[Bugfix] Adjust mllama to regional compilation (vllm-project#15112)

Signed-off-by: Jan Kaniecki <[email protected]>
Signed-off-by: Mu Huai <[email protected]>

Author: jkaniecki

vllm/model_executor/models/mllama.py

@@ -1070,8 +1070,8 @@ def forward(
         inputs_embeds = self.embed_tokens(input_ids)
         hidden_states = inputs_embeds
 
-        for decoder_layer in self.layers:
-            if isinstance(decoder_layer, MllamaCrossAttentionDecoderLayer):
+        for idx, decoder_layer in enumerate(self.layers):
+            if idx in self.cross_attention_layers:
                 if not skip_cross_attention:
                     hidden_states = decoder_layer(
                         hidden_states=hidden_states,
@@ -1081,16 +1081,13 @@ def forward(
                         full_text_row_masked_out_mask=
                         full_text_row_masked_out_mask,
                     )
-            elif isinstance(decoder_layer, LlamaDecoderLayer):
+            else:
                 hidden_states, residual = decoder_layer(
                     positions=positions,
                     hidden_states=hidden_states,
                     residual=None,
                 )
                 hidden_states = hidden_states + residual
-            else:
-                raise ValueError(
-                    f"Unknown decoder layer type {type(decoder_layer)}")
         hidden_states = self.norm(hidden_states)
         return hidden_states
 
@@ -1551,4 +1548,4 @@ def convert_dense_cross_attention_mask_to_tensor(
     full_text_mask = ((mask != ninf).any(dim=-1).type_as(mask)[..., None])
     mask *= full_text_mask
     # (num_prompt_tokens, num_encoder_tokens)
-    return mask
+    return mask