sickn33/agentic-awesome-skills / plugins/aasb-sybil-solutions/skills/sybil-glm53-offload-setup
sybil-glm53-offload-setup
UnverifiedSelect, verify and benchmark a GLM-5.3-Flash EXL3 offload mode on a single supported Linux NVIDIA host.
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 1.0 KB |
| SKILL.md | skill md | 3.8 KB |
| bench/score_ref_panel.py | script | 4.1 KB |
| bench/stream_rate.py | script | 3.4 KB |
| bench/sweep.py | script | 10.6 KB |
| docker/preflight.py | script | 5.5 KB |
| docs/how-it-works.md | markdown | 19.3 KB |
| docs/reference.md | markdown | 43.2 KB |
| docs/results.md | markdown | 11.4 KB |
| docs/setup.md | markdown | 16.7 KB |
| docs/troubleshooting.md | markdown | 5.2 KB |
| reference/glm-5.3-flash-exl3-ref-panel.json | config | 1.1 MB |
bench/stream_rate.py
View on GitHub#!/usr/bin/env python3
"""Streaming delivery check against a running server: the stream must arrive at the generation rate, not lag behind it
and flush at the end (the G072 defect in digest sha256:1159044a: ~14 tok/s delivered while ~29 tok/s were generated).
For each endpoint (OpenAI chat stream, SGLang /generate stream) one greedy long answer is streamed; with t = arrival time
of each content-bearing event and c = cumulative characters received:
half_share = share of the final text that has arrived by the midpoint between first and last event (uniform ~0.5;
the defect gives ~0.25)
tail_share = share of the final text that arrives in the last 1 s (the defect flushes ~50 % there)
PASS when half_share >= 0.40 and tail_share <= 0.10 on every endpoint. Exit code 0 = pass, 1 = fail.
python3 bench/stream_rate.py --url http://127.0.0.1:30000
"""
import argparse, json, sys, time, urllib.request
Q = "Write a long, detailed essay (at least 1200 words) on the history of the printing press."
def stream(url, path, payload):
req = urllib.request.Request(url + path, data=json.dumps(payload).encode(), headers={"Content-Type": "application/json"})
pts, total = [], 0
with urllib.request.urlopen(req, timeout=3600) as r:
for raw in r:
line = raw.decode().strip()
if not line.startswith("data:") or line[5:].strip() == "[DONE]":
continue
ev = json.loads(line[5:].strip())
if "choices" in ev: # OpenAI chunk: deltas
d = ev["choices"][0].get("delta") or {}
k = len(d.get("content") or "") + len(d.get("reasoning_content") or "")
total += k
else: # SGLang /generate: cumulative text
k = len(ev.get("text") or "") - total
total = len(ev.get("text") or "")
if k > 0:
pts.append((time.perf_counter(), total))
return pts
def score(pts):
t0, t1, n = pts[0][0], pts[-1][0], pts[-1][1]
mid = t0 + (t1 - t0) / 2
half = max([c for t, c in pts if t <= mid] or [0]) / n
before_tail = max([c for t, c in pts if t <= t1 - 1.0] or [0])
return {"events": len(pts), "chars": n, "span_s": round(t1 - t0, 2), "half_share": round(half, 3),
"tail_share": round(1 - before_tail / n, 3)}
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--url", default="http://127.0.0.1:30000")
ap.add_argument("--model", default="glm-5.3-flash")
a = ap.parse_args()
ok = True
chat = stream(a.url, "/v1/chat/completions", {"model": a.model, "stream": True, "temperature": 0,
"messages": [{"role": "user", "content": Q}], "chat_template_kwargs": {"enable_thinking": False}})
tok = json.loads(urllib.request.urlopen(urllib.request.Request(
a.url + "/tokenize", data=json.dumps({"prompt": f"[gMASK]<sop><|user|>{Q}<|assistant|><think></think>"}).encode(),
headers={"Content-Type": "application/json"}), timeout=60).read())["tokens"]
gen = stream(a.url, "/generate", {"input_ids": tok, "stream": True, "sampling_params": {"temperature": 0}})
for name, pts in (("chat", chat), ("generate", gen)):
s = score(pts)
s["pass"] = s["half_share"] >= 0.40 and s["tail_share"] <= 0.10
ok &= s["pass"]
print(name, json.dumps(s), flush=True)
print("STREAM_RATE", "PASS" if ok else "FAIL", flush=True)
sys.exit(0 if ok else 1)
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |