Merge operation & multimodal arena (#3425)

Co-authored-by: simon-mo <[email protected]>
Co-authored-by: Christopher Chou <[email protected]>

Author: 3 people

fastchat/constants.py

@@ -21,11 +21,11 @@
 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."
-RATE_LIMIT_MSG = "**RATE LIMIT OF THIS MODEL IS REACHED. PLEASE COME BACK LATER OR USE BATTLE MODE (the 1st tab).**"
+RATE_LIMIT_MSG = "**RATE LIMIT OF THIS MODEL IS REACHED. PLEASE COME BACK LATER OR USE <span style='color: red; font-weight: bold;'>[BATTLE MODE](https://chat.lmsys.org)</span> (the 1st tab).**"
 # Maximum input length
 INPUT_CHAR_LEN_LIMIT = int(os.getenv("FASTCHAT_INPUT_CHAR_LEN_LIMIT", 12000))
 BLIND_MODE_INPUT_CHAR_LEN_LIMIT = int(
-    os.getenv("FASTCHAT_BLIND_MODE_INPUT_CHAR_LEN_LIMIT", 24000)
+    os.getenv("FASTCHAT_BLIND_MODE_INPUT_CHAR_LEN_LIMIT", 30000)
 )
 # Maximum conversation turns
 CONVERSATION_TURN_LIMIT = 50

fastchat/conversation.py

@@ -61,6 +61,7 @@
     "claude-3-haiku-20240307-vertex",
     "claude-3-sonnet-20240229",
     "claude-3-sonnet-20240229-vertex",
+    "claude-3-5-sonnet-20240620",
     "claude-3-opus-20240229",
     "claude-instant-1",
     "claude-instant-1.2",
@@ -80,6 +81,10 @@
     "gpt-4-0125-preview",
     "gpt-4-turbo-browsing",
     "gpt-4-turbo-2024-04-09",
+    "gpt2-chatbot",
+    "im-also-a-good-gpt2-chatbot",
+    "im-a-good-gpt2-chatbot",
+    "gpt-4o-2024-05-13",
 )
 
 
@@ -1111,6 +1116,10 @@ def get_default_conv_template(self, model_path: str) -> Conversation:
             return get_conv_template("api_based_default")
         if "gpt-4-turbo-2024-04-09" in model_path:
             return get_conv_template("gpt-4-turbo-2024-04-09")
+        if "gpt2-chatbot" in model_path:
+            return get_conv_template("gpt-4-turbo-2024-04-09")
+        if "gpt-4o" in model_path:
+            return get_conv_template("gpt-4-turbo-2024-04-09")
         return get_conv_template("chatgpt")
 
 
@@ -1157,6 +1166,8 @@ def get_default_conv_template(self, model_path: str) -> Conversation:
             return get_conv_template("claude-3-haiku-20240307")
         if "claude-3-sonnet" in model_path:
             return get_conv_template("claude-3-sonnet-20240229")
+        if "claude-3-5-sonnet" in model_path:
+            return get_conv_template("claude-3-5-sonnet-20240620")
         if "claude-3-opus" in model_path:
             return get_conv_template("claude-3-opus-20240229")
         return get_conv_template("claude")
@@ -2179,6 +2190,21 @@ def get_default_conv_template(self, model_path: str) -> Conversation:
         return get_conv_template("deepseek-chat")
 
 
+class GeminiAdapter(BaseModelAdapter):
+    """The model adapter for Gemini"""
+
+    def match(self, model_path: str):
+        return "gemini" in model_path.lower() or "bard" in model_path.lower()
+
+    def load_model(self, model_path: str, from_pretrained_kwargs: dict):
+        raise NotImplementedError()
+
+    def get_default_conv_template(self, model_path: str) -> Conversation:
+        if "gemini-1.5-pro" in model_path:
+            return get_conv_template("gemini-1.5-pro")
+        return get_conv_template("gemini")
+
+
 class Yuan2Adapter(BaseModelAdapter):
     """The model adapter for Yuan2.0"""
 
@@ -2470,7 +2496,6 @@ def get_default_conv_template(self, model_path: str) -> Conversation:
 register_model_adapter(InternLMChatAdapter)
 register_model_adapter(StarChatAdapter)
 register_model_adapter(Llama2Adapter)
-register_model_adapter(Llama3Adapter)
 register_model_adapter(CuteGPTAdapter)
 register_model_adapter(OpenOrcaAdapter)
 register_model_adapter(DolphinAdapter)
@@ -2519,6 +2544,7 @@ def get_default_conv_template(self, model_path: str) -> Conversation:
 register_model_adapter(CllmAdapter)
 register_model_adapter(RekaAdapter)
 register_model_adapter(SmaugChatAdapter)
+register_model_adapter(Llama3Adapter)
 
