Commit
·
d74feed
1
Parent(s):
3cce116
app.py
CHANGED
|
@@ -1,259 +1,90 @@
|
|
| 1 |
-
#
|
| 2 |
-
import
|
| 3 |
-
import
|
| 4 |
-
import logging
|
| 5 |
-
from typing import Any, Dict, List, Optional, Tuple
|
| 6 |
|
| 7 |
-
import numpy as np
|
| 8 |
-
|
| 9 |
-
from backends_base import ChatBackend, ImagesBackend # ChatBackend for OA server
|
| 10 |
from config import settings
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
try:
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
device: 'cpu' or 'cuda' (passed to TimesFm if supported by installed lib)
|
| 85 |
-
"""
|
| 86 |
-
self.model_id = model_id or "google/timesfm-2.5-200m-pytorch"
|
| 87 |
-
self.device = device or "cpu"
|
| 88 |
-
self._model = None # lazy init
|
| 89 |
-
|
| 90 |
-
# ---------- internal ----------
|
| 91 |
-
def _ensure_model(self):
|
| 92 |
-
if self._model is not None or not _TIMESFM_AVAILABLE:
|
| 93 |
-
return
|
| 94 |
-
try:
|
| 95 |
-
# minimal init; adjust kwargs if your installed version needs different args
|
| 96 |
-
self._model = TimesFm() # type: ignore
|
| 97 |
-
logger.info("TimesFM model initialized.")
|
| 98 |
-
except Exception as e:
|
| 99 |
-
logger.exception("Failed to initialize TimesFM; will use fallback. %s", e)
|
| 100 |
-
self._model = None
|
| 101 |
-
|
| 102 |
-
# ---------- public helpers ----------
|
| 103 |
-
async def forecast(self, payload: Dict[str, Any]) -> Dict[str, Any]:
|
| 104 |
-
"""
|
| 105 |
-
Unified forecast entrypoint.
|
| 106 |
-
Expected keys (directly in payload OR nested under 'data' OR 'timeseries'):
|
| 107 |
-
- series: list of numbers (or list of dicts holding 'y'/'value')
|
| 108 |
-
- horizon: int (>0)
|
| 109 |
-
- freq: optional string for metadata only
|
| 110 |
-
Returns:
|
| 111 |
-
{
|
| 112 |
-
"model": "...",
|
| 113 |
-
"horizon": int,
|
| 114 |
-
"freq": str|None,
|
| 115 |
-
"forecast": [floats],
|
| 116 |
-
"note": str|None
|
| 117 |
-
}
|
| 118 |
-
"""
|
| 119 |
-
# unwrap if nested
|
| 120 |
-
if "data" in payload and isinstance(payload["data"], dict):
|
| 121 |
-
payload = {**payload, **payload["data"]}
|
| 122 |
-
if "timeseries" in payload and isinstance(payload["timeseries"], dict):
|
| 123 |
-
payload = {**payload, **payload["timeseries"]}
|
| 124 |
-
|
| 125 |
-
series = payload.get("series")
|
| 126 |
-
horizon = int(payload.get("horizon", 0))
|
| 127 |
-
freq = payload.get("freq")
|
| 128 |
-
|
| 129 |
-
y = _parse_series(series)
|
| 130 |
-
if horizon <= 0:
|
| 131 |
-
raise ValueError("horizon must be a positive integer")
|
| 132 |
-
|
| 133 |
-
self._ensure_model()
|
| 134 |
-
|
| 135 |
-
if _TIMESFM_AVAILABLE and self._model is not None:
|
| 136 |
-
# Use real TimesFM
|
| 137 |
-
try:
|
| 138 |
-
# Most TimesFM APIs are batch-oriented; we add a batch dim and remove it later
|
| 139 |
-
# If your installed version differs (e.g., .predict with signature),
|
