Spaces:
Sleeping
Sleeping
Update inference.py
Browse files- inference.py +73 -43
inference.py
CHANGED
|
@@ -1,19 +1,20 @@
|
|
| 1 |
import os
|
| 2 |
-
import asyncio
|
| 3 |
import json
|
|
|
|
| 4 |
from dotenv import load_dotenv
|
| 5 |
from openai import OpenAI
|
| 6 |
from environment import CloudCostEnv
|
| 7 |
-
from flask import Flask, jsonify, request
|
| 8 |
|
| 9 |
load_dotenv()
|
| 10 |
|
| 11 |
-
app = Flask(__name__)
|
| 12 |
-
env = None
|
|
|
|
| 13 |
|
| 14 |
# --- MANDATORY LOGGING FUNCTIONS (Don't Change) ---
|
| 15 |
-
def log_start(task: str,
|
| 16 |
-
print(f"[START] task={task} env={
|
| 17 |
|
| 18 |
def log_step(step: int, action: str, reward: float, done: bool, error=None) -> None:
|
| 19 |
error_val = error if error else "null"
|
|
@@ -27,7 +28,20 @@ def log_end(success: bool, steps: int, score: float, rewards: list) -> None:
|
|
| 27 |
# --- SYSTEM PROMPT ---
|
| 28 |
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."
|
| 29 |
|
| 30 |
-
# ---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
async def get_action_from_llama(state):
|
| 32 |
api_base = os.getenv("API_BASE_URL")
|
| 33 |
api_key = os.getenv("HF_TOKEN")
|
|
@@ -50,52 +64,69 @@ async def get_action_from_llama(state):
|
|
| 50 |
except Exception as e:
|
| 51 |
return 0, f"Error: {str(e)}"
|
| 52 |
|
| 53 |
-
#
|
| 54 |
-
|
| 55 |
-
|
|
|
|
|
|
|
| 56 |
|
| 57 |
-
@app.route('/reset', methods=['POST'])
|
| 58 |
def reset():
|
| 59 |
global env
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
@app.route('/step', methods=['POST'])
|
| 69 |
def step():
|
| 70 |
global env
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
|
| 86 |
-
# ---
|
| 87 |
async def main():
|
| 88 |
-
api_base = os.getenv("API_BASE_URL")
|
| 89 |
-
api_key = os.getenv("HF_TOKEN")
|
| 90 |
model_name = os.getenv("MODEL_NAME")
|
| 91 |
-
|
| 92 |
task_id = "daily_peaks"
|
| 93 |
env_local = CloudCostEnv(task="medium")
|
| 94 |
-
log_start(task=task_id,
|
| 95 |
-
|
| 96 |
rewards = []
|
| 97 |
state = await env_local.reset()
|
| 98 |
-
|
| 99 |
try:
|
| 100 |
for step_num in range(1, 21):
|
| 101 |
action_int, action_text = await get_action_from_llama(state)
|
|
@@ -104,14 +135,13 @@ async def main():
|
|
| 104 |
rewards.append(reward)
|
| 105 |
if done:
|
| 106 |
break
|
| 107 |
-
|
| 108 |
avg_score = sum(rewards) / len(rewards) if rewards else 0
|
| 109 |
-
log_end(success=(avg_score >= 0.7), steps=len(rewards),
|
|
|
|
| 110 |
except Exception as e:
|
| 111 |
log_step(step=0, action="error", reward=0.0, done=True, error=str(e))
|
| 112 |
finally:
|
| 113 |
await env_local.close()
|
| 114 |
|
| 115 |
if __name__ == "__main__":
|
| 116 |
-
|
| 117 |
-
app.run(host="0.0.0.0", port=8080) # <-- runs the web server
|
|
|
|
| 1 |
import os
|
|
|
|
| 2 |
import json
|
| 3 |
+
import asyncio
|
| 4 |
from dotenv import load_dotenv
|
| 5 |
from openai import OpenAI
|
| 6 |
from environment import CloudCostEnv
|
| 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 |
# --- MANDATORY LOGGING FUNCTIONS (Don't Change) ---
|
| 16 |
+
def log_start(task: str, env_name: str, model: str) -> None:
|
| 17 |
+
print(f"[START] task={task} env={env_name} model={model}", flush=True)
|
| 18 |
|
| 19 |
def log_step(step: int, action: str, reward: float, done: bool, error=None) -> None:
|
| 20 |
error_val = error if error else "null"
|
|
|
|
| 28 |
# --- SYSTEM PROMPT ---
|
| 29 |
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."