 # After all adapters, try the default base adapter.
 register_model_adapter(BaseModelAdapter)

fastchat/model/model_adapter.py

@@ -42,6 +42,7 @@ def get_model_info(name: str) -> ModelInfo:
 
 register_model_info(
     [
+        "claude-3-5-sonnet-20240620",
         "claude-3-haiku-20240307",
         "claude-3-sonnet-20240229",
         "claude-3-opus-20240229",
@@ -55,7 +56,139 @@ def get_model_info(name: str) -> ModelInfo:
 )
 
 register_model_info(
-    ["reka-flash", "reka-flash-online"],
+    ["gemma-2-27b-it", "gemma-2-9b-it"],
+    "Gemma 2",
+    "https://ai.google.dev/gemma",
+    "Gemma 2 by Google",
+)
+
+register_model_info(
+    [
+        "gemini-1.5-pro-api-0409-preview",
+        "gemini-1.5-pro-tuned",
+        "gemini-1.5-pro-api-preview",
+        "gemini-1.5-flash-api-preview",
+        "gemini-1.5-flash-api-0514",
+        "gemini-1.5-pro-api-0514",
+        "gemini-advanced-0514",
+    ],
+    "Gemini",
+    "https://deepmind.google/technologies/gemini/",
+    "Gemini by Google",
+)
+
+register_model_info(
+    ["deepseek-coder-v2"],
+    "DeepSeek Coder v2",
+    "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct",
+    "An advanced code model by DeepSeek",
+)
+
+register_model_info(
+    ["nemotron-4-340b"],
+    "Nemotron-4 340B",
+    "https://huggingface.co/nvidia/Nemotron-4-340B-Instruct",
+    "Cutting-edge Open model by Nvidia",
+)
+
+register_model_info(
+    ["llama-3-70b-instruct", "llama-3-8b-instruct"],
+    "Llama 3",
+    "https://ai.meta.com/blog/meta-llama-3/",
+    "Open foundation and chat models by Meta",
+)
+
+register_model_info(
+    [
+        "qwen2-72b-instruct",
+        "qwen-max-0403",
+        "qwen-max-0428",
+    ],
+    "Qwen Max",
+    "https://help.aliyun.com/zh/dashscope/developer-reference/model-introduction",
+    "The Frontier Qwen Model by Alibaba",
+)
+
+register_model_info(
+    [
+        "gpt-4o-2024-05-13",
+    ],
+    "GPT-4o",
+    "https://openai.com/index/hello-gpt-4o/",
+    "The flagship model across audio, vision, and text by OpenAI",
+)
+
+register_model_info(
+    [
+        "gpt-4-turbo",
+        "gpt-4o-2024-05-13",
+        "gpt-4-turbo-2024-04-09",
+        "gpt-4-1106-preview",
+        "gpt-4-0125-preview",
+        "gpt2-chatbot",
+        "im-also-a-good-gpt2-chatbot",
+        "im-a-good-gpt2-chatbot",
+    ],
+    "GPT-4-Turbo",
+    "https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo",
+    "GPT-4-Turbo by OpenAI",
+)
+
+register_model_info(
+    [
+        "gpt-3.5-turbo",
+        "gpt-3.5-turbo-0125",
+        "gpt-3.5-turbo-1106",
+        "gpt-3.5-turbo-0314",
+        "gpt-3.5-turbo-0613",
+    ],
+    "GPT-3.5",
+    "https://platform.openai.com/docs/models/gpt-3-5",
+    "GPT-3.5-Turbo by OpenAI",
+)
+
+register_model_info(
+    ["yi-large-preview", "yi-large"],
+    "Yi-Large",
+    "https://x.com/01AI_Yi/status/1789894091620458667",
+    "State-of-the-art model by 01 AI",
+)
+
+register_model_info(
+    ["yi-1.5-34b-chat", "yi-34b-chat", "yi-6b-chat"],
+    "Yi-Chat",
+    "https://huggingface.co/01-ai/",
+    "A large language model by 01 AI",
+)
+
+register_model_info(
+    [
+        "phi-3-medium-4k-instruct",
+        "phi-3-small-8k-instruct",
+        "phi-3-mini-4k-instruct",
+        "phi-3-mini-128k-instruct",
+    ],
+    "Phi-3",
+    "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/",
+    "A capable and cost-effective small language models (SLMs) by Microsoft",
+)
+
+register_model_info(
+    ["reka-core-20240501"],
+    "Reka Core",
+    "https://www.reka.ai/news/reka-core-our-frontier-class-multimodal-language-model",
+    "Frontier Multimodal Language Model by Reka",
+)
+
+register_model_info(
+    ["gemini-pro", "gemini-pro-dev-api"],
+    "Gemini",
+    "https://blog.google/technology/ai/google-gemini-pro-imagen-duet-ai-update/",
+    "Gemini by Google",
+)
+
+register_model_info(
+    ["reka-flash-preview-20240611", "reka-flash", "reka-flash-online"],
     "Reka Flash",
     "https://www.reka.ai/news/reka-flash-efficient-and-capable-multimodal-language-models",
     "Multimodal model by Reka",
@@ -75,22 +208,23 @@ def get_model_info(name: str) -> ModelInfo:
     "Command-R by Cohere",
 )
 
