fix

fastchat/constants.py

@@ -14,7 +14,7 @@
 MODERATION_MSG = "$MODERATION$ YOUR INPUT VIOLATES OUR CONTENT MODERATION GUIDELINES."
 CONVERSATION_LIMIT_MSG = "YOU HAVE REACHED THE CONVERSATION LENGTH LIMIT. PLEASE CLEAR HISTORY AND START A NEW CONVERSATION."
 INACTIVE_MSG = "THIS SESSION HAS BEEN INACTIVE FOR TOO LONG. PLEASE REFRESH THIS PAGE."
-SLOW_MODEL_MSG = '⚠️  Both models will show the responses all at once. Please stay patient as it may take over 30 seconds.'
+SLOW_MODEL_MSG = "⚠️  Both models will show the responses all at once. Please stay patient as it may take over 30 seconds."
 # Maximum input length
 INPUT_CHAR_LEN_LIMIT = int(os.getenv("FASTCHAT_INPUT_CHAR_LEN_LIMIT", 12000))
 # Maximum conversation turns

fastchat/model/model_registry.py

@@ -44,7 +44,10 @@ def get_model_info(name: str) -> ModelInfo:
     ["gpt-4"], "GPT-4", "https://openai.com/research/gpt-4", "ChatGPT-4 by OpenAI"
 )
 register_model_info(
-    ["gpt-4-turbo"], "GPT-4-Turbo", "https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo", "GPT-4-Turbo by OpenAI"
+    ["gpt-4-turbo"],
+    "GPT-4-Turbo",
+    "https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo",
+    "GPT-4-Turbo by OpenAI",
 )
 register_model_info(
     ["claude-2"],

fastchat/serve/gradio_block_arena_anony.py

@@ -204,7 +204,14 @@ def share_click(state0, state1, model_selector0, model_selector1, request: gr.Re
 }
 
 SAMPLING_BOOST_MODELS = ["openchat-3.5", "gpt-4-turbo", "gpt-3.5-turbo-1106"]
-OUTAGE_MODELS = ["claude-1", "claude-2", "claude-instant-1", "zephyr-7b-alpha", "wizardlm-70b", "falcon-180b-chat"]
+OUTAGE_MODELS = [
+    "claude-1",
+    "claude-2",
+    "claude-instant-1",
+    "zephyr-7b-alpha",
+    "wizardlm-70b",
+    "falcon-180b-chat",
+]
 
 
 def get_sample_weight(model):
@@ -233,11 +240,20 @@ def get_battle_pair():
         "openchat-3.5": {"gpt-3.5-turbo", "llama-2-70b-chat", "zephyr-7b-beta"},
         "qwen-14b-chat": {"vicuna-13b", "llama-2-13b-chat", "llama-2-70b-chat"},
         "zephyr-7b-alpha": {"mistral-7b-instruct", "llama-2-13b-chat"},
-        "zephyr-7b-beta": {"mistral-7b-instruct", "llama-2-13b-chat", "llama-2-7b-chat", "wizardlm-13b"},
+        "zephyr-7b-beta": {
+            "mistral-7b-instruct",
+            "llama-2-13b-chat",
+            "llama-2-7b-chat",
+            "wizardlm-13b",
+        },
         "llama-2-70b-chat": {"gpt-3.5-turbo", "vicuna-33b", "claude-instant-1"},
         "llama-2-13b-chat": {"mistral-7b-instruct", "vicuna-13b", "llama-2-70b-chat"},
         "llama-2-7b-chat": {"mistral-7b-instruct", "vicuna-7b", "llama-2-13b-chat"},
-        "mistral-7b-instruct": {"llama-2-7b-chat", "llama-2-13b-chat", "llama-2-70b-chat"},
+        "mistral-7b-instruct": {
+            "llama-2-7b-chat",
+            "llama-2-13b-chat",
+            "llama-2-70b-chat",
+        },
         "vicuna-33b": {"llama-2-70b-chat", "gpt-3.5-turbo", "claude-instant-1"},
         "vicuna-13b": {"llama-2-13b-chat", "llama-2-70b-chat"},
         "vicuna-7b": {"llama-2-7b-chat", "mistral-7b-instruct", "llama-2-13b-chat"},
@@ -259,8 +275,7 @@ def get_battle_pair():
         if model == chosen_model:
             continue
         weight = get_sample_weight(model)
-        if (weight != 0 and chosen_model in targets and
-                model in targets[chosen_model]):
+        if weight != 0 and chosen_model in targets and model in targets[chosen_model]:
             # boost to 50% chance
             weight = total_weight / len(targets[chosen_model])
         rival_models.append(model)
@@ -314,9 +329,7 @@ def add_text(
     model_list = [states[i].model_name for i in range(num_sides)]
     flagged = moderation_filter(text, model_list)
     if flagged:
-        logger.info(
-            f"violate moderation (anony). ip: {ip}. text: {text}"
-        )
+        logger.info(f"violate moderation (anony). ip: {ip}. text: {text}")
         # overwrite the original text
         text = MODERATION_MSG
 
