hz4th_coder/llm-proxy
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v2.1.0: 能力路由体系 + 模型管理 + 生图/语音/视频端点 + 大量修复

Browse Source 
- 提供商重构: Local Qwen(文本+视觉) / SiliconFlow LLM(文本) / Autodl(文本+视觉) / Autodl Image(生图)
- 能力标签: text/vision/audio_out/audio_in/image_gen/video_gen，模型级可配
- AUTO配置绑定能力: auto/auto-text/auto-vision/auto-image/auto-voice-out/auto-voice-in/auto-video
- 新端点: /v1/images/generations /v1/audio/speech /v1/audio/transcriptions /v1/video/generations
- 模型管理页: 能力切换/增删改/设默认/别名管理
- 修复: 失败切换不再替换用户模型; 熔断冷却后自动恢复; 重试不再白等; embeddings按模型路由
- 修复: 后台模板API路径(/api/admin); 删除v1旧admin目录与双端口混乱
- 修复: run.sh硬编码路径; 死配置RETRY/LOG接线; engine_completions空body防护
- 端口: 16003(16001-16100白名单区间); debug关闭; start.sh部署脚本
This commit is contained in:
hz4th_coder
2026-08-28 11:41:17 +08:00
parent 693f5edf33
commit 178968ba1a
21 changed files with 2331 additions and 3646 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config/settings.py
+173 -30
View File
@@ -1,5 +1,14 @@
"""
大模型API中转系统配置 - 支持动态修改
v2.1.0 - 能力(Capability)体系: 每个模型标记能力，AUTO配置按能力绑定
能力类型:
text 文本推理
vision 视觉能力 (多模态输入)
audio_out 语音输出 (TTS)
audio_in 语音输入 (ASR)
image_gen 图片生成
video_gen 视频生成
"""
import json
@@ -8,28 +17,76 @@ from pathlib import Path
# 配置文件路径
CONFIG_FILE = Path(__file__).parent.parent / 'data' / 'config.json'
# 能力定义
CAPABILITY_DEFS = {
"text": "文本推理",
"vision": "视觉能力",
"audio_out": "语音输出",
"audio_in": "语音输入",
"image_gen": "图片生成",
"video_gen": "视频生成",
}
# 默认上游模型配置
# capabilities: 该提供商下所有模型的默认能力（模型管理页可对单个模型覆盖）
DEFAULT_PROVIDERS = [
{
"id": "local-qwen",
"name": "Local Qwen",
"priority": 1,
"base_url": "http://192.168.2.5:1234/v1",
"api_key": "sk-lm-fuP5tGU8:Hi7YU87jHyDP6Ay8Tl2j",
"models": ["qwen3.5-4b", "qwen3.5", "qwen"],
"default_model": "qwen3.5-4b",
"timeout": 120,
"base_url": "http://121.40.164.32:18003/v1",
"api_key": "sk-xxxx",
"models": [
{"name": "unsloth/Qwen3.8-27B-Q6_K", "capabilities": ["text", "vision"]},
{"name": "unsloth/Qwen3.8-27B-Q4_K_M", "capabilities": ["text", "vision"]},
],
"default_model": "unsloth/Qwen3.8-27B-Q6_K",
"capabilities": ["text", "vision"],
"timeout": 180,
"enabled": True,
},
{
"id": "siliconflow-deepseek",
"name": "SiliconFlow DeepSeek",
"id": "siliconflow-llm",
"name": "SiliconFlow LLM",
"priority": 2,
"base_url": "https://api.siliconflow.cn/v1",
"api_key": "sk-fhpoexpptvjghpnphtaxbkhjwulzovoqfffbckcfscjmwhcg",
"models": ["Pro/deepseek-ai/DeepSeek-V3.2", "deepseek-v3", "deepseek"],
"default_model": "Pro/deepseek-ai/DeepSeek-V3.2",
"timeout": 120,
"models": [
{"name": "deepseek-ai/DeepSeek-V4-Flash", "capabilities": ["text"]},
{"name": "meituan-longcat/LongCat-2.0", "capabilities": ["text"]},
],
"default_model": "deepseek-ai/DeepSeek-V4-Flash",
"capabilities": ["text"],
"timeout": 180,
"enabled": True,
},
{
"id": "autodl",
"name": "Autodl",
"priority": 3,
"base_url": "https://www.autodl.art/api/v1",
"api_key": "F9MBfolzuapqTsD4KmUf9qen720rXvUZ3Sp3IrWiCTukqonx",
"models": [
{"name": "qwen3.6-plus", "capabilities": ["text", "vision"]},
{"name": "GLM-5.3-flash", "capabilities": ["text", "vision"]},
],
"default_model": "GLM-5.3-flash",
