More Qwen3-VL

unsloth/models/_utils.py

@@ -357,6 +357,14 @@ try:
 except:
     pass
 
+# Skipping import of cpp extensions due to incompatible torch version
+try:
+    from torchao import logger as torchao_logger
+    torchao_logger.addFilter(HideLoggingMessage("Skipping import"))
+    del torchao_logger
+except:
+    pass
+
 # Errors out on
 # Some weights of Gemma3nForConditionalGeneration were not initialized from the model checkpoint
 from transformers.modeling_utils import logger as transformers_logger

unsloth/models/llama.py

@@ -1208,7 +1208,8 @@ def CausalLM_fast_forward(fast_forward_inference):
                 # RuntimeError: Triton Error [HIP]: Code: 1, Messsage: invalid argument
                 RETURN_LOGITS = False
             elif bsz*q_len <= 1024:
-                RETURN_LOGITS = True
+                # Uses 800MB more VRAM it seems than fused CE Loss
+                RETURN_LOGITS = False
 
             if not RETURN_LOGITS and labels is not None:
                 n_items = kwargs.get("num_items_in_batch", None)

unsloth/models/mapper.py

@@ -1032,6 +1032,26 @@ __INT_TO_FLOAT_MAPPER = \
         "Qwen/Qwen3-VL-8B-Instruct",
         "unsloth/Qwen3-VL-8B-Instruct-bnb-4bit",
     ),
+    "unsloth/Qwen3-VL-2B-Thinking-unsloth-bnb-4bit" : (
+        "unsloth/Qwen3-VL-2B-Thinking",
+        "Qwen/Qwen3-VL-2B-Thinking",
+        "unsloth/Qwen3-VL-2B-Thinking-bnb-4bit",
+    ),
+    "unsloth/Qwen3-VL-2B-Thinking-unsloth-bnb-4bit" : (
+        "unsloth/Qwen3-VL-2B-Thinking",
+        "Qwen/Qwen3-VL-2B-Thinking",
+        "unsloth/Qwen3-VL-2B-Thinking-bnb-4bit",
+    ),
+    "unsloth/Qwen3-VL-32B-Instruct-unsloth-bnb-4bit" : (
+        "unsloth/Qwen3-VL-32B-Instruct",
+        "Qwen/Qwen3-VL-32B-Instruct",
+        "unsloth/Qwen3-VL-32B-Instruct-bnb-4bit",
+    ),
+    "unsloth/Qwen3-VL-32B-Instruct-unsloth-bnb-4bit" : (
+        "unsloth/Qwen3-VL-32B-Instruct",
+        "Qwen/Qwen3-VL-32B-Instruct",
+        "unsloth/Qwen3-VL-32B-Instruct-bnb-4bit",
+    ),
 }
 
 INT_TO_FLOAT_MAPPER  = {}