File size: 22,471 Bytes
e0b68d2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 | #!/usr/bin/env python
# -*- coding: utf-8 -*-
"""Model management module.
Centrally manages model mapping, MCP server configuration, scene defaults,
model capabilities, and model feature resolution logic. Supports both
hardcoded fallback data and dynamic parsing from online model metadata.
"""
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional
from app.core.config import settings
from app.models.schemas import OpenAIRequest
from app.utils.logger import logger
@dataclass
class ParsedModel:
"""Structured representation of an online model after parsing."""
upstream_id: str
display_name: str
capabilities: Dict[str, Any] = field(default_factory=dict)
mcp_server_ids: List[str] = field(default_factory=list)
params: Dict[str, Any] = field(default_factory=dict)
is_active: bool = True
tags: List[str] = field(default_factory=list)
class ModelManager:
"""Model configuration manager.
Manages four configuration dictionaries:
1. ``model_mapping`` -- external model name -> upstream model ID
2. ``model_mcp_servers`` -- model -> default MCP server list
3. ``model_scene_defaults``-- model -> scene default parameters
4. ``model_capabilities`` -- model -> capability declarations
On startup, hardcoded defaults are loaded. When
:meth:`load_from_online_models` is called with fresh upstream data the
hardcoded values are replaced with dynamically generated variants.
"""
def __init__(self) -> None:
self._init_hardcoded_defaults()
self._parsed: bool = False # True after successful online parse
self._casefold_index: Dict[str, str] = {}
# Apply aliases to hardcoded defaults (online path re-applies after parse)
self.apply_aliases()
# ------------------------------------------------------------------
# Hardcoded fallback (kept for offline / pre-fetch scenarios)
# ------------------------------------------------------------------
def _init_hardcoded_defaults(self) -> None:
"""Initialise all model dictionaries with hardcoded defaults."""
# External model name -> upstream model ID
self.model_mapping: Dict[str, str] = {
settings.GLM45_MODEL: "0727-360B-API",
settings.GLM45_THINKING_MODEL: "0727-360B-API",
settings.GLM45_SEARCH_MODEL: "0727-360B-API",
settings.GLM45_AIR_MODEL: "0727-106B-API",
settings.GLM46V_MODEL: "glm-4.6v",
settings.GLM46V_ADVANCED_SEARCH_MODEL: "glm-4.6v",
settings.GLM5_MODEL: "glm-5",
settings.GLM5_THINKING_MODEL: "glm-5",
settings.GLM5_AGENT_MODEL: "glm-5",
settings.GLM5_ADVANCED_SEARCH_MODEL: "glm-5",
settings.GLM47_MODEL: "glm-4.7",
settings.GLM47_THINKING_MODEL: "glm-4.7",
settings.GLM47_SEARCH_MODEL: "glm-4.7",
settings.GLM47_ADVANCED_SEARCH_MODEL: "glm-4.7",
}
# Default MCP servers per model
self.model_mcp_servers: Dict[str, List[str]] = {
settings.GLM5_ADVANCED_SEARCH_MODEL: ["advanced-search"],
settings.GLM46V_ADVANCED_SEARCH_MODEL: ["advanced-search"],
settings.GLM47_ADVANCED_SEARCH_MODEL: ["advanced-search"],
}
# Scene defaults per model
self.model_scene_defaults: Dict[str, Dict[str, Any]] = {
settings.GLM5_AGENT_MODEL: {
"flags": ["general_agent"],
"enable_thinking": True,
"auto_web_search": False,
},
settings.GLM5_ADVANCED_SEARCH_MODEL: {
"enable_thinking": True,
"web_search": True,
"auto_web_search": True,
},
settings.GLM46V_ADVANCED_SEARCH_MODEL: {
"enable_thinking": True,
"web_search": True,
"auto_web_search": True,
},
settings.GLM47_ADVANCED_SEARCH_MODEL: {
"enable_thinking": True,
"web_search": True,
"auto_web_search": True,
},
settings.GLM45_SEARCH_MODEL: {
"web_search": True,
"auto_web_search": True,
},
settings.GLM47_SEARCH_MODEL: {
"web_search": True,
"auto_web_search": True,
},
}
# Model capability declarations
self.model_capabilities: Dict[str, Dict[str, bool]] = {
settings.GLM45_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": False,
},
settings.GLM45_THINKING_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": True,
},
settings.GLM45_SEARCH_MODEL: {
"tool_use": False, "vision": False,
"web_search": True, "thinking": False,
},
settings.GLM45_AIR_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": False,
},
settings.GLM46V_MODEL: {
"tool_use": False, "vision": True,
"web_search": False, "thinking": False,
},
settings.GLM46V_ADVANCED_SEARCH_MODEL: {
"tool_use": False, "vision": True,
"web_search": True, "thinking": True,
},
settings.GLM5_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": False,
},
settings.GLM5_THINKING_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": True,
},
settings.GLM5_AGENT_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": True,
"agent": True,
