98 lines
3.8 KiB
Python
98 lines
3.8 KiB
Python
from aiohttp import web
|
|
import json
|
|
import asyncio
|
|
|
|
from ..web_agent import WebAgent
|
|
from ..web_io_buffer import WebIOBuffer
|
|
from .util import wrap_async
|
|
|
|
class Api:
|
|
def __init__(self, app: web.Application, agent: WebAgent, io_buffer: WebIOBuffer):
|
|
self._app = app
|
|
self._agent = agent
|
|
self._io_buffer = io_buffer
|
|
|
|
self._init_routes()
|
|
|
|
def _init_routes(self):
|
|
"""Initialize REST API and WebSocket routes."""
|
|
self._app.router.add_post("/api/inference/{llm}", self._run_inference)
|
|
self._app.router.add_post("/api/approve/{llm}", self._approve_response)
|
|
self._app.router.add_post("/api/context", self._modify_context)
|
|
self._app.router.add_post("/api/input", self._send_input)
|
|
self._app.router.add_post("/api/clear", self._clear_output)
|
|
self._app.router.add_get("/api/output/{llm}", self._get_output)
|
|
self._app.router.add_get("/api/llms", self._get_llms)
|
|
|
|
@property
|
|
def app(self):
|
|
return self._app
|
|
|
|
async def _run_inference(self, request: web.Request) -> web.Response:
|
|
"""Start inference on specified LLM."""
|
|
llm_name = request.match_info["llm"]
|
|
try:
|
|
await asyncio.get_event_loop().run_in_executor(None, self._agent.run_inference, llm_name)
|
|
return web.Response(status=200)
|
|
except (ValueError, RuntimeError) as e:
|
|
return web.Response(status=400, text=str(e))
|
|
|
|
async def _approve_response(self, request: web.Request) -> web.Response:
|
|
"""Approve response from specified LLM."""
|
|
llm_name = request.match_info["llm"]
|
|
data = await request.json()
|
|
response = data.get("response")
|
|
if not response:
|
|
return web.Response(status=400, text="Missing response in request body")
|
|
try:
|
|
self._agent.approve_response(llm_name, response)
|
|
return web.Response(status=200)
|
|
except ValueError as e:
|
|
return web.Response(status=400, text=str(e))
|
|
|
|
async def _modify_context(self, request: web.Request) -> web.Response:
|
|
"""Modify the current context."""
|
|
data = await request.json()
|
|
context = data.get("context")
|
|
if not context:
|
|
return web.Response(status=400, text="Missing context in request body")
|
|
self._agent.modify_context(context)
|
|
return web.Response(status=200)
|
|
|
|
async def _send_input(self, request: web.Request) -> web.Response:
|
|
"""Send input to the IO buffer."""
|
|
data = await request.json()
|
|
input_text = data.get("input")
|
|
if not input_text:
|
|
return web.Response(status=400, text="Missing input in request body")
|
|
self._io_buffer.append_stdin(input_text)
|
|
return web.Response(status=200)
|
|
|
|
async def _clear_output(self, request: web.Request) -> web.Response:
|
|
"""Clear the stdout buffer."""
|
|
self._io_buffer.clear_stdout()
|
|
return web.Response(status=200)
|
|
|
|
async def _get_output(self, request: web.Request) -> web.Response:
|
|
"""Get complete output for specified LLM."""
|
|
llm_name = request.match_info["llm"]
|
|
try:
|
|
output = self._agent.get_output(llm_name)
|
|
return web.Response(
|
|
text=json.dumps({"output": output}),
|
|
content_type="application/json"
|
|
)
|
|
except ValueError as e:
|
|
return web.Response(status=400, text=str(e))
|
|
|
|
async def _get_llms(self, request: web.Request) -> web.Response:
|
|
"""Get all LLMs and their current states."""
|
|
states = self._agent.llms
|
|
return web.Response(
|
|
text=json.dumps({
|
|
name: state.name
|
|
for name, state in states.items()
|
|
}),
|
|
content_type="application/json"
|
|
)
|