|
| 30 |
|
| 31 |
+
# --- GET EVENT LOOP (fixes asyncio + Flask conflict) ---
|
| 32 |
+
def get_loop():
|
| 33 |
+
global event_loop
|
| 34 |
+
try:
|
| 35 |
+
if event_loop is None or event_loop.is_closed():
|
| 36 |
+
event_loop = asyncio.new_event_loop()
|
| 37 |
+
asyncio.set_event_loop(event_loop)
|
| 38 |
+
return event_loop
|
| 39 |
+
except Exception:
|
| 40 |
+
event_loop = asyncio.new_event_loop()
|
| 41 |
+
asyncio.set_event_loop(event_loop)
|
| 42 |
+
return event_loop
|
| 43 |
+
|
| 44 |
+
# --- LLM ACTION ---
|
| 45 |
async def get_action_from_llama(state):
|
| 46 |
api_base = os.getenv("API_BASE_URL")
|
| 47 |
api_key = os.getenv("HF_TOKEN")
|
|
|
|
| 64 |
except Exception as e:
|
| 65 |
return 0, f"Error: {str(e)}"
|
| 66 |
|
| 67 |
+
# --- ROUTES ---
|
| 68 |
+
|
| 69 |
+
@app.route('/health', methods=['GET'])
|
| 70 |
+
def health():
|
| 71 |
+
return jsonify({"status": "ok"}), 200
|
| 72 |
|
| 73 |
+
@app.route('/reset', methods=['POST'])
|
| 74 |
def reset():
|
| 75 |
global env
|
| 76 |
+
try:
|
| 77 |
+
loop = get_loop()
|
| 78 |
+
env = CloudCostEnv(task="medium")
|
| 79 |
+
state = loop.run_until_complete(env.reset())
|
| 80 |
+
log_start(
|
| 81 |
+
task="daily_peaks",
|
| 82 |
+
env_name="CloudEnv-v1",
|
| 83 |
+
model=os.getenv("MODEL_NAME", "unknown")
|
| 84 |
+
)
|
| 85 |
+
if hasattr(state, 'model_dump'):
|
| 86 |
+
state_data = state.model_dump()
|
| 87 |
+
elif hasattr(state, 'model_dump_json'):
|
| 88 |
+
state_data = json.loads(state.model_dump_json())
|
| 89 |
+
else:
|
| 90 |
+
state_data = str(state)
|
| 91 |
+
return jsonify({"status": "ok", "state": state_data}), 200
|
| 92 |
+
except Exception as e:
|
| 93 |
+
print(f"[ERROR] reset failed: {str(e)}", flush=True)
|
| 94 |
+
return jsonify({"status": "error", "message": str(e)}), 500
|
| 95 |
|
| 96 |
@app.route('/step', methods=['POST'])
|
| 97 |
def step():
|
| 98 |
global env
|
| 99 |
+
try:
|
| 100 |
+
if env is None:
|
| 101 |
+
return jsonify({"status": "error", "message": "Call /reset first"}), 400
|
| 102 |
+
data = request.get_json()
|
| 103 |
+
action = int(data.get("action", 0))
|
| 104 |
+
loop = get_loop()
|
| 105 |
+
state, reward, done = loop.run_until_complete(env.step(action))
|
| 106 |
+
log_step(step=1, action=str(action), reward=reward, done=done)
|
| 107 |
+
if hasattr(state, 'model_dump'):
|
| 108 |
+
state_data = state.model_dump()
|
| 109 |
+
elif hasattr(state, 'model_dump_json'):
|
| 110 |
+
state_data = json.loads(state.model_dump_json())
|
| 111 |
+
else:
|
| 112 |
+
state_data = str(state)
|
| 113 |
+
return jsonify({
|
| 114 |
+
"state": state_data,
|
| 115 |
+
"reward": reward,
|
| 116 |
+
"done": done
|
| 117 |
+
}), 200
|
| 118 |
+
except Exception as e:
|
| 119 |
+
print(f"[ERROR] step failed: {str(e)}", flush=True)
|
| 120 |
+
return jsonify({"status": "error", "message": str(e)}), 500
|
| 121 |
|
| 122 |
+
# --- TERMINAL TESTING (unchanged) ---
|
| 123 |
async def main():
|
|
|
|
|
|
|
| 124 |
model_name = os.getenv("MODEL_NAME")
|
|
|
|
| 125 |
task_id = "daily_peaks"
|
| 126 |
env_local = CloudCostEnv(task="medium")
|
| 127 |
+
log_start(task=task_id, env_name="CloudEnv-v1", model=model_name)
|
|
|
|
| 128 |
rewards = []
|
| 129 |
state = await env_local.reset()
|
|
|
|
| 130 |
try:
|
| 131 |
for step_num in range(1, 21):
|
| 132 |
action_int, action_text = await get_action_from_llama(state)
|
|
|
|
| 135 |
rewards.append(reward)
|
| 136 |
if done:
|
| 137 |
break
|
|
|
|
| 138 |
avg_score = sum(rewards) / len(rewards) if rewards else 0
|
| 139 |
+
log_end(success=(avg_score >= 0.7), steps=len(rewards),
|
| 140 |
+
score=avg_score, rewards=rewards)
|
| 141 |
except Exception as e:
|
| 142 |
log_step(step=0, action="error", reward=0.0, done=True, error=str(e))
|
| 143 |
finally:
|
| 144 |
await env_local.close()
|
| 145 |
|
| 146 |
if __name__ == "__main__":
|
| 147 |
+
app.run(host="0.0.0.0", port=7860, debug=False)
|
|
|