diff options
| author | Sj-Si <sjw.jetty@gmail.com> | 2024-01-11 16:37:35 -0500 |
|---|---|---|
| committer | Sj-Si <sjw.jetty@gmail.com> | 2024-01-11 16:37:35 -0500 |
| commit | 036500223de0a3caaa86360a8ad3ed301e4367b0 (patch) | |
| tree | f05f0d5fc503d9c35d57bad077a5dab1dfd6569e /extensions-builtin/Lora/network_oft.py | |
| parent | 0726a6e12e85a37d1e514f5603acf9f058c11783 (diff) | |
| parent | cb5b335acddd126d4f6c990982816c06beb0d6ae (diff) | |
Merge changes from dev
Diffstat (limited to 'extensions-builtin/Lora/network_oft.py')
| -rw-r--r-- | extensions-builtin/Lora/network_oft.py | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index fa647020..342fcd0d 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -56,7 +56,7 @@ class NetworkModuleOFT(network.NetworkModule): self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) def calc_updown(self, orig_weight): - oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + oft_blocks = self.oft_blocks.to(orig_weight.device) eye = torch.eye(self.block_size, device=self.oft_blocks.device) if self.is_kohya: @@ -66,7 +66,7 @@ class NetworkModuleOFT(network.NetworkModule): block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) oft_blocks = torch.matmul(eye + block_Q, (eye - block_Q).float().inverse()) - R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + R = oft_blocks.to(orig_weight.device) # This errors out for MultiheadAttention, might need to be handled up-stream merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) @@ -77,6 +77,6 @@ class NetworkModuleOFT(network.NetworkModule): ) merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') - updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + updown = merged_weight.to(orig_weight.device) - orig_weight.to(merged_weight.dtype) output_shape = orig_weight.shape return self.finalize_updown(updown, orig_weight, output_shape) |