Add Anthropic API

0_🏠_Home.py

@@ -49,6 +49,7 @@ st.page_link("pages/2_🎨_Text_to_Image.py", label="Text to Image", icon="🎨"
 st.markdown("""
 ## Services
 
+- [Anthropic](https://docs.anthropic.com/en/api/getting-started)
 - [Black Forest Labs](https://docs.bfl.ml)
 - [fal.ai](https://fal.ai/docs)
 - [Hugging Face](https://huggingface.co/docs/api-inference/index)

README.md

@@ -27,6 +27,7 @@ models:
 Setting keys as environment variables persists them so you don't have to enter them on every page load.
 
 ```bash
+ANTHROPIC_API_KEY=...
 BFL_API_KEY=...
 FAL_KEY=...
 HF_TOKEN=...

lib/api.py

@@ -4,23 +4,33 @@ import time
 
 import httpx
 import streamlit as st
-from openai import APIError, OpenAI
+from anthropic import Anthropic
+from anthropic import APIError as AnthropicAPIError
+from openai import APIError as OpenAIAPIError
+from openai import OpenAI
 from PIL import Image
 
 from .config import config
 
 
-def txt2txt_generate(api_key, service, model, parameters, **kwargs):
+def txt2txt_generate(api_key, service, 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)
+        base_url = f"{base_url}/{parameters['model']}/v1"
 
     try:
-        stream = client.chat.completions.create(stream=True, model=model, **parameters, **kwargs)
-        return st.write_stream(stream)
-    except APIError as e:
+        if service == "anthropic":
+            client = Anthropic(api_key=api_key)
+            with client.messages.stream(**parameters, **kwargs) as stream:
+                return st.write_stream(stream.text_stream)
+        else:
+            client = OpenAI(api_key=api_key, base_url=base_url)
+            stream = client.chat.completions.create(stream=True, **parameters, **kwargs)
+            return st.write_stream(stream)
+    except AnthropicAPIError as e:
+        return e.message
+    except OpenAIAPIError as e:
         # OpenAI uses this message for streaming errors and attaches response.error to error.body
         # https://github.com/openai/openai-python/blob/v1.0.0/src/openai/_streaming.py#L59
         return e.body if e.message == "An error occurred during streaming" else e.message

lib/config.py

@@ -97,6 +97,15 @@ class AppConfig:
     services: Dict[str, ServiceConfig]
 
 
+_anthropic_text_kwargs = {
+    "system_prompt": TEXT_SYSTEM_PROMPT,
+    "max_tokens": 512,
+    "max_tokens_range": (512, 4096),
+    "temperature": 0.5,
+    "temperature_range": (0.0, 1.0),
+    "parameters": ["max_tokens", "temperature"],
+}
+
 _hf_text_kwargs = {
     "system_prompt": TEXT_SYSTEM_PROMPT,
     "frequency_penalty": 0.0,
@@ -129,7 +138,7 @@ _pplx_text_kwargs = {
     "max_tokens_range": (512, 4096),
     "temperature": 1.0,
     "temperature_range": (0.0, 2.0),
-    "parameters": ["max_tokens", "temperature", "frequency_penalty", "seed"],
+    "parameters": ["max_tokens", "temperature", "frequency_penalty"],
 }
 
 config = AppConfig(
@@ -153,6 +162,17 @@ config = AppConfig(
         "sync_mode",
     ],
     services={
+        "anthropic": ServiceConfig(
+            name="Anthropic",
+            url="https://api.anthropic.com/v1",
+            api_key=os.environ.get("ANTHROPIC_API_KEY"),
+            text={
+                "claude-3-haiku-20240307": TextModelConfig("Claude 3 Haiku", **_anthropic_text_kwargs),
+                "claude-3-opus-20240229": TextModelConfig("Claude 3 Opus", **_anthropic_text_kwargs),
+                "claude-3-sonnet-20240229": TextModelConfig("Claude 3 Sonnet", **_anthropic_text_kwargs),
+                "claude-3-5-sonnet-20240620": TextModelConfig("Claude 3.5 Sonnet", **_anthropic_text_kwargs),
+            },
+        ),
         "bfl": ServiceConfig(
             name="Black Forest Labs",
             url="https://api.bfl.ml/v1",

pages/1_💬_Text_Generation.py

@@ -10,6 +10,9 @@ st.set_page_config(
     layout=config.layout,
 )
 
+if "api_key_anthropic" not in st.session_state:
+    st.session_state.api_key_anthropic = ""
+
 if "api_key_hf" not in st.session_state:
     st.session_state.api_key_hf = ""
 
@@ -80,7 +83,7 @@ st.html("""
 """)
 
 # Build parameters from preset by rendering the appropriate input widgets
-parameters = {}
+parameters = {"model": model}
 for param in model_config.parameters:
     if param == "max_tokens":
         parameters[param] = st.sidebar.slider(
@@ -92,6 +95,7 @@ for param in model_config.parameters:
             disabled=st.session_state.running,
             help="Maximum number of tokens to generate (default: 512)",
         )
+
     if param == "temperature":
         parameters[param] = st.sidebar.slider(
             "Temperature",
@@ -102,6 +106,7 @@ for param in model_config.parameters:
             disabled=st.session_state.running,
             help="Used to modulate the next token probabilities (default: 1.0)",
         )
+
     if param == "frequency_penalty":
         parameters[param] = st.sidebar.slider(
             "Frequency Penalty",
@@ -112,6 +117,7 @@ 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",
@@ -122,6 +128,7 @@ for param in model_config.parameters:
             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",
@@ -180,7 +187,12 @@ if prompt := st.chat_input(
     if button_container:
         button_container.empty()
 
-    messages = [{"role": "system", "content": system}]
+    if service == "anthropic":
+        messages = []
+        parameters["system"] = system
+    else:
+        messages = [{"role": "system", "content": system}]
+
     messages.extend([{"role": m["role"], "content": m["content"]} for m in st.session_state.txt2txt_messages])
     messages.append({"role": "user", "content": prompt})
     parameters["messages"] = messages

requirements.txt

@@ -1,3 +1,4 @@
+anthropic==0.36.0
 h2
 httpx
 openai==1.41.0