sonnet to haiku (#305)

Author: Erick Friis

MODIFY.md

@@ -73,7 +73,7 @@ The LLM is used inside the `/chat` endpoint for generating the final answer, and
 Without any modification, we offer a few LLM providers out of the box:
 
 - `gpt-3.5-turbo-0125` by OpenAI
-- `claude-3-sonnet-20240229` by Anthropic
+- `claude-3-haiku-20240307` by Anthropic
 - `mixtral-8x7b` by Fireworks
 - `gemini-pro` by Google
 - `command` by Cohere
@@ -97,8 +97,8 @@ llm = ChatOpenAI(
     # When configuring the end runnable, we can then use this id to configure this field
     ConfigurableField(id="llm"),
     default_key="openai_gpt_3_5_turbo",
-    anthropic_claude_3_sonnet=ChatAnthropic(
-        model="claude-3-sonnet-20240229",
+    anthropic_claude_3_haiku=ChatAnthropic(
+        model="claude-3-haiku-20240307",
         max_tokens=16384,
         temperature=0,
         anthropic_api_key=os.environ.get("ANTHROPIC_API_KEY", "not_provided"),
@@ -138,7 +138,7 @@ response_synthesizer = (
     default_response_synthesizer.configurable_alternatives(
         ConfigurableField("llm"),
         default_key="openai_gpt_3_5_turbo",
-        anthropic_claude_3_sonnet=default_response_synthesizer,
+        anthropic_claude_3_haiku=default_response_synthesizer,
         ...
         local_ollama=default_response_synthesizer,
     )

backend/chain.py

@@ -222,7 +222,7 @@ def cohere_response_synthesizer(input: dict) -> RunnableSequence:
         default_response_synthesizer.configurable_alternatives(
             ConfigurableField("llm"),
             default_key="openai_gpt_3_5_turbo",
-            anthropic_claude_3_sonnet=default_response_synthesizer,
+            anthropic_claude_3_haiku=default_response_synthesizer,
             fireworks_mixtral=default_response_synthesizer,
             google_gemini_pro=default_response_synthesizer,
             cohere_command=cohere_response_synthesizer,
@@ -237,8 +237,8 @@ def cohere_response_synthesizer(input: dict) -> RunnableSequence:
 
 
 gpt_3_5 = ChatOpenAI(model="gpt-3.5-turbo-0125", temperature=0, streaming=True)
-claude_3_sonnet = ChatAnthropic(
-    model="claude-3-sonnet-20240229",
+claude_3_haiku = ChatAnthropic(
+    model="claude-3-haiku-20240307",
     temperature=0,
     max_tokens=4096,
     anthropic_api_key=os.environ.get("ANTHROPIC_API_KEY", "not_provided"),
@@ -266,12 +266,12 @@ def cohere_response_synthesizer(input: dict) -> RunnableSequence:
     # When configuring the end runnable, we can then use this id to configure this field
     ConfigurableField(id="llm"),
     default_key="openai_gpt_3_5_turbo",
-    anthropic_claude_3_sonnet=claude_3_sonnet,
+    anthropic_claude_3_haiku=claude_3_haiku,
     fireworks_mixtral=fireworks_mixtral,
     google_gemini_pro=gemini_pro,
     cohere_command=cohere_command,
 ).with_fallbacks(
-    [gpt_3_5, claude_3_sonnet, fireworks_mixtral, gemini_pro, cohere_command]
+    [gpt_3_5, claude_3_haiku, fireworks_mixtral, gemini_pro, cohere_command]
 )
 
 retriever = get_retriever()

frontend/app/components/ChatWindow.tsx

@@ -29,7 +29,7 @@ import { apiBaseUrl } from "../utils/constants";
 
 const MODEL_TYPES = [
   "openai_gpt_3_5_turbo",
-  "anthropic_claude_3_sonnet",
+  "anthropic_claude_3_haiku",
   "google_gemini_pro",
   "fireworks_mixtral",
   "cohere_command",
@@ -255,7 +255,7 @@ export function ChatWindow(props: { conversationId: string }) {
                 width={"240px"}
               >
                 <option value="openai_gpt_3_5_turbo">GPT-3.5-Turbo</option>
-                <option value="anthropic_claude_3_sonnet">Claude 3 Sonnet</option>
+                <option value="anthropic_claude_3_haiku">Claude 3 Haiku</option>
                 <option value="google_gemini_pro">Google Gemini Pro</option>
                 <option value="fireworks_mixtral">
                   Mixtral (via Fireworks.ai)