diff --git a/unsloth/__init__.py b/unsloth/__init__.py
index abee9c9e04..458c2696bc 100644
--- a/unsloth/__init__.py
+++ b/unsloth/__init__.py
@@ -62,9 +62,13 @@ os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
 
 try:
     import torch
-except:
-    raise ImportError("Pytorch is not installed. Go to https://pytorch.org/.\n"\
-                      "We have some installation instructions on our Github page.")
+except ModuleNotFoundError:
+    raise ImportError(
+        "Unsloth: Pytorch is not installed. Go to https://pytorch.org/.\n"\
+        "We have some installation instructions on our Github page."
+    )
+except Exception as exception:
+    raise exception
 pass
 
 # Hugging Face Hub faster downloads (only enable during Colab and Kaggle sessions)
diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index 2214ff80d4..bf5216b228 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "2024.10.5"
+__version__ = "2024.10.6"
 
 __all__ = [
     "prepare_model_for_kbit_training",
@@ -1172,10 +1172,10 @@ pass
 
 def patch_gradient_accumulation_fix(Trainer):
     # Fixes gradient accumulation 
+    import inspect
     if hasattr(Trainer, "get_batch_samples"):
-        from inspect import getsource
         if \
-            not getsource(Trainer.get_batch_samples).strip()\
+            not inspect.getsource(Trainer.get_batch_samples).strip()\
             .endswith("return batch_samples, num_items_in_batch"):
 
             raise NotImplementedError("Unsloth: Please make a Github issue immediately!!")
@@ -1198,4 +1198,33 @@ def patch_gradient_accumulation_fix(Trainer):
             '`pip install --upgrade --no-cache-dir unsloth git+https://github.com/huggingface/transformers.git git+https://github.com/huggingface/trl.git`'
         )
     pass
+
+    # Also fix up loss scaling ie negate loss *= self.args.gradient_accumulation_steps
+    if "num_items_in_batch" not in inspect.signature(Trainer.training_step).parameters: return
+
+    function = inspect.getsource(Trainer.training_step)
+    where = function.find("def")
+    function = function.split("\n")
+    function = "\n".join(x[where:] for x in function)
+
+    # Import all variables that need importing
+    import transformers.trainer
+    items_in_trainer = dir(transformers.trainer)
+    good_items = []
+    for item in items_in_trainer:
+        # TODO: Support Deepspeed
+        if item.startswith(("deepspeed", "xm", "met", "smp")): continue
+        if item in function: good_items.append(item)
+    pass
+    exec("from transformers.trainer import (" + ", ".join(x for x in good_items) + ")", globals())
+
+    # Accelerate does / self.args.gradient_accumulation_steps internally, so if we already
+    # summed it up and did the division before hand, we have to negate it.
+    function = function.replace(
+        "loss *= self.args.gradient_accumulation_steps",
+        "if num_items_in_batch is not None: loss *= self.args.gradient_accumulation_steps",
+    )
+    function = function.replace("def training_step", "def _unsloth_training_step", 1)
+    exec(function, globals())
+    Trainer.training_step = _unsloth_training_step
 pass
diff --git a/unsloth/save.py b/unsloth/save.py
index ab30e0fea5..ccda79aeee 100644
--- a/unsloth/save.py
+++ b/unsloth/save.py
@@ -145,7 +145,7 @@ pass
 
 def _merge_lora(layer, name):
 
-    bias = None
+    bias = getattr(layer, "bias", None)
     if isinstance(layer, (Bnb_Linear4bit, Peft_Linear4bit, Peft_Linear)):
         # Is LoRA so we need to merge!
         W, quant_state, A, B, s, bias = get_lora_parameters_bias(layer)
diff --git a/unsloth/tokenizer_utils.py b/unsloth/tokenizer_utils.py
index 4b9fd5e133..8806f1e743 100644
--- a/unsloth/tokenizer_utils.py
+++ b/unsloth/tokenizer_utils.py
@@ -914,7 +914,9 @@ def patch_sft_trainer_tokenizer():
 
         check_text = \
         "\n"\
-        "if 'tokenizer' not in locals(): tokenizer = processing_class\n"\
+        "if 'tokenizer'          not in locals(): tokenizer = processing_class\n"\
+        "if 'formatting_func'    not in locals(): raise RuntimeError('Unsloth: Please file a bug report - `formatting_func` does not exist!')\n"\
+        "if 'dataset_text_field' not in locals(): raise RuntimeError('Unsloth: Please file a bug report - `dataset_text_field` does not exist!')\n"\
         "test_text = dataset[0][dataset_text_field] if (formatting_func is None and dataset_text_field is not None) else formatting_func(dataset[0])[0]\n"\
         "chat_template = getattr(tokenizer, 'chat_template', None)\n"\
         "chat_template = '' if chat_template is None else chat_template\n"\
@@ -1017,7 +1019,10 @@ pass
 for trainer_name in ("SFTTrainer", "DPOTrainer", "KTOTrainer"):
     trainer_text = patch_trl_tokenizer_processing_class(trainer_name)
     if trainer_text is None: continue
-    exec(trainer_text, globals())
+    try:
+        exec(trainer_text, globals())
+    except:
+        raise RuntimeError(f"Unsloth: Please file a bug report! Error patching {trainer_name}")
     exec(f"trl.trainer.{trainer_name} = Unsloth{trainer_name}", globals())
 pass