"capabilities": ["text", "vision"],
"timeout": 180,
"enabled": True,
},
{
"id": "autodl-image",
"name": "Autodl Image",
"priority": 4,
"base_url": "https://www.autodl.art/api/v1",
"api_key": "F9MBfolzuapqTsD4KmUf9qen720rXvUZ3Sp3IrWiCTukqonx",
"models": [
{"name": "Qwen-Image", "capabilities": ["image_gen"]},
],
"default_model": "Qwen-Image",
"capabilities": ["image_gen"],
"timeout": 180,
"enabled": True,
},
]
@@ -37,24 +94,80 @@ DEFAULT_PROVIDERS = [
# 默认模型别名
DEFAULT_MODEL_ALIASES = {
"auto": "auto",
"qwen": "qwen3.5-4b",
"qwen3.5": "qwen3.5-4b",
"qwen3.5-4b": "qwen3.5-4b",
"deepseek": "Pro/deepseek-ai/DeepSeek-V3.2",
"deepseek-v3": "Pro/deepseek-ai/DeepSeek-V3.2",
"deepseek-v3.2": "Pro/deepseek-ai/DeepSeek-V3.2",
"auto-text": "auto-text",
"auto-vision": "auto-vision",
"auto-image": "auto-image",
"auto-voice-out": "auto-voice-out",
"auto-voice-in": "auto-voice-in",
"auto-video": "auto-video",
"auto-vlm": "auto-vision", # 兼容旧配置
"qwen": "unsloth/Qwen3.8-27B-Q6_K",
"qwen3.8": "unsloth/Qwen3.8-27B-Q6_K",
"local": "unsloth/Qwen3.8-27B-Q6_K",
"deepseek": "deepseek-ai/DeepSeek-V4-Flash",
"deepseek-v4-flash": "deepseek-ai/DeepSeek-V4-Flash",
"longcat": "meituan-longcat/LongCat-2.0",
"glm": "GLM-5.3-flash",
"qwen3.6-plus": "qwen3.6-plus",
"gpt-4": "GLM-5.3-flash",
}
# 默认Auto配置（自定义候选模型和优先级）
# 默认Auto配置：每个auto配置固定绑定一个能力
# capability: 固定此auto的功能类型（取自 CAPABILITY_DEFS）
# providers: 候选提供商（* 表示所有启用的、且具备该能力模型的提供商）
DEFAULT_AUTO_PROFILES = {
"auto": {
"name": "默认Auto",
"description": "所有启用的提供商按优先级自动选择",
"providers": ["*"], # * 表示所有启用的提供商
"strategy": "priority", # priority | random | round-robin
}
"description": "文本推理 - 自动选择可用提供商",
"capability": "text",
"providers": ["*"],
"strategy": "priority",
},
"auto-text": {
"name": "文本推理",
"description": "纯文本推理，自动选择文本模型",
"capability": "text",
"providers": ["*"],
"strategy": "priority",
},
"auto-vision": {
"name": "视觉能力",
"description": "多模态视觉理解，自动选择视觉模型",
"capability": "vision",
"providers": ["*"],
"strategy": "priority",
},
"auto-image": {
"name": "图片生成",
"description": "文生图，自动选择生图模型",
"capability": "image_gen",
"providers": ["*"],
"strategy": "priority",
},
"auto-voice-out": {
"name": "语音输出",
"description": "语音合成(TTS)，自动选择语音输出模型",
"capability": "audio_out",
"providers": ["*"],
"strategy": "priority",
},
"auto-voice-in": {
"name": "语音输入",
"description": "语音识别(ASR)，自动选择语音输入模型",
"capability": "audio_in",
"providers": ["*"],
"strategy": "priority",
},
"auto-video": {
"name": "视频生成",
"description": "文生视频，自动选择视频生成模型",
"capability": "video_gen",
"providers": ["*"],
"strategy": "priority",
},
}
def load_config():
"""加载配置"""
if CONFIG_FILE.exists():
@@ -68,16 +181,19 @@ def load_config():
"model_aliases": DEFAULT_MODEL_ALIASES,
}
def save_config(config):
"""保存配置"""
CONFIG_FILE.parent.mkdir(parents=True, exist_ok=True)
CONFIG_FILE.write_text(json.dumps(config, ensure_ascii=False, indent=2), encoding='utf-8')
def get_providers():
"""获取提供商列表"""
config = load_config()
return config.get("providers", DEFAULT_PROVIDERS)
def get_provider(provider_id):
"""获取单个提供商"""
providers = get_providers()