| 140 |
-
# change these two lines accordingly:
|
| 141 |
-
y_batch = y[None, :]
|
| 142 |
-
preds = self._model.predict(y_batch, horizon=horizon) # type: ignore
|
| 143 |
-
# preds shape => (1, horizon)
|
| 144 |
-
fc = np.asarray(preds).reshape(-1).tolist()
|
| 145 |
-
note = None
|
| 146 |
-
except Exception as e:
|
| 147 |
-
logger.exception("TimesFM predict failed; falling back. %s", e)
|
| 148 |
-
fc = _fallback_forecast(y, horizon).tolist()
|
| 149 |
-
note = "fallback_used_due_to_predict_error"
|
| 150 |
-
else:
|
| 151 |
-
# Fallback path
|
| 152 |
-
fc = _fallback_forecast(y, horizon).tolist()
|
| 153 |
-
note = "fallback_used_timesfm_missing"
|
| 154 |
-
|
| 155 |
-
return {
|
| 156 |
-
"model": self.model_id,
|
| 157 |
-
"horizon": horizon,
|
| 158 |
-
"freq": freq,
|
| 159 |
-
"forecast": fc,
|
| 160 |
-
"note": note,
|
| 161 |
-
}
|
| 162 |
-
|
| 163 |
-
# ---------- ChatBackend interface (for oa_server) ----------
|
| 164 |
-
async def stream(self, request: Dict[str, Any]):
|
| 165 |
-
"""
|
| 166 |
-
OA-compatible streaming shim:
|
| 167 |
-
- Extracts forecast inputs from request (or from last user message JSON).
|
| 168 |
-
- Runs forecast() and yields ONE OpenAI-style chat chunk whose content
|
| 169 |
-
is a compact JSON string with the forecast result.
|
| 170 |
-
"""
|
| 171 |
-
rid = f"chatcmpl-timesfm-{int(time.time())}"
|
| 172 |
-
now = int(time.time())
|
| 173 |
-
|
| 174 |
-
# try to gather payload
|
| 175 |
-
payload: Dict[str, Any] = {}
|
| 176 |
-
|
| 177 |
-
# 1) allow direct shape: {series, horizon, ...} / or under 'data'/'timeseries'
|
| 178 |
-
if isinstance(request, dict):
|
| 179 |
-
payload = dict(request) # shallow copy
|
| 180 |
-
|
| 181 |
-
# 2) optionally parse last user message if it's JSON
|
| 182 |
-
try:
|
| 183 |
-
msgs = request.get("messages") if isinstance(request, dict) else None
|
| 184 |
-
if isinstance(msgs, list) and msgs:
|
| 185 |
-
for m in reversed(msgs):
|
| 186 |
-
if isinstance(m, dict) and m.get("role") == "user":
|
| 187 |
-
c = m.get("content")
|
| 188 |
-
if isinstance(c, str):
|
| 189 |
-
c_str = c.strip()
|
| 190 |
-
if (c_str.startswith("{") and c_str.endswith("}")) or (
|
| 191 |
-
c_str.startswith("[") and c_str.endswith("]")
|
| 192 |
-
):
|
| 193 |
-
# try parse JSON content
|
| 194 |
-
parsed = json.loads(c_str)
|
| 195 |
-
if isinstance(parsed, dict):
|
| 196 |
-
payload.update(parsed)
|
| 197 |
-
break
|
| 198 |
-
except Exception:
|
| 199 |
-
# non-fatal: keep whatever we had
|
| 200 |
-
pass
|
| 201 |
-
|
| 202 |
-
# run forecast
|
| 203 |
-
try:
|
| 204 |
-
result = await self.forecast(payload)
|
| 205 |
-
except Exception as e:
|
| 206 |
-
# return an error chunk in OpenAI shape
|
| 207 |
-
err = {"error": str(e)}
|
| 208 |
-
content = json.dumps(err, separators=(",", ":"), ensure_ascii=False)
