1

zRzRzRzRzRzRzR · zRzRzRzRzRzRzR · commit 1d91a24ec855 · 2025-02-19T19:52:01.000+08:00
diff --git a/scripts/convert_cogview4_to_diffusers_megatron.py b/scripts/convert_cogview4_to_diffusers_megatron.py
@@ -215,7 +215,7 @@ def convert_megatron_transformer_checkpoint_to_diffusers(
         # Attention Output
         new_state_dict[block_prefix + "attn1.to_out.0.weight"] = mega[
             f"decoder.layers.{i}.self_attention.linear_proj.weight"
-        ].T
+        ]
         new_state_dict[block_prefix + "attn1.to_out.0.bias"] = mega[
             f"decoder.layers.{i}.self_attention.linear_proj.bias"
         ]

Original file line number	Diff line number	Diff line change
`@@ -215,7 +215,7 @@ def convert_megatron_transformer_checkpoint_to_diffusers(`
`215`	`215`	`# Attention Output`
`216`	`216`	`new_state_dict[block_prefix + "attn1.to_out.0.weight"] = mega[`
`217`	`217`	`f"decoder.layers.{i}.self_attention.linear_proj.weight"`
`218`		`- ].T`
	`218`	`+ ]`
`219`	`219`	`new_state_dict[block_prefix + "attn1.to_out.0.bias"] = mega[`
`220`	`220`	`f"decoder.layers.{i}.self_attention.linear_proj.bias"`
`221`	`221`	`]`