Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Load six older models in 4bit instead of a missing repo (#13941)
Nilayaf0e2e1d04f9
diff --git a/tests/test_bad_mappings_redirect.py b/tests/test_bad_mappings_redirect.pyindex fb8fd803b6a..d6f92dcf989 100644--- a/tests/test_bad_mappings_redirect.py+++ b/tests/test_bad_mappings_redirect.py@@ -3,6 +3,8 @@import astimport os+import pytest+_MODELS = os.path.join(os.path.dirname(__file__), os.pardir, "unsloth", "models")@@ -47,3 +49,20 @@ def test_bad_mapping_redirects_keep_canonical_repo_case():get_model_name, _ = _load_get_model_name()assert get_model_name("Qwen/Qwen3-32B", load_in_4bit = True) == "unsloth/Qwen3-32B-bnb-4bit"assert get_model_name("Qwen/Qwen3-30B-A3B", load_in_4bit = True) == "unsloth/Qwen3-30B-A3B"+++ "name, expected",+ [+ ("upstage/SOLAR-10.7B-v1.0", "upstage/SOLAR-10.7B-v1.0"),+ ("allenai/Llama-3.1-Tulu-3-8B", "unsloth/Llama-3.1-Tulu-3-8B"),+ ("unsloth/Llama-3.1-Tulu-3-8B", "unsloth/Llama-3.1-Tulu-3-8B"),+ ("allenai/Llama-3.1-Tulu-3-70B", "allenai/Llama-3.1-Tulu-3-70B"),+ ("NousResearch/Hermes-3-Llama-3.1-405B", "NousResearch/Hermes-3-Llama-3.1-405B"),+ ("open-thoughts/OpenThinker-7B", "open-thoughts/OpenThinker-7B"),+ ("mistralai/Pixtral-12B-Base-2409", "unsloth/Pixtral-12B-Base-2409-bnb-4bit"),+ ],+)+def test_4bit_load_avoids_unpublished_bnb_repos(name, expected):+ get_model_name, _ = _load_get_model_name()+ assert get_model_name(name, load_in_4bit = True) == expecteddiff --git a/unsloth/models/loader_utils.py b/unsloth/models/loader_utils.pyindex 2e3dc2e21ee..e0947574c90 100644--- a/unsloth/models/loader_utils.py+++ b/unsloth/models/loader_utils.py@@ -59,6 +59,12 @@ BAD_MAPPINGS = {"unsloth/Qwen3-30B-A3B-bnb-4bit".lower(): "unsloth/Qwen3-30B-A3B", # We rather do it on the fly"unsloth/Qwen3-30B-A3B-Base-unsloth-bnb-4bit".lower(): "unsloth/Qwen3-30B-A3B-Base", # HF loads MoEs too slowly"unsloth/Qwen3-30B-A3B-Base-bnb-4bit".lower(): "unsloth/Qwen3-30B-A3B-Base", # We rather do it on the fly+ "unsloth/solar-10.7b-bnb-4bit".lower(): "upstage/SOLAR-10.7B-v1.0",+ "unsloth/Hermes-3-Llama-3.1-405B-bnb-4bit".lower(): "NousResearch/Hermes-3-Llama-3.1-405B",+ "unsloth/Pixtral-12B-2409-Base-bnb-4bit".lower(): "unsloth/Pixtral-12B-Base-2409-bnb-4bit",+ "unsloth/Llama-3.1-Tulu-3-8B-bnb-4bit".lower(): "unsloth/Llama-3.1-Tulu-3-8B",+ "unsloth/Llama-3.1-Tulu-3-70B-bnb-4bit".lower(): "allenai/Llama-3.1-Tulu-3-70B",+ "unsloth/OpenThinker-7B-unsloth-bnb-4bit".lower(): "open-thoughts/OpenThinker-7B",}