloth/Qwen3-4B-Instruct-2507-bnb-4bit",
    ),
    "qwen3-thinking": (
        "unsloth/QwQ-32B-Preview-bnb-4bit",
        "unsloth/QwQ-32B-Preview",
        "Qwen/QwQ-32B-Preview",
        "unsloth/QwQ-32B-unsloth-bnb-4bit",
        "unsloth/QwQ-32B",
        "Qwen/QwQ-32B",
        "unsloth/QwQ-32B-bnb-4bit",
        "unsloth/Qwen3-4B-Thinking-2507-unsloth-bnb-4bit",
        "unsloth/Qwen3-4B-Thinking-2507",
        "Qwen/Qwen3-4B-Thinking-2507",
        "unsloth/Qwen3-4B-Thinking-2507-bnb-4bit",
        "unsloth/Qwen3-30B-A3B-Thinking-2507",
        "Qwen/Qwen3-30B-A3B-Thinking-2507",
        "Qwen/Qwen3.6-35B-A3B",
        "unsloth/Qwen3.6-35B-A3B",
        "Qwen/Qwen3.6-27B",
        "unsloth/Qwen3.6-27B",
    ),
    "qwen3.5": (
        "unsloth/Qwen3.5-0.8B",
        "unsloth/Qwen3.5-2B",
        "unsloth/Qwen3.5-4B",
        "unsloth/Qwen3.5-9B",
        "unsloth/Qwen3.5-27B",
        "unsloth/Qwen3.5-35B-A3B",
    ),
    "zephyr": (
        "unsloth/zephyr-sft-bnb-4bit",
        "unsloth/zephyr-sft",
        "HuggingFaceH4/mistral-7b-sft-beta",
    ),
    "chatml": (
        "unsloth/yi-6b-bnb-4bit",
        "unsloth/yi-6b",
        "01-ai/Yi-6B",
        "unsloth/Hermes-2-Pro-Mistral-7B-bnb-4bit",
        "unsloth/Hermes-2-Pro-Mistral-7B",
        "NousResearch/Hermes-2-Pro-Mistral-7B",
        "unsloth/OpenHermes-2.5-Mistral-7B-bnb-4bit",
        "unsloth/OpenHermes-2.5-Mistral-7B",
        "teknium/OpenHermes-2.5-Mistral-7B",
    ),
    "gpt-oss": (
        "unsloth/gpt-oss-20b-unsloth-bnb-4bit",
        "unsloth/gpt-oss-20b",
        "openai/gpt-oss-20b",
        "unsloth/gpt-oss-20b-unsloth-bnb-4bit",
        "unsloth/gpt-oss-120b-unsloth-bnb-4bit",
        "unsloth/gpt-oss-120b",
        "openai/gpt-oss-120b",
        "unsloth/gpt-oss-120b-unsloth-bnb-4bit",
    ),
    "starling": (
        "unsloth/Starling-LM-7B-beta-bnb-4bit",
        "unsloth/Starling-LM-7B-beta",
        "Nexusflow/Starling-LM-7B-beta",
    ),
    "yi-chat": (
        "unsloth/yi-34b-chat-bnb-4bit",
        "01-ai/Yi-6B-Chat",
        "01-ai/Yi-34B-Chat",
    ),
    "glm": (
        "unsloth/GLM-4.7-Flash-unsloth-bnb-4bit",
        "unsloth/GLM-4.7-Flash",
        "THUDM/GLM-4.7-Flash",
        "unsloth/GLM-4.7-Flash-bnb-4bit",
    ),
    "lfm-2": (
        "unsloth/LFM2-1.2B",
        "LiquidAI/LFM2-1.2B",
        "unsloth/LFM2-1.2B-unsloth-bnb-4bit",
    ),
    "lfm-2.5": (
        "unsloth/LFM2.5-1.2B-Instruct",
        "LiquidAI/LFM2.5-1.2B-Instruct",
    ),
}

MODEL_TO_TEMPLATE_MAPPER = {}

for key, values in TEMPLATE_TO_MODEL_MAPPER.items():
    for value in values:
        MODEL_TO_TEMPLATE_MAPPER[value] = key

    # Also map lowercased names.
    lowered_key = key.lower()
    for value in values:
        MODEL_TO_TEMPLATE_MAPPER[value.lower()] = lowered_key


def is_gpt_oss_model_name(name: str) -> bool:
    """Name-based check for gpt-oss / harmony models.

