Update UI and new models (lm-sys#2762)

silk55 · Jan 26, 2024 · 3f0dc7e · 3f0dc7e
1 parent 636d847
commit 3f0dc7e
Show file tree

Hide file tree

Showing 8 changed files with 213 additions and 134 deletions.
diff --git a/fastchat/model/model_adapter.py b/fastchat/model/model_adapter.py
@@ -53,7 +53,10 @@
 ANTHROPIC_MODEL_LIST = (
     "claude-1",
     "claude-2",
+    "claude-2.0",
+    "claude-2.1",
     "claude-instant-1",
+    "claude-instant-1.2",
 )
 
 

diff --git a/fastchat/model/model_registry.py b/fastchat/model/model_registry.py
@@ -40,17 +40,17 @@ def get_model_info(name: str) -> ModelInfo:
     "https://platform.openai.com/docs/models/gpt-3-5",
     "GPT-3.5-Turbo-1106 by OpenAI",
 )
-register_model_info(
-    ["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",
 )
 register_model_info(
-    ["claude-2"],
+    ["gpt-4"], "GPT-4", "https://openai.com/research/gpt-4", "ChatGPT-4 by OpenAI"
+)
+register_model_info(
+    ["claude-2.1", "claude-2.0"],
     "Claude",
     "https://www.anthropic.com/index/claude-2",
     "Claude 2 by Anthropic",
@@ -62,16 +62,34 @@ def get_model_info(name: str) -> ModelInfo:
     "Claude by Anthropic",
 )
 register_model_info(
-    ["claude-instant-1"],
+    ["claude-instant-1", "claude-instant-1.2"],
     "Claude Instant",
     "https://www.anthropic.com/index/introducing-claude",
     "Claude Instant by Anthropic",
 )
 register_model_info(
-    ["palm-2"],
-    "PaLM 2 Chat",
-    "https://cloud.google.com/vertex-ai/docs/release-notes#May_10_2023",
-    "PaLM 2 for Chat (chat-bison@001) by Google",
+    ["openhermes-2.5-mistral-7b"],
+    "OpenHermes-2.5-Mistral-7B",
+    "https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B",
+    "a mistral-based model fine-tuned on 1M GPT-4 outputs",
+)
+register_model_info(
+    ["starling-lm-7b-alpha"],
+    "Starling-LM-7B-alpha",
+    "https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha",
+    "an open model trained using RLAIF by Berkeley",
+)
+register_model_info(
+    ["tulu-2-dpo-70b"],
+    "Tulu 2",
+    "https://huggingface.co/allenai/tulu-2-dpo-70b",
+    "Tulu 2 by UW/AllenAI",
+)
+register_model_info(
+    ["yi-34b-chat", "yi-6b-chat"],
+    "Yi-Chat",
+    "https://huggingface.co/01-ai/Yi-34B-Chat",
+    "A large language model by 01 AI",
 )
 register_model_info(
     [
@@ -92,6 +110,18 @@ def get_model_info(name: str) -> ModelInfo:
     "https://ai.meta.com/llama/",
     "open foundation and fine-tuned chat models by Meta",
 )
+register_model_info(
+    ["chatglm3-6b", "chatglm2-6b", "chatglm-6b"],
+    "ChatGLM",
+    "https://chatglm.cn/blog",
+    "an open bilingual dialogue language model by Tsinghua University",
+)
+register_model_info(
+    ["openchat-3.5"],
+    "OpenChat 3.5",
+    "https://github.com/imoneoi/openchat",
+    "an open model fine-tuned on Mistral-7B using C-RLFT",
+)
 register_model_info(
     ["mistral-7b-instruct"],
     "Mistral",
@@ -158,12 +188,6 @@ def get_model_info(name: str) -> ModelInfo:
     "https://huggingface.co/BlinkDL/rwkv-4-raven",
     "an RNN with transformer-level LLM performance",
 )
-register_model_info(
-    ["chatglm-6b", "chatglm2-6b"],
-    "ChatGLM",
-    "https://chatglm.cn/blog",
-    "an open bilingual dialogue language model by Tsinghua University",
-)
 register_model_info(
     ["alpaca-13b"],
     "Alpaca",
@@ -183,16 +207,10 @@ def get_model_info(name: str) -> ModelInfo:
     "an Open Assistant for everyone by LAION",
 )
 register_model_info(
-    ["openchat-3.5"],
-    "OpenChat 3.5",
-    "https://github.com/imoneoi/openchat",
-    "OpenChat 3.5 is a versatile, open-source language model fine-tuned using C-RLFT",
-)
-register_model_info(
-    ["starling-lm-7b-alpha"],
-    "Starling-LM-7B-alpha",
-    "https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha",
-    "an open model trained using RLAIF by Berkeley",
+    ["palm-2"],
+    "PaLM 2 Chat",
+    "https://cloud.google.com/vertex-ai/docs/release-notes#May_10_2023",
+    "PaLM 2 for Chat (chat-bison@001) by Google",
 )
 register_model_info(
     ["llama-7b", "llama-13b"],
@@ -403,10 +421,3 @@ def get_model_info(name: str) -> ModelInfo:
     "https://huggingface.co/BAAI/AquilaChat2-34B",
     "Chat models developed by BAAI team",
 )
-
-register_model_info(
-    ["Yi-34B-Chat", "Yi-6B-Chat"],
-    "Yi-Chat",
-    "https://huggingface.co/01-ai",
-    "A large language model by 01.AI.",
-)
diff --git a/fastchat/serve/api_provider.py b/fastchat/serve/api_provider.py
@@ -22,6 +22,15 @@ def openai_api_stream_iter(
 ):
     import openai
 
