rui40000andClaude Fable 5
ca61c48dbd
- 一个节点调用 ZenMux 聚合平台全部文本类模型（20 厂商 / 138 模型） - vendor→model 二级下拉：前端 web/zenmux_cascade.js 按厂商实时收窄模型列表 - 模型选项后缀标注输入/输出价格（USD/百万token），如 [入$0.2/M 出$1.25/M] - 默认模型 openai/gpt-5.4-nano；具备 api_key/system/user prompt/seed/ temperature/top_p/max_tokens/多模态图像×6/detail/base_url/proxy 等完整参数 - 模型清单随包分发（models_snapshot.json），离线可列出； build_snapshot.py 可随时重拉最新价格 - VALIDATE_INPUTS 宽松校验：价格快照更新后旧工作流的模型标签仍可解析执行 - base_url 默认值沿用仓库约定不含 "://"，规避 ComfyUI 前端吞协议片段的老问题 - 导出 WEB_DIRECTORY 以装载前端级联脚本 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
92 lines
3.1 KiB
Python
92 lines
3.1 KiB
Python
# -*- coding: utf-8 -*-
|
||
"""
|
||
从 ZenMux 拉取模型列表，提炼成节点需要的精简快照。
|
||
|
||
运行：python build_snapshot.py
|
||
产出：models_snapshot.json（随节点分发，保证离线可用）
|
||
|
||
只保留文本类模型（输出模态含 text），并预先算好基础输入/输出价格，
|
||
避免节点运行时每次都解析 ZenMux 那套分段定价结构。
|
||
"""
|
||
import json
|
||
import os
|
||
import sys
|
||
|
||
import requests
|
||
|
||
# Windows GBK 控制台打印中文/特殊字符会崩，强制 UTF-8
|
||
if hasattr(sys.stdout, "reconfigure"):
|
||
sys.stdout.reconfigure(encoding="utf-8", errors="replace")
|
||
|
||
API = "https://zenmux.ai/api/v1/models"
|
||
HERE = os.path.dirname(os.path.abspath(__file__))
|
||
OUT = os.path.join(HERE, "models_snapshot.json")
|
||
|
||
|
||
def base_price(pricings, key):
|
||
"""取某项价格的基础段（prompt_tokens 从 0 起的那一段），单位 USD / 百万 token。"""
|
||
seg = (pricings or {}).get(key)
|
||
if not seg:
|
||
return None
|
||
vals = [s for s in seg if isinstance(s, dict) and "value" in s]
|
||
if not vals:
|
||
return None
|
||
base = None
|
||
for s in vals:
|
||
cond = s.get("conditions", {}).get("prompt_tokens", {})
|
||
if cond.get("gte", 0) in (0, None):
|
||
base = s
|
||
break
|
||
base = base or vals[0]
|
||
try:
|
||
return float(base["value"])
|
||
except (TypeError, ValueError, KeyError):
|
||
return None
|
||
|
||
|
||
def build(raw_models):
|
||
out = []
|
||
for m in raw_models:
|
||
out_mod = m.get("output_modalities", [])
|
||
if "text" not in out_mod: # 只要文本类模型
|
||
continue
|
||
pr = m.get("pricings") or {}
|
||
out.append({
|
||
"id": m["id"],
|
||
"display_name": m.get("display_name", m["id"]),
|
||
# 厂商一律取模型 id 的前缀（"openai/gpt-5.4-nano" -> "openai"），
|
||
# 与前端级联 JS 的过滤规则（label.split("/")[0]）保持同一真源。
|
||
"vendor": m["id"].split("/")[0] if "/" in m["id"] else m.get("owned_by", "other"),
|
||
"input_price": base_price(pr, "prompt"),
|
||
"output_price": base_price(pr, "completion"),
|
||
"context_length": m.get("context_length"),
|
||
"input_modalities": m.get("input_modalities", []),
|
||
})
|
||
# 按厂商、再按 id 排序，方便前端级联展示
|
||
out.sort(key=lambda x: (x["vendor"].lower(), x["id"].lower()))
|
||
return out
|
||
|
||
|
||
def main():
|
||
print(f"拉取 {API} ...")
|
||
r = requests.get(API, timeout=30)
|
||
r.raise_for_status()
|
||
data = r.json()["data"]
|
||
models = build(data)
|
||
payload = {
|
||
"source": API,
|
||
"count": len(models),
|
||
"models": models,
|
||
}
|
||
with open(OUT, "w", encoding="utf-8") as f:
|
||
json.dump(payload, f, ensure_ascii=False, indent=1)
|
||
print(f"写入 {OUT}：{len(models)} 个文本模型")
|
||
vendors = sorted(set(m["vendor"] for m in models))
|
||
print(f"厂商 {len(vendors)} 个：{', '.join(vendors)}")
|
||
assert any(m["id"] == "openai/gpt-5.4-nano" for m in models), "默认模型缺失！"
|
||
print("默认模型 openai/gpt-5.4-nano 存在 ✓")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
sys.exit(main())
|