From 16dea671cda55d22a4b9eef697d9fedeef518ce2 Mon Sep 17 00:00:00 2001 From: DefiWimar7 Date: Thu, 28 Aug 2025 10:44:28 +0800 Subject: [PATCH] chore: Fix Typos --- unsloth/models/falcon_h1.py | 2 +- unsloth/models/gemma.py | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/unsloth/models/falcon_h1.py b/unsloth/models/falcon_h1.py index 5643e3dfe9..83eb538f0c 100644 --- a/unsloth/models/falcon_h1.py +++ b/unsloth/models/falcon_h1.py @@ -582,7 +582,7 @@ def _fast_prepare_inputs_for_generation( position_ids=None, use_cache=True, **kwargs,): - # Overwitten -- has a unique cache type, `FalconHybridMambaAttentionDynamicCache` + # Overwritten -- has a unique cache type, `FalconHybridMambaAttentionDynamicCache` empty_past_kv = past_key_values is None # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens diff --git a/unsloth/models/gemma.py b/unsloth/models/gemma.py index e43b205ecd..fc784be09d 100644 --- a/unsloth/models/gemma.py +++ b/unsloth/models/gemma.py @@ -246,7 +246,7 @@ class GemmaFixedRotaryEmbedding(torch.nn.Module): # in FP32. They are applied (multiplied) in FP32 as well. self.current_rope_size = seq_len - # The difference is we do division explicity instead of t * (1/x) ie we do t/x. + # The difference is we do division explicitly instead of t * (1/x) ie we do t/x. freq_exponents = (2.0 / self.dim) * ( torch.arange(self.dim // 2, dtype = torch.int64, device = "cpu").float() ) @@ -310,7 +310,7 @@ class GemmaFixedLinearScalingRotaryEmbedding(GemmaFixedRotaryEmbedding): # in FP32. They are applied (multiplied) in FP32 as well. self.current_rope_size = seq_len - # The difference is we do division explicity instead of t * (1/x) ie we do t/x. + # The difference is we do division explicitly instead of t * (1/x) ie we do t/x. freq_exponents = (2.0 / self.dim) * ( torch.arange(self.dim // 2, dtype = torch.int64, device = "cpu").float() )