|
| 209 |
-
yield {
|
| 210 |
-
"id": rid,
|
| 211 |
-
"object": "chat.completion.chunk",
|
| 212 |
-
"created": now,
|
| 213 |
-
"model": self.model_id,
|
| 214 |
-
"choices": [
|
| 215 |
-
{
|
| 216 |
-
"index": 0,
|
| 217 |
-
"delta": {"role": "assistant", "content": content},
|
| 218 |
-
"finish_reason": "stop",
|
| 219 |
-
}
|
| 220 |
-
],
|
| 221 |
-
}
|
| 222 |
-
return
|
| 223 |
-
|
| 224 |
-
# success: compact JSON content so your .NET can parse
|
| 225 |
-
content = json.dumps(
|
| 226 |
-
{
|
| 227 |
-
"model": result.get("model"),
|
| 228 |
-
"horizon": result.get("horizon"),
|
| 229 |
-
"freq": result.get("freq"),
|
| 230 |
-
"forecast": result.get("forecast"),
|
| 231 |
-
"note": result.get("note"),
|
| 232 |
-
"backend": "timesfm",
|
| 233 |
-
},
|
| 234 |
-
separators=(",", ":"),
|
| 235 |
-
ensure_ascii=False,
|
| 236 |
-
)
|
| 237 |
-
|
| 238 |
-
yield {
|
| 239 |
-
"id": rid,
|
| 240 |
-
"object": "chat.completion.chunk",
|
| 241 |
-
"created": now,
|
| 242 |
-
"model": self.model_id,
|
| 243 |
-
"choices": [
|
| 244 |
-
{
|
| 245 |
-
"index": 0,
|
| 246 |
-
"delta": {"role": "assistant", "content": content},
|
| 247 |
-
"finish_reason": "stop",
|
| 248 |
-
}
|
| 249 |
-
],
|
| 250 |
-
}
|
| 251 |
-
|
| 252 |
-
|
| 253 |
-
# Optional: keep an images stub to satisfy oa_server wiring if needed elsewhere
|
| 254 |
-
class StubImagesBackend(ImagesBackend):
|
| 255 |
-
async def generate_b64(self, request: Dict[str, Any]) -> str:
|
| 256 |
-
logger.warning("Image generation not supported in TimesFM backend.")
|
| 257 |
-
return (
|
| 258 |
-
"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR4nGP4BwQACfsD/etCJH0AAAAASUVORK5CYII="
|
| 259 |
-
)
|
|
|
|
| 1 |
+
# app.py
|
| 2 |
+
import asyncio, logging
|
| 3 |
+
import gradio as gr
|
|
|
|
|
|
|
| 4 |
|
|
|
|
|
|
|
|
|
|
| 5 |
from config import settings
|
| 6 |
+
from rabbit_base import RabbitBase
|
| 7 |
+
from listener import RabbitListenerBase
|
| 8 |
+
from rabbit_repo import RabbitRepo
|
| 9 |
+
from oa_server import OpenAIServers
|
| 10 |
+
#from vllm_backend import VLLMChatBackend, StubImagesBackend
|
| 11 |
+
#from transformers_backend import TransformersChatBackend, StubImagesBackend
|
| 12 |
+
#from hf_backend import HFChatBackend, StubImagesBackend
|
| 13 |
+
from hf_backend import StubImagesBackend
|
| 14 |
+
from timesfm_backend import TimesFMBackend
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
logging.basicConfig(
|
| 18 |
+
level=logging.INFO,
|
| 19 |
+
format="%(asctime)s [%(levelname)s] %(name)s: %(message)s"
|
| 20 |
+
)
|
| 21 |
+
log = logging.getLogger("app")
|
| 22 |
+
|
| 23 |
+
# ----------------- Hugging Face Spaces helpers -----------------
|
| 24 |
try:
|
| 25 |
+
import spaces
|
| 26 |
+
|
| 27 |
+
@spaces.GPU(duration=60)
|
| 28 |
+
def gpu_entrypoint() -> str:
|
| 29 |
+