+    is_azure = False
+    if "azure" in model_name:
+        is_azure = True
+        openai.api_type = "azure"
+        openai.api_version = "2023-07-01-preview"
+    else:
+        openai.api_type = "open_ai"
+        openai.api_version = None
+
     openai.api_base = api_base or "https://api.openai.com/v1"
     openai.api_key = api_key or os.environ["OPENAI_API_KEY"]
     if model_name == "gpt-4-turbo":
@@ -37,21 +46,31 @@ def openai_api_stream_iter(
     }
     logger.info(f"==== request ====\n{gen_params}")
 
-    res = openai.ChatCompletion.create(
-        model=model_name,
-        messages=messages,
-        temperature=temperature,
-        max_tokens=max_new_tokens,
-        stream=True,
-    )
+    if is_azure:
+        res = openai.ChatCompletion.create(
+            engine=model_name,
+            messages=messages,
+            temperature=temperature,
+            max_tokens=max_new_tokens,
+            stream=True,
+        )
+    else:
+        res = openai.ChatCompletion.create(
+            model=model_name,
+            messages=messages,
+            temperature=temperature,
+            max_tokens=max_new_tokens,
+            stream=True,
+        )
     text = ""
     for chunk in res:
-        text += chunk["choices"][0]["delta"].get("content", "")
-        data = {
-            "text": text,
-            "error_code": 0,
-        }
-        yield data
+        if len(chunk["choices"]) > 0:
+            text += chunk["choices"][0]["delta"].get("content", "")
+            data = {
+                "text": text,
+                "error_code": 0,
+            }
+            yield data
 
 
 def anthropic_api_stream_iter(model_name, prompt, temperature, top_p, max_new_tokens):

