Update Visual-CoT demo - 2025-10-12 23:10:09

Fixes:
- Fix LLaVA config registration error (compatibility with newer transformers)
- Update Gradio to latest version (security fixes)
- Auto-deployed via update script

llava/model/language_model/modeling_llamantk.py

@@ -144,6 +144,7 @@ def call_flash_attn_qkvpacked(qkv, cu_seqlens, max_seqlen, dropout_p=0.0, softma
             cu_seqlens_k=cu_seqlens,
             max_seqlen_q=max_seqlen,
             max_seqlen_k=max_seqlen,
+            is_causal=causal,  # Add causal mask parameter
         )[0]
         return output
     elif HAS_FLASH_ATTN: