Spaces:
Sleeping
Sleeping
Update inference.py
Browse files- inference.py +17 -42
inference.py
CHANGED
|
@@ -4,15 +4,14 @@ import asyncio
|
|
| 4 |
import traceback
|
| 5 |
from dotenv import load_dotenv
|
| 6 |
from openai import OpenAI
|
|
|
|
| 7 |
from flask import Flask, jsonify, request
|
| 8 |
|
| 9 |
load_dotenv()
|
| 10 |
|
| 11 |
app = Flask(__name__)
|
| 12 |
env = None
|
| 13 |
-
event_loop = None
|
| 14 |
|
| 15 |
-
# --- LOGGING ---
|
| 16 |
def log_start(task, env_name, model):
|
| 17 |
print(f"[START] task={task} env={env_name} model={model}", flush=True)
|
| 18 |
|
|
@@ -26,19 +25,16 @@ def log_end(success, steps, score, rewards):
|
|
| 26 |
|
| 27 |
SYSTEM_PROMPT = "You are a Cloud Infrastructure AI. Manage the cluster by choosing actions: 0(Stay), 1(Add Server), 2(Remove), 3(Toggle Spot). Reply with ONLY the integer."
|
| 28 |
|
| 29 |
-
def
|
| 30 |
-
global event_loop
|
| 31 |
try:
|
| 32 |
loop = asyncio.get_event_loop()
|
| 33 |
if loop.is_closed():
|
| 34 |
-
raise RuntimeError
|
| 35 |
-
return loop
|
| 36 |
except Exception:
|
| 37 |
-
|
| 38 |
-
asyncio.set_event_loop(
|
| 39 |
-
return
|
| 40 |
-
|
| 41 |
-
# --- ROUTES ---
|
| 42 |
|
| 43 |
@app.route('/health', methods=['GET'])
|
| 44 |
def health():
|
|
@@ -49,23 +45,13 @@ def reset():
|
|
| 49 |
global env
|
| 50 |
print("[DEBUG] /reset called", flush=True)
|
| 51 |
try:
|
| 52 |
-
# Step 1 - import environment
|
| 53 |
-
print("[DEBUG] importing CloudCostEnv...", flush=True)
|
| 54 |
-
from environment import CloudCostEnv
|
| 55 |
-
print("[DEBUG] CloudCostEnv imported OK", flush=True)
|
| 56 |
-
|
| 57 |
-
# Step 2 - create env
|
| 58 |
-
print("[DEBUG] creating env...", flush=True)
|
| 59 |
env = CloudCostEnv(task="medium")
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
print(f"[DEBUG] reset done, state type: {type(state)}", flush=True)
|
| 67 |
-
|
| 68 |
-
# Step 4 - serialize state
|
| 69 |
if hasattr(state, 'model_dump'):
|
| 70 |
state_data = state.model_dump()
|
| 71 |
elif hasattr(state, 'model_dump_json'):
|
|
@@ -75,12 +61,6 @@ def reset():
|
|
| 75 |
else:
|
| 76 |
state_data = str(state)
|
| 77 |
|
| 78 |
-
log_start(
|
| 79 |
-
task="daily_peaks",
|
| 80 |
-
env_name="CloudEnv-v1",
|
| 81 |
-
model=os.getenv("MODEL_NAME", "unknown")
|
| 82 |
-
)
|
| 83 |
-
|
| 84 |
print("[DEBUG] /reset success", flush=True)
|
| 85 |
return jsonify({"status": "ok", "state": state_data}), 200
|
| 86 |
|
|
@@ -96,18 +76,13 @@ def reset():
|
|
| 96 |
@app.route('/step', methods=['POST'])
|
| 97 |
def step():
|
| 98 |
global env
|
| 99 |
-
print("[DEBUG] /step called", flush=True)
|
| 100 |
try:
|
| 101 |
if env is None:
|
| 102 |
-
return jsonify({
|
| 103 |
-
"status": "error",
|
| 104 |
-
"message": "Call /reset first"
|
| 105 |
-
}), 400
|
| 106 |
|
| 107 |
data = request.get_json()
|
| 108 |
action = int(data.get("action", 0))
|
| 109 |
-
|
| 110 |
-