    Used by the in-process backend and the parent orchestrator to detect
    harmony models without an IPC round-trip.
    """
    name = (name or "").lower()
    if not name:
        return False
    try:
        if MODEL_TO_TEMPLATE_MAPPER.get(name) == "gpt-oss":
            return True
        for key, tmpl in MODEL_TO_TEMPLATE_MAPPER.items():
            if tmpl == "gpt-oss" and (key in name or name in key):
                return True
    except Exception:
        pass
    return "gpt-oss" in name


TEMPLATE_TO_RESPONSES_MAPPER = {
    "gemma-4-thinking": {
        "instruction": "<|turn>user\n",
        "response": "<|turn>model\n",
    },
    "gemma-4": {
        "instruction": "<|turn>user\n",
        "response": "<|turn>model\n",
    },
    "gemma-3": {
        "instruction": "<start_of_turn>user\n",
        "response": "<start_of_turn>model\n",
    },
    "gemma3n": {
        "instruction": "<start_of_turn>user\n",
        "response": "<start_of_turn>model\n",
    },
    "qwen3.5": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    "qwen3-instruct": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    # No "<think>" suffix: Qwen3-Thinking-2507 strips it from non-final turns
    # and QwQ renders none, so a marker holding it masks those responses.
    "qwen3-thinking": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    "qwen3": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    "qwen2.5": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    "llama-3.2": {
        "instruction": "<|start_header_id|>user<|end_header_id|>\n\n",
        "response": "<|start_header_id|>assistant<|end_header_id|>\n\n",
    },
    "llama-3.3": {
        "instruction": "<|start_header_id|>user<|end_header_id|>\n\n",
        "response": "<|start_header_id|>assistant<|end_header_id|>\n\n",
    },
    "llama-3.1": {
        "instruction": "<|start_header_id|>user<|end_header_id|>\n\n",
        "response": "<|start_header_id|>assistant<|end_header_id|>\n\n",
    },
    "llama3": {
        "instruction": "<|start_header_id|>user<|end_header_id|>\n\n",
        "response": "<|start_header_id|>assistant<|end_header_id|>\n\n",
    },
    "phi-3": {
        "instruction": "<|user|>\n",
        "response": "<|assistant|>\n",
    },
    "phi-3.5": {
        "instruction": "<|user|>\n",
        "response": "<|assistant|>\n",
    },
    "phi-4": {
        "instruction": "<|im_start|>user<|im_sep|>",
        "response": "<|im_start|>assistant<|im_sep|>",
    },
    # No surrounding spaces: in Mistral v0.3 they fold into neighbouring text
    # tokens ("[INST]"/"[/INST]" are single special tokens), so padded strings
    # never match and everything masks. Same for Llama-2's SentencePiece.
    "mistral": {
        "instruction": "[INST]",
        "response": "[/INST]",
    },
    "llama": {
        # <s>-anchored: llama-2 tokenizes [INST] after <s> as bare "[" on
        # transformers 5.x (standalone gives space-prefixed "▁["), so an
        # unanchored marker misses every turn boundary there.
        "instruction": "<s>[INST]",
        "response": "[/INST]",
    },
    "chatml": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    # Leading newline required: Zephyr's role tags are plain text, and
    # SentencePiece tokenizes "<|assistant|>" differently at text start than
    # after "</s>\n". Without the "\n" anchor the markers never match real
    # turns, so every assistant token masks.
    "zephyr": {
        "instruction": "\n<|user|>\n",
        "response": "\n<|assistant|>\n",
    },
    "unsloth": {
        "instruction": ">>> User:",
        "response": ">>> Assistant:",
    },
    "vicuna": {
        "instruction": "USER:",
        "response": "ASSISTANT:",
    },
    "alpaca": {
        "instruction": "### Instruction:\n",
        "response": "### Response:\n",
    },
    "gemma": {
        "instruction": "<start_of_turn>user\n",
        "response": "<start_of_turn>model\n",
    },
    "gemma2": {
        "instruction": "<start_of_turn>user\n",
        "response": "<start_of_turn>model\n",
    },
    "gpt-oss": {
        "instruction": "<|start|>user<|message|>",
        "response": "<|start|>assistant<|channel|>final<|message|>",
    },
    "lfm-2": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    "lfm-2.5": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    # No trailing space: SentencePiece folds it into the next content token
    # ("▁Hello"), so the padded marker never matches and masks everything.
    "starling": {
        "instruction": "GPT4 Correct User:",
        "response": "GPT4 Correct Assistant:",
    },
    "yi-chat": {
        "instruction": "<|im_start|>user\n",
        "response": "<|im_start|>assistant\n",
    },
    # "[gMASK]<sop>" appears once at text start, so a marker holding it matches
    # no later user turn; "<think>" is scaffolding GLM-4.x renders as a lone
    # "</think>" on non-final turns, so "<|assistant|><think>" never matches.
    "glm": {
        "instruction": "<|user|>",
        "response": "<|assistant|>",
    },
}