From 33a72ba122ed8f68b64ee34d8e8986a356281e37 Mon Sep 17 00:00:00 2001 From: Daniel Han-Chen Date: Sun, 25 Feb 2024 03:50:59 +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 6a44137402..fa2785faf3 100644 --- a/unsloth/models/gemma.py +++ b/unsloth/models/gemma.py @@ -84,6 +84,7 @@ class FastGemmaRotaryEmbedding(torch.nn.Module): 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) + print(freqs.to(torch.int64)) emb = torch.cat((freqs, freqs), dim=-1) self.cos_cached = emb.cos().to(dtype=torch.bfloat16) @@ -104,7 +105,6 @@ class FastGemmaRotaryEmbedding(torch.nn.Module): self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64, device="cuda").float() / self.dim) ) - print(self.inv_freq) t = torch.arange(self.max_position_embeddings, device="cpu", dtype=torch.int64).float().to("cuda").unsqueeze(0) inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to("cuda") position_ids_expanded = t[:, None, :].float()