@@ -86,25 +202,27 @@ def get_provider(provider_id):
return p
return None
def add_provider(provider):
"""添加提供商"""
config = load_config()
providers = config.get("providers", [])
# 生成ID
if not provider.get("id"):
provider["id"] = provider["name"].lower().replace(" ", "-").replace(".", "-")
providers.append(provider)
config["providers"] = providers
save_config(config)
return provider
def update_provider(provider_id, data):
"""更新提供商"""
config = load_config()
providers = config.get("providers", [])
for i, p in enumerate(providers):
if p.get("id") == provider_id:
providers[i] = {**p, **data}
@@ -113,6 +231,7 @@ def update_provider(provider_id, data):
return providers[i]
return None
def delete_provider(provider_id):
"""删除提供商"""
config = load_config()
@@ -122,26 +241,29 @@ def delete_provider(provider_id):
save_config(config)
return True
def update_priority(provider_ids):
"""更新优先级顺序"""
config = load_config()
providers = config.get("providers", [])
# 按新顺序设置优先级
for i, pid in enumerate(provider_ids):
for p in providers:
if p.get("id") == pid:
p["priority"] = i + 1
config["providers"] = providers
save_config(config)
return providers
def get_model_aliases():
"""获取模型别名"""
config = load_config()
return config.get("model_aliases", DEFAULT_MODEL_ALIASES)
def update_model_alias(alias, target):
"""更新模型别名"""
config = load_config()
@@ -151,16 +273,33 @@ def update_model_alias(alias, target):
save_config(config)
return aliases
def delete_model_alias(alias):
"""删除模型别名"""
if alias == "auto":
return False # 不能删除默认的auto
config = load_config()
aliases = config.get("model_aliases", {})
if alias in aliases:
del aliases[alias]
config["model_aliases"] = aliases
save_config(config)
return True
return False
def get_auto_profiles():
"""获取Auto配置列表"""
config = load_config()
return config.get("auto_profiles", DEFAULT_AUTO_PROFILES)
def get_auto_profile(profile_name):
"""获取单个Auto配置"""
profiles = get_auto_profiles()
return profiles.get(profile_name)
def add_auto_profile(profile_name, profile_data):
"""添加Auto配置"""
config = load_config()
@@ -170,6 +309,7 @@ def add_auto_profile(profile_name, profile_data):
save_config(config)
return profiles
def update_auto_profile(profile_name, profile_data):
"""更新Auto配置"""
config = load_config()
@@ -181,6 +321,7 @@ def update_auto_profile(profile_name, profile_data):
return profiles[profile_name]
return None
def delete_auto_profile(profile_name):
"""删除Auto配置"""
if profile_name == "auto":
@@ -194,6 +335,7 @@ def delete_auto_profile(profile_name):
return True
return False
# 初始化配置
config = load_config()
UPSTREAM_PROVIDERS = config.get("providers", DEFAULT_PROVIDERS)
@@ -202,8 +344,8 @@ MODEL_ALIASES = config.get("model_aliases", DEFAULT_MODEL_ALIASES)
# 服务配置
SERVER_CONFIG = {
"host": "0.0.0.0",
"port": 19007,
"debug": True,
"port": 16003,
"debug": False,
}
# 日志配置
@@ -213,14 +355,15 @@ LOG_CONFIG = {
"log_errors": True,
}
# 重试配置
# 重试/熔断配置
RETRY_CONFIG = {
"max_retries": 3,
"retry_delay": 1,
"cooldown_seconds": 60, # 熔断冷却期，到期自动恢复(半开)
"retry_on_errors": [
"connection_error",
"timeout",
"rate_limit",
"server_error",
],
}
}