Spaces:
Running
Running
File size: 7,749 Bytes
0a2e8cf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 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 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 | """The HTTP surface: POST /v1/systemone (decisions), POST /v1/prewarm, GET /v1/version, GET /healthz, GET /readyz, GET /metrics,
POST /v1/chat/completions (passthrough for clients that also type text). Bearer auth on /v1/* when a token is configured."""
from __future__ import annotations
import hashlib
import json
import logging
import os
import time
import uuid
from contextlib import asynccontextmanager
from pathlib import Path
from typing import Any
from fastapi import APIRouter, Depends, FastAPI, HTTPException, Request, Response
from fastapi.responses import JSONResponse
from prometheus_client import CONTENT_TYPE_LATEST, Counter, Gauge, Histogram, generate_latest
from pydantic import BaseModel, Field, field_validator
from . import __version__
from .backends.base import BackendError
from .readout import BadQuestion, Readout, ReadoutIncomplete
log = logging.getLogger("openjev")
REQS = Counter("openjev_requests_total", "systemone requests", ["status"])
QUESTIONS = Counter("openjev_questions_total", "questions answered", ["type"])
TOKENS = Counter("openjev_prompt_tokens_total", "prompt tokens sent to the model")
LAT = Histogram("openjev_request_seconds", "systemone request latency", buckets=(0.05, 0.1, 0.2, 0.3, 0.5, 0.75, 1, 1.5, 2, 3, 5, 8, 13))
INFLIGHT = Gauge("openjev_inflight_requests", "requests in flight")
ANSWER_MODULES = ("prompt.py", "readout.py") # the code that determines answers; its hash is reported by /v1/version
class Question(BaseModel):
type: str
instructions: Any = None
criteria: Any = None
@field_validator("type")
@classmethod
def _type(cls, v):
if v not in ("choice", "score", "noul"):
raise ValueError("type must be one of choice, score, noul")
return v
class SystemOneRequest(BaseModel):
model: str | None = None
state: Any
questions: dict[str, Question] = Field(min_length=1)
class PrewarmRequest(BaseModel):
state: Any
def http_error(status: int, message: str) -> HTTPException:
return HTTPException(status, {"code": status, "message": message})
def version_info(readout: Readout, *, backend_name: str, model_dir: str, extra: dict | None) -> dict:
"""What GET /v1/version reports: the profile constants, the backend and a hash of the answer-determining code."""
p = readout.profile
here = Path(__file__).parent
code = b"".join((here / m).read_bytes() for m in ANSWER_MODULES)
return {
"server": "openjev-server",
"version": __version__,
"backend": backend_name,
"model_dir": os.path.basename(model_dir.rstrip("/")) or model_dir,
"profile": p.name,
"T": p.temp,
"noul_t": p.noul_t,
"noul_bias": p.noul_bias,
"flags": {
"perms": p.perms,
"stagger": p.stagger,
"compact": p.compact,
"compact_cap": p.compact_cap,
"layout": p.layout,
"pad": p.pad,
"targeted": True,
"instr_style": p.instr_style,
},
"readout_sha256": hashlib.sha256(code).hexdigest(),
**(extra or {}),
}
def bearer_auth(token: str):
"""A dependency that requires `Authorization: Bearer <token>` when a token is configured (no token: open)."""
async def auth(request: Request):
if token and request.headers.get("Authorization", "") != f"Bearer {token}":
raise http_error(401, "missing or invalid bearer token")
return auth
def ops_router(readout: Readout, backend_name: str) -> APIRouter:
"""Unauthenticated operations endpoints: liveness, readiness, Prometheus metrics."""
r = APIRouter()
@r.get("/healthz")
async def healthz():
return {"ok": True}
@r.get("/readyz")
async def readyz():
ok = await readout.backend.healthy()
return JSONResponse({"ok": ok, "backend": backend_name}, status_code=200 if ok else 503)
@r.get("/metrics")
async def metrics():
return Response(generate_latest(), media_type=CONTENT_TYPE_LATEST)
return r
def api_router(readout: Readout, version: dict, model_string: str, token: str) -> APIRouter:
"""The /v1 API: version, prewarm, systemone, chat passthrough."""
r = APIRouter(prefix="/v1", dependencies=[Depends(bearer_auth(token))])
@r.get("/version")
@r.post("/version")
async def get_version():
return version
@r.post("/prewarm")
async def prewarm(body: PrewarmRequest):
t0 = time.perf_counter()
try:
n = await readout.prewarm(body.state)
except (BackendError, ReadoutIncomplete) as e:
raise http_error(502, str(e)[:200]) from e
log.info(json.dumps({"prewarm_tokens": n, "ms": round((time.perf_counter() - t0) * 1000)}))
return {"ok": True, "prompt_tokens": n}
@r.post("/systemone")
async def systemone(body: SystemOneRequest):
rid = f"oj-{uuid.uuid4().hex[:16]}"
t0 = time.perf_counter()
INFLIGHT.inc()
try:
answers, toks = await readout.answer_all(body.state, {k: q.model_dump() for k, q in body.questions.items()})
except BadQuestion as e:
REQS.labels("422").inc()
raise http_error(422, str(e)) from e
except ReadoutIncomplete as e:
REQS.labels("502").inc()
raise http_error(502, f"readout incomplete: {e}") from e
except BackendError as e:
REQS.labels("502").inc()
raise http_error(502, str(e)[:300]) from e
finally:
INFLIGHT.dec()
record(rid, answers, toks, time.perf_counter() - t0)
return {"id": rid, "model": model_string, "answers": answers, "usage": {"input_tokens": toks, "output_tokens": 0}}
@r.post("/chat/completions")
async def chat(request: Request):
fn = getattr(readout.backend, "passthrough", None)
if fn is None:
raise http_error(501, "this backend has no chat passthrough")
code, data = await fn(await request.json())
return Response(data, status_code=code, media_type="application/json")
return r
def record(rid: str, answers: dict[str, dict], toks: int, dt: float) -> None:
"""Metrics and the one-line JSON log of a served request."""
LAT.observe(dt)
REQS.labels("200").inc()
TOKENS.inc(toks)
for a in answers.values():
QUESTIONS.labels(a["type"]).inc()
ops = {q: a.get("choice") for q, a in answers.items() if "choice" in a}
log.info(json.dumps({"id": rid, "questions": len(answers), "tokens": toks, "ms": round(dt * 1000), "ops": ops}))
def build_app(readout: Readout, *, token: str = "", model_dir: str = "", backend_name: str = "", extra_version: dict | None = None) -> FastAPI:
version = version_info(readout, backend_name=backend_name, model_dir=model_dir, extra=extra_version)
p = readout.profile
model_string = f"{version['model_dir']} T={p.temp} noul={p.noul_t},{p.noul_bias} profile={p.name} openjev-server@{__version__}"
@asynccontextmanager
async def lifespan(app: FastAPI):
yield
close = getattr(readout.backend, "aclose", None)
if close:
await close()
app = FastAPI(title="openjev-server", version=__version__, lifespan=lifespan, docs_url="/docs", redoc_url=None)
@app.exception_handler(HTTPException)
async def _http_exc(request, exc: HTTPException):
detail = exc.detail if isinstance(exc.detail, dict) else {"code": exc.status_code, "message": str(exc.detail)}
return JSONResponse({"error": detail}, status_code=exc.status_code)
app.include_router(ops_router(readout, backend_name))
app.include_router(api_router(readout, version, model_string, token))
return app
|