return "gpu: ready"
|
| 30 |
+
|
| 31 |
+
except Exception:
|
| 32 |
+
def gpu_entrypoint() -> str:
|
| 33 |
+
return "gpu: not available (CPU only)"
|
| 34 |
+
|
| 35 |
+
# ----------------- RabbitMQ wiring -----------------
|
| 36 |
+
publisher = RabbitRepo(external_source="openai.mq.server")
|
| 37 |
+
resolver = (lambda name: "direct" if name.startswith("oa.") else settings.RABBIT_EXCHANGE_TYPE)
|
| 38 |
+
base = RabbitBase(exchange_type_resolver=resolver)
|
| 39 |
+
|
| 40 |
+
servers = OpenAIServers(
|
| 41 |
+
publisher,
|
| 42 |
+
chat_backend=TimesFMBackend(),
|
| 43 |
+
images_backend=StubImagesBackend()
|
| 44 |
+
)
|
| 45 |
+
|
| 46 |
+
handlers = {
|
| 47 |
+
"oaChatCreate": servers.handle_chat_create,
|
| 48 |
+
"oaImagesGenerate": servers.handle_images_generate,
|
| 49 |
+
}
|
| 50 |
+
|
| 51 |
+
DECLS = [
|
| 52 |
+
{"ExchangeName": "oa.chat.create", "FuncName": "oaChatCreate",
|
| 53 |
+
"MessageTimeout": 600_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
|
| 54 |
+
{"ExchangeName": "oa.images.generate", "FuncName": "oaImagesGenerate",
|
| 55 |
+
"MessageTimeout": 600_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
|
| 56 |
+
]
|
| 57 |
+
|
| 58 |
+
listener = RabbitListenerBase(base, instance_name=settings.RABBIT_INSTANCE_NAME, handlers=handlers)
|
| 59 |
+
|
| 60 |
+
# ----------------- Startup init -----------------
|
| 61 |
+
async def _startup_init():
|
| 62 |
+
try:
|
| 63 |
+
await base.connect() # connect to RabbitMQ
|
| 64 |
+
await listener.start(DECLS) # start queue listeners
|
| 65 |
+
return "OpenAI MQ + vLLM: ready"
|
| 66 |
+
except Exception as e:
|
| 67 |
+
log.exception("Startup init failed")
|
| 68 |
+
return f"ERROR: {e}"
|
| 69 |
+
|
| 70 |
+
async def ping():
|
| 71 |
+
return "ok"
|
| 72 |
+
|
| 73 |
+
# ----------------- Gradio UI -----------------
|
| 74 |
+
with gr.Blocks(title="OpenAI over RabbitMQ (local vLLM)", theme=gr.themes.Soft()) as demo:
|
| 75 |
+
gr.Markdown("## OpenAI-compatible over RabbitMQ — using vLLM locally inside Space")
|
| 76 |
+
with gr.Tabs():
|
| 77 |
+
with gr.Tab("Service"):
|
| 78 |
+
btn = gr.Button("Ping")
|
| 79 |
+
out = gr.Textbox(label="Ping result")
|
| 80 |
+
btn.click(ping, inputs=None, outputs=out)
|
| 81 |
+
init_status = gr.Textbox(label="Startup status", interactive=False)
|
| 82 |
+
demo.load(fn=_startup_init, inputs=None, outputs=init_status)
|
| 83 |
+
|
| 84 |
+
with gr.Tab("@spaces.GPU Probe"):
|
| 85 |
+
gpu_btn = gr.Button("GPU Ready Probe", variant="primary")
|
| 86 |
+
gpu_out = gr.Textbox(label="GPU Probe Result", interactive=False)
|
| 87 |
+
gpu_btn.click(gpu_entrypoint, inputs=None, outputs=gpu_out)
|
| 88 |
+
|
| 89 |
+
if __name__ == "__main__":
|
| 90 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True, debug=True, mcp_server=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|