diff --git a/fastchat/serve/gradio_block_arena_anony.py b/fastchat/serve/gradio_block_arena_anony.py
@@ -29,6 +29,7 @@
     acknowledgment_md,
     ip_expiration_dict,
     get_ip,
+    get_model_description_md,
 )
 from fastchat.utils import (
     build_logger,
@@ -163,20 +164,26 @@ def share_click(state0, state1, model_selector0, model_selector1, request: gr.Re
     "gpt-4-turbo": 4,
     "gpt-3.5-turbo": 2,
     "gpt-3.5-turbo-1106": 2,
-    "claude-2": 8,
+    "claude-2.1": 4,
+    "claude-2.0": 2,
     "claude-1": 2,
-    "claude-instant-1": 8,
+    "claude-instant-1": 4,
+    "openhermes-2.5-mistral-7b": 2,
+    "wizardlm-70b": 2,
+    "starling-lm-7b-alpha": 2,
+    "tulu-2-dpo-70b": 2,
+    "yi-34b-chat": 2,
     "zephyr-7b-beta": 2,
     "openchat-3.5": 2,
+    "chatglm3-6b": 2,
     # tier 1
-    "deluxe-chat-v1.1": 2,
+    "deluxe-chat-v1.1": 4,
     "palm-2": 1.5,
     "llama-2-70b-chat": 1.5,
     "llama-2-13b-chat": 1.5,
     "codellama-34b-instruct": 1.5,
-    "vicuna-33b": 8,
+    "vicuna-33b": 4,
     "vicuna-13b": 1.5,
-    "wizardlm-70b": 1.5,
     "wizardlm-13b": 1.5,
     "qwen-14b-chat": 1.5,
     "mistral-7b-instruct": 1.5,
@@ -205,14 +212,21 @@ def share_click(state0, state1, model_selector0, model_selector1, request: gr.Re
 
 # target model sampling weights will be boosted.
 BATTLE_TARGETS = {
-    "gpt-4": {"claude-2"},
-    "gpt-4-turbo": {"gpt-4", "gpt-3.5-turbo"},
-    "gpt-3.5-turbo": {"claude-instant-1", "gpt-4", "claude-2"},
-    "claude-2": {"gpt-4", "gpt-3.5-turbo", "claude-1"},
-    "claude-1": {"claude-2", "gpt-4", "gpt-3.5-turbo"},
-    "claude-instant-1": {"gpt-3.5-turbo", "claude-2"},
-    "deluxe-chat-v1.1": {"gpt-4"},
+    "gpt-4": {"claude-2.1", "gpt-4-turbo"},
+    "gpt-4-turbo": {"gpt-4", "gpt-3.5-turbo", "gpt-3.5-turbo-1106", "claude-2.1"},
+    "gpt-3.5-turbo": {"claude-instant-1", "gpt-4", "claude-2.1"},
+    "gpt-3.5-turbo-1106": {"claude-instant-1", "gpt-3.5-turbo"},
+    "claude-2.1": {"gpt-4-turbo", "gpt-4", "claude-1"},
+    "claude-2.0": {"gpt-4-turbo", "gpt-4", "claude-1"},
+    "claude-1": {"claude-2.1", "gpt-4", "gpt-3.5-turbo"},
+    "claude-instant-1": {"gpt-3.5-turbo-1106", "claude-2.1"},
+    "deluxe-chat-v1.1": {"gpt-4", "gpt-4-turbo"},
+    "openhermes-2.5-mistral-7b": {"gpt-3.5-turbo", "openchat-3.5", "zephyr-7b-beta"},
+    "starling-lm-7b-alpha": {"gpt-3.5-turbo", "openchat-3.5", "zephyr-7b-beta"},
+    "tulu-2-dpo-70b": {"gpt-3.5-turbo", "vicuna-33b", "claude-instant-1"},
+    "yi-34b-chat": {"gpt-3.5-turbo", "vicuna-33b", "claude-instant-1"},
     "openchat-3.5": {"gpt-3.5-turbo", "llama-2-70b-chat", "zephyr-7b-beta"},
+    "chatglm3-6b": {"yi-34b-chat", "qwen-14b-chat"},
     "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": {
@@ -236,10 +250,24 @@ def share_click(state0, state1, model_selector0, model_selector1, request: gr.Re
     "palm-2": {"llama-2-13b-chat", "gpt-3.5-turbo"},
 }
 
-SAMPLING_BOOST_MODELS = ["openchat-3.5", "gpt-4-turbo", "gpt-3.5-turbo-1106"]
+SAMPLING_BOOST_MODELS = [
+    "tulu-2-dpo-70b",
+    "yi-34b-chat",
+    "claude-2.1",
+    "wizardlm-70b",
+    "starling-lm-7b-alpha",
+    "openhermes-2.5-mistral-7b",
+    "gpt-3.5-turbo-1106",
+    # "openchat-3.5",
+    # "gpt-4-turbo",
+    # "claude-1",
+]
 
