mirror of
https://github.com/katanemo/plano.git
synced 2026-05-24 14:05:14 +02:00
Pass tool call and app function response back in metadata (#193)
This commit is contained in:
parent
62a000036e
commit
dd1c7be706
8 changed files with 169 additions and 112 deletions
|
|
@ -2,14 +2,21 @@ import json
|
|||
import os
|
||||
from openai import OpenAI, DefaultHttpxClient
|
||||
import gradio as gr
|
||||
import logging as log
|
||||
import logging
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
|
||||
logging.basicConfig(
|
||||
level=logging.INFO,
|
||||
format="%(asctime)s - %(levelname)s - %(message)s",
|
||||
)
|
||||
|
||||
log = logging.getLogger(__name__)
|
||||
|
||||
CHAT_COMPLETION_ENDPOINT = os.getenv("CHAT_COMPLETION_ENDPOINT")
|
||||
ARCH_STATE_HEADER = "x-arch-state"
|
||||
log.info("CHAT_COMPLETION_ENDPOINT: ", CHAT_COMPLETION_ENDPOINT)
|
||||
log.info(f"CHAT_COMPLETION_ENDPOINT: {CHAT_COMPLETION_ENDPOINT}")
|
||||
|
||||
client = OpenAI(
|
||||
api_key="--",
|
||||
|
|
@ -23,23 +30,19 @@ def predict(message, state):
|
|||
state["history"] = []
|
||||
history = state.get("history")
|
||||
history.append({"role": "user", "content": message})
|
||||
log.info("history: ", history)
|
||||
log.info(f"history: {history}")
|
||||
|
||||
# Custom headers
|
||||
custom_headers = {
|
||||
"x-arch-deterministic-provider": "openai",
|
||||
}
|
||||
|
||||
metadata = None
|
||||
if "arch_state" in state:
|
||||
metadata = {ARCH_STATE_HEADER: state["arch_state"]}
|
||||
|
||||
try:
|
||||
raw_response = client.chat.completions.with_raw_response.create(
|
||||
model="--",
|
||||
messages=history,
|
||||
temperature=1.0,
|
||||
metadata=metadata,
|
||||
# metadata=metadata,
|
||||
extra_headers=custom_headers,
|
||||
)
|
||||
except Exception as e:
|
||||
|
|
@ -49,26 +52,35 @@ def predict(message, state):
|
|||
log.info("Error calling gateway API: {}".format(e.message))
|
||||
raise gr.Error("Error calling gateway API: {}".format(e.message))
|
||||
|
||||
log.info("raw_response: ", raw_response.text)
|
||||
log.error(f"raw_response: {raw_response.text}")
|
||||
response = raw_response.parse()
|
||||
|
||||
# extract arch_state from metadata and store it in gradio session state
|
||||
# this state must be passed back to the gateway in the next request
|
||||
response_json = json.loads(raw_response.text)
|
||||
arch_state = None
|
||||
if response_json:
|
||||
metadata = response_json.get("metadata", {})
|
||||
if metadata:
|
||||
arch_state = metadata.get(ARCH_STATE_HEADER, None)
|
||||
if arch_state:
|
||||
state["arch_state"] = arch_state
|
||||
# load arch_state from metadata
|
||||
arch_state_str = response_json.get("metadata", {}).get(ARCH_STATE_HEADER, "{}")
|
||||
# parse arch_state into json object
|
||||
arch_state = json.loads(arch_state_str)
|
||||
# load messages from arch_state
|
||||
arch_messages_str = arch_state.get("messages", "[]")
|
||||
# parse messages into json object
|
||||
arch_messages = json.loads(arch_messages_str)
|
||||
# append messages from arch gateway to history
|
||||
for message in arch_messages:
|
||||
history.append(message)
|
||||
|
||||
content = response.choices[0].message.content
|
||||
|
||||
history.append({"role": "assistant", "content": content, "model": response.model})
|
||||
|
||||
# for gradio UI we don't want to show raw tool calls and messages from developer application
|
||||
# so we're filtering those out
|
||||
history_view = [h for h in history if h["role"] != "tool" and "content" in h]
|
||||
messages = [
|
||||
(history[i]["content"], history[i + 1]["content"])
|
||||
for i in range(0, len(history) - 1, 2)
|
||||
(history_view[i]["content"], history_view[i + 1]["content"])
|
||||
for i in range(0, len(history_view) - 1, 2)
|
||||
]
|
||||
return messages, state
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue