mirror of
https://github.com/hanzoai/trolley-bench.git
synced 2026-10-30 15:00:49 +00:00
Zachary Landes
247bd2614b
* Packaged the benchmark's working copy with its frozen prompts, runners, raw answers, analysis and built page as committed evidence * Rewrote run-machine paths and left out share credentials, drafts and operator notes; PACKAGING.md lists every exclusion and change * Fixed the models.yaml generator listing rerun models twice, and let the Jev runner take its key from TYPESAFE_API_KEY
32 lines
1.2 KiB
Python
32 lines
1.2 KiB
Python
"""A /v1/systemone front for JevK5's documented llama.cpp path (jevk5.gguf.JevK5GGUF), on 127.0.0.1.
|
|
|
|
JevK5 reads Nouls natively; this only turns one request into one JevK5GGUF.decide call per
|
|
question, at the temperature the JevK5 README gives for jevk5-4b-v0.3-Q8_0.gguf (1.22, knockout 0.93).
|
|
Usage: python jevk5_server.py <port> <llama-server url> (with jevk5 importable)
|
|
"""
|
|
import json
|
|
import sys
|
|
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
|
|
|
|
from jevk5 import JevK5GGUF
|
|
|
|
model = JevK5GGUF(url=sys.argv[2], temperature=1.22, knockout_temperature=0.93)
|
|
|
|
|
|
class Handler(BaseHTTPRequestHandler):
|
|
def do_POST(self):
|
|
body = json.loads(self.rfile.read(int(self.headers["content-length"])))
|
|
answers = {name: model.decide(body["state"], q) for name, q in body["questions"].items()}
|
|
out = json.dumps({"answers": answers, "model": "JevK5 v0.3 4B, jevk5-4b-v0.3-Q8_0.gguf"}).encode()
|
|
self.send_response(200)
|
|
self.send_header("content-type", "application/json")
|
|
self.send_header("content-length", str(len(out)))
|
|
self.end_headers()
|
|
self.wfile.write(out)
|
|
|
|
def log_message(self, *a):
|
|
pass
|
|
|
|
|
|
ThreadingHTTPServer(("127.0.0.1", int(sys.argv[1])), Handler).serve_forever()
|