 # outage models won't be sampled.
-OUTAGE_MODELS = []
+OUTAGE_MODELS = [
+    "zephyr-7b-alpha",
+    "falcon-180b-chat",
+]
 
 
 def get_sample_weight(model):
@@ -442,6 +470,9 @@ def build_side_by_side_ui_anony(models):
     gr.Markdown(notice_markdown, elem_id="notice_markdown")
 
     with gr.Box(elem_id="share-region-anony"):
+        with gr.Accordion("🔍 Expand to see 20+ Arena players", open=False):
+            model_description_md = get_model_description_md(models)
+            gr.Markdown(model_description_md, elem_id="model_description_markdown")
         with gr.Row():
             for i in range(num_sides):
                 label = "Model A" if i == 0 else "Model B"
@@ -470,15 +501,13 @@ def build_side_by_side_ui_anony(models):
             )
 
     with gr.Row():
-        with gr.Column(scale=20):
-            textbox = gr.Textbox(
-                show_label=False,
-                placeholder="👉 Enter your prompt and press ENTER",
-                container=False,
-                elem_id="input_box",
-            )
-        with gr.Column(scale=1, min_width=50):
-            send_btn = gr.Button(value="Send", variant="primary")
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt and press ENTER",
+            container=True,
+            elem_id="input_box",
+        )
+        send_btn = gr.Button(value="Send", variant="primary", scale=0)
 
     with gr.Row() as button_row:
         clear_btn = gr.Button(value="🎲 New Round", interactive=False)
@@ -511,7 +540,7 @@ def build_side_by_side_ui_anony(models):
             label="Max output tokens",
         )
 
-    gr.Markdown(acknowledgment_md)
+    gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
     # Register listeners
     btn_list = [

diff --git a/fastchat/serve/gradio_block_arena_named.py b/fastchat/serve/gradio_block_arena_named.py
@@ -284,10 +284,7 @@ def build_side_by_side_ui_named(models):
     model_selectors = [None] * num_sides
     chatbots = [None] * num_sides
 
-    model_description_md = get_model_description_md(models)
-    notice = gr.Markdown(
-        notice_markdown + model_description_md, elem_id="notice_markdown"
-    )
+    notice = gr.Markdown(notice_markdown, elem_id="notice_markdown")
 
     with gr.Box(elem_id="share-region-named"):
         with gr.Row():
@@ -300,6 +297,10 @@ def build_side_by_side_ui_named(models):
                         show_label=False,
                         container=False,
                     )
+        with gr.Row():
+            with gr.Accordion("🔍 Expand to see 20+ model descriptions", open=False):
+                model_description_md = get_model_description_md(models)
+                gr.Markdown(model_description_md, elem_id="model_description_markdown")
 
         with gr.Row():
             for i in range(num_sides):
@@ -322,19 +323,16 @@ def build_side_by_side_ui_named(models):
             )
 
     with gr.Row():
-        with gr.Column(scale=20):
-            textbox = gr.Textbox(
-                show_label=False,
-                placeholder="Enter your prompt here and press ENTER",
-                container=False,
-                elem_id="input_box",
-            )
-        with gr.Column(scale=1, min_width=50):
-            send_btn = gr.Button(value="Send", variant="primary")
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt and press ENTER",
+            elem_id="input_box",
+        )
+        send_btn = gr.Button(value="Send", variant="primary", scale=0)
 
     with gr.Row() as button_row:
-        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
         clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)
+        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
         share_btn = gr.Button(value="📷  Share")
 
     with gr.Accordion("Parameters", open=False) as parameter_row:
@@ -363,7 +361,7 @@ def build_side_by_side_ui_named(models):
             label="Max output tokens",
         )
 
-    gr.Markdown(acknowledgment_md)
+    gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
     # Register listeners
     btn_list = [