@@ -541,7 +554,9 @@ def build_side_by_side_ui_anony(models):
         flash_buttons, [], btn_list
     )
     clear_btn.click(
-        clear_history, None, states + chatbots + model_selectors + [textbox] + btn_list + [slow_warning]
+        clear_history,
+        None,
+        states + chatbots + model_selectors + [textbox] + btn_list + [slow_warning],
     )
 
     share_js = """
@@ -575,7 +590,9 @@ def build_side_by_side_ui_anony(models):
         states + [temperature, top_p, max_output_tokens],
         states + chatbots + btn_list,
     ).then(
-        flash_buttons, [], btn_list,
+        flash_buttons,
+        [],
+        btn_list,
     )
 
     send_btn.click(

fastchat/serve/gradio_block_arena_named.py

@@ -174,9 +174,7 @@ def add_text(
     model_list = [states[i].model_name for i in range(num_sides)]
     flagged = moderation_filter(text, model_list)
     if flagged:
-        logger.info(
-            f"violate moderation (named). ip: {ip}. text: {text}"
-        )
+        logger.info(f"violate moderation (named). ip: {ip}. text: {text}")
         # overwrite the original text
         text = MODERATION_MSG
 

fastchat/serve/gradio_web_server.py

@@ -239,7 +239,7 @@ def clear_history(request: gr.Request):
 
 def get_ip(request: gr.Request):
     if "cf-connecting-ip" in request.headers:
-        ip = request.headers['cf-connecting-ip']
+        ip = request.headers["cf-connecting-ip"]
     else:
         ip = request.client.host
     return ip
@@ -259,9 +259,7 @@ def add_text(state, model_selector, text, request: gr.Request):
 
     flagged = moderation_filter(text, [state.model_name])
     if flagged:
-        logger.info(
-            f"violate moderation. ip: {ip}. text: {text}"
-        )
+        logger.info(f"violate moderation. ip: {ip}. text: {text}")
         # overwrite the original text
         text = MODERATION_MSG
 
@@ -559,8 +557,8 @@ def get_model_description_md(models):
         ct += 1
     return model_description_md
 
-def build_about():
 
+def build_about():
     about_markdown = f"""
 # About Us
 Chatbot Arena is an open-source research project developed by members from [LMSYS](https://lmsys.org/about/) and UC Berkeley [SkyLab](https://sky.cs.berkeley.edu/).  Our mission is to build an open crowdsourced platform to collect human feedback and evaluate LLMs under real-world scenarios. We open-source our code at [GitHub](https://github.com/lm-sys/FastChat) and release chat and human feedback datasets [here](https://github.com/lm-sys/FastChat/blob/main/docs/dataset_release.md). We invite everyone to join us in this journey!
@@ -592,10 +590,11 @@ def build_about():
 </div>
 """
 
-    #state = gr.State()
+    # state = gr.State()
     gr.Markdown(about_markdown, elem_id="about_markdown")
 
-    #return [state]
+    # return [state]
+
 
 def build_single_model_ui(models, add_promotion_links=False):
     promotion = (

fastchat/serve/gradio_web_server_multi.py

@@ -84,7 +84,12 @@ def load_demo(url_params, request: gr.Request):
     if args.anony_only_for_proprietary_model:
         # Only enable these models in anony battles.
         if args.add_chatgpt:
-            models_anony += ["gpt-4", "gpt-3.5-turbo", "gpt-4-turbo", "gpt-3.5-turbo-1106"]
+            models_anony += [
+                "gpt-4",
+                "gpt-3.5-turbo",
+                "gpt-4-turbo",
+                "gpt-3.5-turbo-1106",
+            ]
         if args.add_claude:
             models_anony += ["claude-2", "claude-1", "claude-instant-1"]
         if args.add_palm:
@@ -127,7 +132,6 @@ def build_demo(models, elo_results_file, leaderboard_table_file):
             with gr.Tab("About Us", id=4):
                 about = build_about()
 
-
         url_params = gr.JSON(visible=False)
 
         if args.model_list_mode not in ["once", "reload"]:

fastchat/serve/monitor/monitor.py

@@ -39,7 +39,7 @@ def make_leaderboard_md(elo_results):
 - [MT-Bench](https://arxiv.org/abs/2306.05685) - a set of challenging multi-turn questions. We use GPT-4 to grade the model responses.
 - [MMLU](https://arxiv.org/abs/2009.03300) (5-shot) - a test to measure a model's multitask accuracy on 57 tasks.
 
-💻 Code: The Arena Elo ratings are computed by this [notebook]({notebook_url}). The MT-bench scores (single-answer grading on a scale of 10) are computed by [fastchat.llm_judge](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge). The MMLU scores are mostly computed by [InstructEval](https://github.com/declare-lab/instruct-eval). Higher values are better for all benchmarks. Empty cells mean not available. Last updated: October, 2023.
+💻 Code: The Arena Elo ratings are computed by this [notebook]({notebook_url}). The MT-bench scores (single-answer grading on a scale of 10) are computed by [fastchat.llm_judge](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge). The MMLU scores are mostly computed by [InstructEval](https://github.com/declare-lab/instruct-eval). Higher values are better for all benchmarks. Empty cells mean not available. Last updated: November, 2023.
 """
     return leaderboard_md
 
@@ -218,7 +218,6 @@ def build_leaderboard_tab(elo_results_file, leaderboard_table_file):
     else:
         pass
 
-
     leader_component_values[:] = [md, p1, p2, p3, p4]
 
     """
@@ -253,6 +252,7 @@ def build_leaderboard_tab(elo_results_file, leaderboard_table_file):
     # return [md_1, plot_1, plot_2, plot_3, plot_4]
     return [md_1]
 
+
 def build_demo(elo_results_file, leaderboard_table_file):
     from fastchat.serve.gradio_web_server import block_css
 

fastchat/utils.py

@@ -148,6 +148,7 @@ def oai_moderation(text):
     Check whether the text violates OpenAI moderation API.
     """
     import openai
+
     openai.api_base = "https://api.openai.com/v1"
     openai.api_key = os.environ["OPENAI_API_KEY"]