Fix llava_next for llama 3.2 vision cross attention states (#641)

predibase · Oct 15, 2024 · 4c84b4f · 4c84b4f
1 parent bea8834
commit 4c84b4f
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/server/lorax_server/models/custom_modeling/llava_next.py b/server/lorax_server/models/custom_modeling/llava_next.py
@@ -258,6 +258,7 @@ def forward(
             input_lengths=input_lengths,
             max_s=max_s,
             prefill_cache_indices=None,
+            cross_attention_states=None,
             adapter_data=adapter_data,
         )
         if lm_head_indices is not None: