Add OpenAI API

0_🏠_Home.py

@@ -52,6 +52,7 @@ st.markdown("""
 - [Black Forest Labs](https://docs.bfl.ml)
 - [fal.ai](https://fal.ai/docs)
 - [Hugging Face](https://huggingface.co/docs/api-inference/index)
+- [OpenAI](https://platform.openai.com/docs/api-reference/introduction)
 - [Perplexity](https://docs.perplexity.ai/home)
 - [together.ai](https://docs.together.ai/docs/introduction)
 """)
@@ -59,7 +60,7 @@ st.markdown("""
 st.markdown("""
 ## Usage
 
-Choose a task. Select a service. Enter your API key (refresh browser to clear).
+Select a task. Choose a service. Enter your API key (refresh browser to clear).
 
 I recommend [duplicating this space](https://huggingface.co/spaces/adamelliotfields/api-inference?duplicate=true) **privately** and persisting your keys as secrets. See [`README.md`](https://huggingface.co/spaces/adamelliotfields/api-inference/blob/main/README.md).
 """)

README.md

@@ -30,6 +30,7 @@ Setting keys as environment variables persists them so you don't have to enter t
 BFL_API_KEY=...
 FAL_KEY=...
 HF_TOKEN=...
+OPENAI_API_KEY=...
 PPLX_API_KEY=...
 TOGETHER_API_KEY=...
 ```
@@ -49,7 +50,7 @@ source .venv/bin/activate
 uv pip install -r requirements.txt
 
 # run
-python app.py
+python 0_🏠_Home.py
 ```
 
 ## Development

lib/api.py

@@ -12,6 +12,7 @@ from .config import config
 
 def txt2txt_generate(api_key, service, model, parameters, **kwargs):
     base_url = config.services[service].url
+
     if service == "hf":
         base_url = f"{base_url}/{model}/v1"
     client = OpenAI(api_key=api_key, base_url=base_url)

lib/config.py

@@ -53,6 +53,8 @@ class TextModelConfig(ModelConfig):
     system_prompt: Optional[str] = None
     frequency_penalty: Optional[float] = None
     frequency_penalty_range: Optional[tuple[float, float]] = None
+    presence_penalty: Optional[float] = None
+    presence_penalty_range: Optional[tuple[float, float]] = None
     max_tokens: Optional[int] = None
     max_tokens_range: Optional[tuple[int, int]] = None
     temperature: Optional[float] = None
@@ -106,6 +108,19 @@ _hf_text_kwargs = {
     "parameters": ["max_tokens", "temperature", "frequency_penalty", "seed"],
 }
 
+_openai_text_kwargs = {
+    "system_prompt": TEXT_SYSTEM_PROMPT,
+    "frequency_penalty": 0.0,
+    "frequency_penalty_range": (-2.0, 2.0),
+    "presence_penalty": 0.0,
+    "presence_penalty_range": (-2.0, 2.0),
+    "max_tokens": 512,
+    "max_tokens_range": (512, 4096),
+    "temperature": 1.0,
+    "temperature_range": (0.0, 2.0),
+    "parameters": ["max_tokens", "temperature", "frequency_penalty", "presence_penalty", "seed"],
+}
+
 _pplx_text_kwargs = {
     "system_prompt": TEXT_SYSTEM_PROMPT,
     "frequency_penalty": 1.0,
@@ -302,18 +317,9 @@ config = AppConfig(
             url="https://api-inference.huggingface.co/models",
             api_key=os.environ.get("HF_TOKEN"),
             text={
-                "codellama/codellama-34b-instruct-hf": TextModelConfig(
-                    "Code Llama 34B",
-                    **_hf_text_kwargs,
-                ),
-                "meta-llama/llama-2-13b-chat-hf": TextModelConfig(
-                    "Meta Llama 2 13B",
-                    **_hf_text_kwargs,
-                ),
-                "mistralai/mistral-7b-instruct-v0.2": TextModelConfig(
-                    "Mistral 0.2 7B",
-                    **_hf_text_kwargs,
-                ),
+                "codellama/codellama-34b-instruct-hf": TextModelConfig("Code Llama 34B", **_hf_text_kwargs),
+                "meta-llama/llama-2-13b-chat-hf": TextModelConfig("Meta Llama 2 13B", **_hf_text_kwargs),
+                "mistralai/mistral-7b-instruct-v0.2": TextModelConfig("Mistral 0.2 7B", **_hf_text_kwargs),
                 "nousresearch/nous-hermes-2-mixtral-8x7b-dpo": TextModelConfig(
                     "Nous Hermes 2 Mixtral 8x7B",
                     **_hf_text_kwargs,
@@ -365,6 +371,20 @@ config = AppConfig(
                 ),
             },
         ),
+        "openai": ServiceConfig(
+            name="OpenAI",
+            url="https://api.openai.com/v1",
+            api_key=os.environ.get("OPENAI_API_KEY"),
+            text={
+                "chatgpt-4o-latest": TextModelConfig("ChatGPT-4o", **_openai_text_kwargs),
+                "gpt-3.5-turbo": TextModelConfig("GPT-3.5 Turbo", **_openai_text_kwargs),
+                "gpt-4-turbo": TextModelConfig("GPT-4 Turbo", **_openai_text_kwargs),
+                "gpt-4o": TextModelConfig("GPT-4o", **_openai_text_kwargs),
+                "gpt-4o-mini": TextModelConfig("GPT-4o mini", **_openai_text_kwargs),
+                "o1-preview": TextModelConfig("o1-preview", **_openai_text_kwargs),
+                "o1-mini": TextModelConfig("o1-mini", **_openai_text_kwargs),
+            },
+        ),
         "pplx": ServiceConfig(
             name="Perplexity",
             url="https://api.perplexity.ai",
@@ -392,7 +412,7 @@ config = AppConfig(
                 ),
             },
         ),
-        # TODO: text models
+        # TODO: text models, more image models
         "together": ServiceConfig(
             name="Together",
             url="https://api.together.xyz/v1/images/generations",

pages/1_💬_Text_Generation.py

@@ -4,17 +4,18 @@ import streamlit as st
 
 from lib import config, txt2txt_generate
 
-# config
 st.set_page_config(
     page_title=f"{config.title} | Text Generation",
     page_icon=config.logo,
     layout=config.layout,
 )
 
-# initialize state
 if "api_key_hf" not in st.session_state:
     st.session_state.api_key_hf = ""
 
+if "api_key_openai" not in st.session_state:
+    st.session_state.api_key_openai = ""
+
 if "api_key_pplx" not in st.session_state:
     st.session_state.api_key_pplx = ""
 
@@ -27,7 +28,6 @@ if "txt2txt_messages" not in st.session_state:
 if "txt2txt_seed" not in st.session_state:
     st.session_state.txt2txt_seed = 0
 
-# sidebar
 st.logo(config.logo)
 st.sidebar.header("Settings")
 
@@ -74,7 +74,12 @@ system = st.sidebar.text_area(
     disabled=st.session_state.running,
 )
 
-# build parameters from preset
+st.html("""
+    <h1>Text Generation</h1>
+    <p>Chat with large language models.</p>
+""")
+
+# Build parameters from preset by rendering the appropriate input widgets
 parameters = {}
 for param in model_config.parameters:
     if param == "max_tokens":
@@ -107,6 +112,16 @@ for param in model_config.parameters:
             disabled=st.session_state.running,
             help="Penalize new tokens based on their existing frequency in the text (default: 0.0)",
         )
+    if param == "presence_penalty":
+        parameters[param] = st.sidebar.slider(
+            "Presence Penalty",
+            step=0.1,
+            value=model_config.presence_penalty,
+            min_value=model_config.presence_penalty_range[0],
+            max_value=model_config.presence_penalty_range[1],
+            disabled=st.session_state.running,
+            help="Penalize new tokens based on their presence in the text so far (default: 0.0)",
+        )
     if param == "seed":
         parameters[param] = st.sidebar.number_input(
             "Seed",
@@ -117,18 +132,12 @@ for param in model_config.parameters:
             help="Make a best effort to sample deterministically (default: -1)",
         )
 
-# heading
-st.html("""
-    <h1>Text Generation</h1>
-    <p>Chat with large language models.</p>
-""")
-
-# chat messages
+# Chat messages
 for message in st.session_state.txt2txt_messages:
     with st.chat_message(message["role"]):
         st.markdown(message["content"])
 
-# button row
+# Buttons for deleting last message or clearing all messages
 if st.session_state.txt2txt_messages:
     button_container = st.empty()
     with button_container.container():
@@ -156,7 +165,7 @@ if st.session_state.txt2txt_messages:
 else:
     button_container = None
 
-# chat input
+# Chat input
 if prompt := st.chat_input(
     "What would you like to know?",
     on_submit=lambda: setattr(st.session_state, "running", True),

pages/2_🎨_Text_to_Image.py

@@ -10,7 +10,6 @@ st.set_page_config(
     layout=config.layout,
 )
 
-# Initialize Streamlit session state
 if "api_key_bfl" not in st.session_state:
     st.session_state.api_key_bfl = ""
 
@@ -72,7 +71,6 @@ model = st.sidebar.selectbox(
 
 model_config = service_config.image[model]
 
-# heading
 st.html("""
     <h1>Text to Image</h1>
     <p>Generate an image from a text prompt.</p>