Merge branch 'main' into deploy

.github/workflows/spelling.yml

@@ -0,0 +1,10 @@
+name: Spell Check
+
+on: ["push", "pull_request"]
+
+jobs:
+  typos:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: crate-ci/typos@master

src/logic/tools/fetch.py

@@ -14,7 +14,7 @@
 
 
 class Browser(AbstractTool):
-    """Fetch an url. The only paremeter `url`. URLs on the page can be used for reference. If you want to know any realworld fact, you should search Google."""
+    """Fetch an url. The only parameter `url`. URLs on the page can be used for reference. If you want to know any realworld fact, you should search Google."""
 
     name = "fetch"
 

src/routes/run.py

@@ -4,7 +4,7 @@
 from typing import Annotated, Literal, cast
 
 from fastapi import APIRouter, Depends, Request
-from fastapi.responses import PlainTextResponse, StreamingResponse
+from fastapi.responses import PlainTextResponse
 from promplate import Node
 from pydantic import BaseModel, Field
 
@@ -13,6 +13,7 @@
 from ..utils.config import env
 from ..utils.http import forward_headers
 from ..utils.llm import Model, find_llm, groq, openai
+from ..utils.response import make_response
 from .sse import non_duplicated_event_stream
 
 run_router = APIRouter(tags=["call"])
@@ -72,7 +73,7 @@ async def make_stream():
             print_exc(file=stderr)
             yield str(e), "error"
 
-    return StreamingResponse(make_stream(), media_type="text/event-stream")
+    return await make_response(make_stream(), "text/event-stream")
 
 
 @run_router.put(f"{env.base}/single/{{template}}")
@@ -95,4 +96,4 @@ async def make_stream():
                 yield cast(str, c.result).removeprefix(last)
                 last = c.result
 
-    return StreamingResponse(make_stream(), media_type="text/plain")
+    return await make_response(make_stream())

src/templates/schema/code_interpreter.md

@@ -4,7 +4,7 @@ There is only 1 parameters:
 
 - `code: str`: the python code to `exec()`
 
-For example, if you want to caculate 0.409 \* 2023, you can make a call like this:
+For example, if you want to calculate 0.409 \* 2023, you can make a call like this:
 
 ```json
 {

src/utils/llm/__init__.py

@@ -24,6 +24,7 @@
     "claude-3-sonnet-20240229",
     "claude-3-opus-20240229",
     "gemma-7b-it",
+    "gemma2-9b-it",
     "llama3-8b-8192",
     "llama3-70b-8192",
     "llama2-70b-4096",

src/utils/response.py

@@ -0,0 +1,23 @@
+from operator import call
+from traceback import print_exc
+
+from fastapi.responses import PlainTextResponse, StreamingResponse
+from starlette.responses import AsyncContentStream
+
+
+async def make_response(stream: AsyncContentStream, media_type="text/plain"):
+    it = aiter(stream)
+
+    try:
+        first_chunk = await anext(it)
+    except Exception as e:
+        print_exc()
+        return PlainTextResponse("\n".join(map(str, e.args)).strip(), getattr(e, "status_code", 500))
+
+    @call
+    async def _():
+        yield first_chunk
+        async for i in it:
+            yield i
+
+    return StreamingResponse(_, media_type=media_type)