state, reward, done = loop.run_until_complete(env.step(action))
|
| 111 |
log_step(step=1, action=str(action), reward=reward, done=done)
|
| 112 |
|
| 113 |
if hasattr(state, 'model_dump'):
|
|
@@ -135,5 +110,5 @@ def step():
|
|
| 135 |
}), 500
|
| 136 |
|
| 137 |
if __name__ == "__main__":
|
| 138 |
-
print("[BOOT]
|
| 139 |
-
app.run(host="0.0.0.0", port=
|
|
|
|
| 4 |
import traceback
|
| 5 |
from dotenv import load_dotenv
|
| 6 |
from openai import OpenAI
|
| 7 |
+
from environment import CloudCostEnv
|
| 8 |
from flask import Flask, jsonify, request
|
| 9 |
|
| 10 |
load_dotenv()
|
| 11 |
|
| 12 |
app = Flask(__name__)
|
| 13 |
env = None
|
|
|
|
| 14 |
|
|
|
|
| 15 |
def log_start(task, env_name, model):
|
| 16 |
print(f"[START] task={task} env={env_name} model={model}", flush=True)
|
| 17 |
|
|
|
|
| 25 |
|
| 26 |
SYSTEM_PROMPT = "You are a Cloud Infrastructure AI. Manage the cluster by choosing actions: 0(Stay), 1(Add Server), 2(Remove), 3(Toggle Spot). Reply with ONLY the integer."
|
| 27 |
|
| 28 |
+
def run_async(coro):
|
|
|
|
| 29 |
try:
|
| 30 |
loop = asyncio.get_event_loop()
|
| 31 |
if loop.is_closed():
|
| 32 |
+
raise RuntimeError
|
| 33 |
+
return loop.run_until_complete(coro)
|
| 34 |
except Exception:
|
| 35 |
+
loop = asyncio.new_event_loop()
|
| 36 |
+
asyncio.set_event_loop(loop)
|
| 37 |
+
return loop.run_until_complete(coro)
|
|
|
|
|
|
|
| 38 |
|
| 39 |
@app.route('/health', methods=['GET'])
|
| 40 |
def health():
|
|
|
|
| 45 |
global env
|
| 46 |
print("[DEBUG] /reset called", flush=True)
|
| 47 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
env = CloudCostEnv(task="medium")
|
| 49 |
+
state = run_async(env.reset())
|
| 50 |
+
log_start(
|
| 51 |
+
task="daily_peaks",
|
| 52 |
+
env_name="CloudEnv-v1",
|
| 53 |
+
model=os.getenv("MODEL_NAME", "unknown")
|
| 54 |
+
)
|
|
|
|
|
|
|
|
|
|
| 55 |
if hasattr(state, 'model_dump'):
|
| 56 |
state_data = state.model_dump()
|
| 57 |
elif hasattr(state, 'model_dump_json'):
|
|
|
|
| 61 |
else:
|
| 62 |
state_data = str(state)
|
| 63 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
print("[DEBUG] /reset success", flush=True)
|
| 65 |
return jsonify({"status": "ok", "state": state_data}), 200
|
| 66 |
|
|
|
|
| 76 |
@app.route('/step', methods=['POST'])
|
| 77 |
def step():
|
| 78 |
global env
|
|
|
|
| 79 |
try:
|
| 80 |
if env is None:
|
| 81 |
+
return jsonify({"status": "error", "message": "Call /reset first"}), 400
|
|
|
|
|
|
|
|
|
|
| 82 |
|
| 83 |
data = request.get_json()
|
| 84 |
action = int(data.get("action", 0))
|
| 85 |
+
state, reward, done = run_async(env.step(action))
|
|
|
|
| 86 |
log_step(step=1, action=str(action), reward=reward, done=done)
|
| 87 |
|
| 88 |
if hasattr(state, 'model_dump'):
|
|
|
|
| 110 |
}), 500
|
| 111 |
|
| 112 |
if __name__ == "__main__":
|
| 113 |
+
print("[BOOT] Flask starting on port 8080", flush=True)
|
| 114 |
+
app.run(host="0.0.0.0", port=8080, debug=False)
|