From 7f5e3f0f82b74c9229d7ebf64ebf70dc9fe2d91e Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Tue, 23 Jul 2024 11:16:31 -0700 Subject: [PATCH] Update llama.py --- unsloth/models/llama.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py index d2bbb5a3a5..de9eb80dae 100644 --- a/unsloth/models/llama.py +++ b/unsloth/models/llama.py @@ -1080,7 +1080,7 @@ class LlamaExtendedRotaryEmbedding(torch.nn.Module): ) inv_freq = self.apply_scaling(inv_freq) self.register_buffer("inv_freq", inv_freq, persistent = False) - print(__FILE__, __LINE__) + print(__LINE__) # Build here to make `torch.jit.trace` work. self._set_cos_sin_cache(seq_len=self.current_rope_size, device=device, dtype=torch.get_default_dtype()) @@ -1090,7 +1090,7 @@ class LlamaExtendedRotaryEmbedding(torch.nn.Module): # Note: on the original Llama codebase, these tensors are created on the target device (and not on CPU) and # in FP32. They are applied (multiplied) in FP32 as well. self.current_rope_size = seq_len - print(__FILE__, __LINE__) + print(__LINE__) t = torch.arange(self.current_rope_size, device="cpu", dtype=torch.int64).float()