From 765e54fe5550fca9cb289ca655b1c85e1e4b0087 Mon Sep 17 00:00:00 2001 From: Daniel Han-Chen Date: Sun, 25 Feb 2024 03:38:18 +1100 Subject: [PATCH] Update gemma.py --- unsloth/models/gemma.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/unsloth/models/gemma.py b/unsloth/models/gemma.py index c4ec330b91..14cf7f5092 100644 --- a/unsloth/models/gemma.py +++ b/unsloth/models/gemma.py @@ -80,7 +80,7 @@ class FastGemmaRotaryEmbedding(torch.nn.Module): self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64, device="cpu").float() / self.dim) ) t = torch.arange(self.max_position_embeddings, device="cpu", dtype=torch.int64).float().to("cuda").unsqueeze(0) - inv_freq_expanded = inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to("cuda") + inv_freq_expanded = inv_freq[None, :, None].float().expand(1, -1, 1).to("cuda") position_ids_expanded = t[:, None, :].float() freqs = (inv_freq_expanded @ position_ids_expanded).transpose(1, 2) emb = torch.cat((freqs, freqs), dim=-1)