Add FastChat xgen inference fix patch

deployment/benchmark.Dockerfile

@@ -30,6 +30,11 @@ RUN cd /workspace \
       && rm -r .git \
       && pip install -e .
 
+# Apply patches
+# Salesforce xgen inference fix (https://github.com/lm-sys/FastChat/pull/2350)
+RUN cd /root/.local/miniconda3/lib/python3.9/site-packages/fastchat \
+      && patch -p1 < /workspace/leaderboard/deployment/fastchat_xgen_fix.patch
+
 # Where all the weights downloaded from Hugging Face Hub will go to
 ENV TRANSFORMERS_CACHE=/data/leaderboard/hfcache
 

deployment/fastchat_xgen_fix.patch

@@ -0,0 +1,18 @@
+--- a/fastchat/conversation.py
++++ b/fastchat/conversation.py
+@@ -787,12 +787,12 @@ def get_conv_template(name: str) -> Conversation:
+     Conversation(
+         name="xgen",
+         system_message="A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\n\n",
+-        roles=("### Human: ", "###"),
++        roles=("### Human", "### Assistant"),
+         messages=(),
+         offset=0,
+-        sep_style=SeparatorStyle.NO_COLON_SINGLE,
++        sep_style=SeparatorStyle.ADD_COLON_SINGLE,
+         sep="\n",
+-        stop_token_ids=[50256, 0, 1, 2],
++        stop_token_ids=[50256],
+         stop_str="<|endoftext|>",
+     )
+ )