Bug Fixes

pyproject.toml

@@ -39,7 +39,7 @@ triton = [
     "triton @ https://github.com/woct0rdho/triton-windows/releases/download/v3.1.0-windows.post5/triton-3.1.0-cp312-cp312-win_amd64.whl ; python_version=='3.12' and platform_system == 'Windows'",
 ]
 huggingface = [
-    "unsloth_zoo>=2025.2.6",
+    "unsloth_zoo>=2025.2.7",
     "packaging",
     "tyro",
     "transformers>=4.46.1,!=4.47.0",
@@ -50,7 +50,7 @@ huggingface = [
     "wheel>=0.42.0",
     "numpy",
     "accelerate>=0.34.1",
-    "trl>=0.7.9,!=0.9.0,!=0.9.1,!=0.9.2,!=0.9.3",
+    "trl>=0.7.9,!=0.9.0,!=0.9.1,!=0.9.2,!=0.9.3,!=0.15.0",
     "peft>=0.7.1,!=0.11.0",
     "protobuf<4.0.0",
     "huggingface_hub",
@@ -348,7 +348,7 @@ colab-ampere-torch220 = [
     "flash-attn>=2.6.3",
 ]
 colab-new = [
-    "unsloth_zoo>=2025.2.6",
+    "unsloth_zoo>=2025.2.7",
     "packaging",
     "tyro",
     "transformers>=4.46.1,!=4.47.0",
@@ -366,7 +366,7 @@ colab-new = [
 ]
 colab-no-deps = [
     "accelerate>=0.34.1",
-    "trl>=0.7.9,!=0.9.0,!=0.9.1,!=0.9.2,!=0.9.3",
+    "trl>=0.7.9,!=0.9.0,!=0.9.1,!=0.9.2,!=0.9.3,!=0.15.0",
     "peft>=0.7.1",
     "xformers",
     "bitsandbytes>=0.46.1",

unsloth/models/_utils.py

@@ -143,6 +143,11 @@ def filter(self, x): return not (self.text in x.getMessage())
 transformers_training_args_logger.addFilter(HideLoggingMessage("torch.distributed"))
 del transformers_training_args_logger
 
+# No label_names provided for model class
+from transformers.trainer import logger as transformers_trainer_logger
+transformers_trainer_logger.addFilter(HideLoggingMessage("No label_names"))
+del transformers_trainer_logger
+
 # Using the default loss: `ForCausalLMLoss`.
 try:
     from transformers.modeling_utils import logger as transformers_modeling_utils_logger

unsloth/models/llama.py

@@ -725,8 +725,9 @@ def LlamaModel_fast_forward(
             past_key_values_length,
             sliding_window = getattr(self.config, "sliding_window", None),
         )
-        if attention_mask is not None:
-            attention_mask = attention_mask.to(torch.bool)
+        # Must NOT convert to bool - weirdly this causes stuff to error out!
+        # if attention_mask is not None:
+        #     attention_mask = attention_mask.to(torch.bool)
     pass
 
     hidden_states = inputs_embeds

unsloth/models/rl_replacements.py

@@ -164,10 +164,11 @@ def grpo_trainer__prepare_inputs(function_name, function):
 # Remove _move_model_to_vllm
 def grpo_trainer__move_model_to_vllm(function_name, function):
     if  function_name != "_move_model_to_vllm": return function
+
+    def _move_model_to_vllm(self, *args, **kwargs): return None
 
-    # .*? matches first match. .+? matches final match.
-    replacement = "def _move_model_to_vllm(self, *args, **kwargs): return None\n"
-    return " "*function.find("def") + replacement
+    function = inspect.getsource(_move_model_to_vllm)
+    return function
 pass
 RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer__move_model_to_vllm)