Typos

unsloth/models/falcon_h1.py

@@ -726,7 +726,7 @@ class FastFalconH1Model(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.falcon_h1.modeling_falcon_h1

unsloth/models/gemma.py

@@ -416,7 +416,7 @@ class FastGemmaModel(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.gemma.modeling_gemma

unsloth/models/gemma2.py

@@ -550,7 +550,7 @@ class FastGemma2Model(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.gemma2.modeling_gemma2

unsloth/models/llama.py

@@ -1533,7 +1533,7 @@ def PeftModel_fast_forward(
 
 # Solves https://github.com/unslothai/unsloth/issues/168
 # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-# Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+# Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
 # https://github.com/huggingface/transformers/pull/27931
 # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
 class LlamaRotaryEmbedding(torch.nn.Module):

unsloth/models/mistral.py

@@ -439,7 +439,7 @@ class FastMistralModel(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.mistral.modeling_mistral

unsloth/models/qwen2.py

@@ -60,7 +60,7 @@ class FastQwen2Model(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.qwen2.modeling_qwen2

unsloth/models/qwen3.py

@@ -451,7 +451,7 @@ class FastQwen3Model(FastLlamaModel):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py
         import transformers.models.qwen3.modeling_qwen3

unsloth/models/qwen3_moe.py

@@ -200,7 +200,7 @@ class FastQwen3MoeModel(FastQwen3Model):
 
         # Solves https://github.com/unslothai/unsloth/issues/168
         # Static KV Cache was introduced in 4.38.0, causing training to be much slower.
-        # Inferene can now be CUDAGraphed, but we shall retain the old rotary embeddings.
+        # Inference can now be CUDAGraphed, but we shall retain the old rotary embeddings.
         # https://github.com/huggingface/transformers/pull/27931
         # https://github.com/huggingface/transformers/blob/v4.37.2/src/transformers/models/llama/modeling_llama.py\
         import transformers.models.qwen3_moe.modeling_qwen3_moe