-register_model_info(
-    [
-        "zephyr-orpo-141b-A35b-v0.1",
-    ],
-    "Zephyr 141B-A35B",
-    "https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1",
-    "ORPO fine-tuned of Mixtral-8x22B-v0.1",
-)
-
 register_model_info(
     ["gemma-1.1-7b-it", "gemma-1.1-2b-it", "gemma-7b-it", "gemma-2b-it"],
     "Gemma",
     "https://blog.google/technology/developers/gemma-open-models/",
     "Gemma by Google",
 )
 
+register_model_info(
+    [
+        "codestral-2405",
+        "mixtral-8x22b-instruct-v0.1",
+    ],
+    "Mixtral of experts",
+    "https://mistral.ai/news/mixtral-8x22b/",
+    "A Mixture-of-Experts model by Mistral AI",
+)
+
 register_model_info(
     [
         "mixtral-8x7b-instruct-v0.1",
@@ -105,6 +239,15 @@ def get_model_info(name: str) -> ModelInfo:
     "A Mixture-of-Experts model by Mistral AI",
 )
 
+register_model_info(
+    [
+        "qwen1.5-110b-chat",
+    ],
+    "Qwen 1.5",
+    "https://qwenlm.github.io/blog/qwen1.5-110b/",
+    "The First 100B+ Model of the Qwen1.5 Series",
+)
+
 register_model_info(
     [
         "qwen1.5-72b-chat",
@@ -121,14 +264,36 @@ def get_model_info(name: str) -> ModelInfo:
     "A large language model by Alibaba Cloud",
 )
 
+register_model_info(
+    ["glm-4-0520", "glm-4-0116"],
+    "GLM-4",
+    "https://zhipuai.cn/devday",
+    "Next-Gen Foundation Model by Zhipu AI",
+)
+
+register_model_info(
+    ["snowflake-arctic-instruct"],
+    "Snowflake Arctic Instruct",
+    "https://www.snowflake.com/blog/arctic-open-efficient-foundation-language-models-snowflake/",
+    "Dense-MoE transformer by Snowflake AI",
+)
 
 register_model_info(
-    ["dbrx-instruct"],
+    ["dbrx-instruct", "dbrx-next"],
     "DBRX Instruct",
     "https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm",
     "DBRX by Databricks Mosaic AI",
 )
 
+register_model_info(
+    [
+        "zephyr-orpo-141b-A35b-v0.1",
+    ],
+    "Zephyr 141B-A35B",
+    "https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1",
+    "ORPO fine-tuned of Mixtral-8x22B-v0.1",
+)
+
 register_model_info(
     ["starling-lm-7b-beta", "starling-lm-7b-alpha"],
     "Starling-LM-7B",
@@ -152,8 +317,6 @@ def get_model_info(name: str) -> ModelInfo:
 
 register_model_info(
     [
-        "gemini-pro",
-        "gemini-pro-dev-api",
         "gemini-1.0-pro-vision",
         "gemini-1.5-pro-preview-0409",
     ],
@@ -176,38 +339,13 @@ def get_model_info(name: str) -> ModelInfo:
     "A model trained using depth up-scaling by Upstage AI",
 )
 
-register_model_info(
-    [
-        "gpt-4-turbo",
-        "gpt-4-turbo-2024-04-09",
-        "gpt-4-1106-preview",
-        "gpt-4-0125-preview",
-    ],
-    "GPT-4-Turbo",
-    "https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo",
-    "GPT-4-Turbo by OpenAI",
-)
-
 register_model_info(
     ["gpt-4-turbo-browsing"],
     "GPT-4-Turbo with browsing",
     "https://platform.openai.com/docs/assistants/overview",
     "GPT-4-Turbo with browsing by OpenAI",
 )
 
-register_model_info(
-    [
-        "gpt-3.5-turbo",
-        "gpt-3.5-turbo-0125",
-        "gpt-3.5-turbo-1106",
-        "gpt-3.5-turbo-0314",
-        "gpt-3.5-turbo-0613",
-    ],
-    "GPT-3.5",
-    "https://platform.openai.com/docs/models/gpt-3-5",
-    "GPT-3.5-Turbo by OpenAI",
-)
-
 register_model_info(
     ["gpt-4", "gpt-4-0314", "gpt-4-0613"],
     "GPT-4",
@@ -250,13 +388,6 @@ def get_model_info(name: str) -> ModelInfo:
     "A chat assistant fine-tuned on user-shared conversations by LMSYS",
 )
 
-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(
     [
         "codellama-70b-instruct",
@@ -755,10 +886,3 @@ def get_model_info(name: str) -> ModelInfo:
     "https://huggingface.co/cllm",
     "consistency-llm is a new generation of parallel decoder LLMs with fast generation speed.",
 )
-
-register_model_info(
-    ["reka-flash", "reka-flash-20240226"],
-    "Reka Flash",
-    "https://reka.ai/reka-flash",
-    "Multimodal model by Reka",
-)