Fix model bug.
This commit is contained in:
parent
72718e6b72
commit
b248d1d890
|
@ -305,8 +305,7 @@ class QwenRunner:
|
|||
layernorm_output = block.ln_1(hidden_states)
|
||||
|
||||
attn_outputs = self.forwardAttention(block.attn, layernorm_output, rotary_pos_emb_list)
|
||||
attn_output = attn_outputs[0]
|
||||
layernorm_input = attn_output + hidden_states
|
||||
layernorm_input = attn_outputs + hidden_states
|
||||
|
||||
layernorm_output = block.ln_2(layernorm_input)
|
||||
a1 = block.mlp.w1(layernorm_output)
|
||||
|
|
Loading…
Reference in New Issue