},
settings.GLM5_ADVANCED_SEARCH_MODEL: {
"tool_use": False, "vision": False,
"web_search": True, "thinking": True,
},
settings.GLM47_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": False,
},
settings.GLM47_THINKING_MODEL: {
"tool_use": False, "vision": False,
"web_search": False, "thinking": True,
},
settings.GLM47_SEARCH_MODEL: {
"tool_use": False, "vision": False,
"web_search": True, "thinking": False,
},
settings.GLM47_ADVANCED_SEARCH_MODEL: {
"tool_use": False, "vision": False,
"web_search": True, "thinking": True,
},
}
# Hardcoded supported model list (fallback order)
self._hardcoded_supported_models: List[str] = [
settings.GLM45_MODEL,
settings.GLM45_THINKING_MODEL,
settings.GLM45_SEARCH_MODEL,
settings.GLM45_AIR_MODEL,
settings.GLM46V_MODEL,
settings.GLM46V_ADVANCED_SEARCH_MODEL,
settings.GLM5_MODEL,
settings.GLM5_THINKING_MODEL,
settings.GLM5_AGENT_MODEL,
settings.GLM5_ADVANCED_SEARCH_MODEL,
settings.GLM47_MODEL,
settings.GLM47_THINKING_MODEL,
settings.GLM47_SEARCH_MODEL,
settings.GLM47_ADVANCED_SEARCH_MODEL,
]
# Will be populated by load_from_online_models
self._dynamic_supported_models: List[str] = []
# ------------------------------------------------------------------
# Online model parsing
# ------------------------------------------------------------------
def load_from_online_models(self, online_models: List[Dict[str, Any]]) -> None:
"""Parse online model metadata and generate variant configurations.
For each non-blacklisted active model, up to five variants are
generated (base, Thinking, Search, Agent, advanced-search) depending
on the upstream capability flags. The four core dictionaries are
fully replaced on success.
Args:
online_models: List of model dicts as returned by
:meth:`upstream.UpstreamClient.get_online_models`.
"""
# Read blacklist from settings (comma-separated upstream IDs)
blacklist_raw = getattr(settings, "MODEL_BLACKLIST", "glm-4-flash")
blacklist = {
item.strip().lower()
for item in blacklist_raw.split(",")
if item.strip()
}
# Parse each raw dict into a ParsedModel
parsed: List[ParsedModel] = []
skipped_inactive = 0
skipped_blacklisted = 0
for raw in online_models:
upstream_id = raw.get("id", "")
if not upstream_id:
continue
# Blacklist check (case-insensitive on upstream ID)
if upstream_id.lower() in blacklist:
skipped_blacklisted += 1
continue
is_active = raw.get("is_active", True)
if not is_active:
skipped_inactive += 1
continue
display_name = raw.get("name") or upstream_id
capabilities = raw.get("capabilities") or {}
mcp_server_ids = raw.get("mcpServerIds") or []
tags = raw.get("tags") or []
parsed.append(ParsedModel(
upstream_id=upstream_id,
display_name=display_name,
capabilities=capabilities,
mcp_server_ids=mcp_server_ids,
params={},
is_active=is_active,
tags=tags,
))
# Build fresh dictionaries from parsed models
new_mapping: Dict[str, str] = {}
new_capabilities: Dict[str, Dict[str, bool]] = {}
new_mcp_servers: Dict[str, List[str]] = {}
new_scene_defaults: Dict[str, Dict[str, Any]] = {}
new_supported: List[str] = []
total_variants = 0
for model in parsed:
variants = self._generate_variants(model)
for v_name, v_caps, v_scene, v_mcp in variants:
new_mapping[v_name] = model.upstream_id
new_capabilities[v_name] = v_caps
if v_scene:
new_scene_defaults[v_name] = v_scene
if v_mcp:
new_mcp_servers[v_name] = v_mcp
new_supported.append(v_name)
total_variants += 1
# Atomically replace all dictionaries
self.model_mapping = new_mapping
self.model_capabilities = new_capabilities
self.model_mcp_servers = new_mcp_servers
self.model_scene_defaults = new_scene_defaults
self._dynamic_supported_models = new_supported
self._parsed = True
logger.debug(
"Online model parsing complete: %d models parsed, "
"%d variants generated, %d blacklisted, %d inactive",
len(parsed), total_variants,
skipped_blacklisted, skipped_inactive,
)
# Apply user-defined aliases on top of generated variants
self.apply_aliases()
@staticmethod
def _generate_variants(
model: ParsedModel,
) -> List[tuple]:
"""Generate variant tuples for a single parsed model.
Returns:
List of (variant_name, capabilities_dict, scene_defaults_dict,
mcp_servers_list) tuples.
"""
caps = model.capabilities
name = model.display_name
name_lower = name.lower()
mcp_ids = model.mcp_server_ids
# Derive base capability flags from upstream capabilities
has_return_fc = bool(caps.get("returnFc", False))
has_vision = bool(caps.get("vision", False))
has_think = bool(caps.get("think", False))
has_web_search = bool(caps.get("web_search", False))
has_agent_mode = bool(caps.get("agent_mode", False))
has_deep_web_search = "deep-web-search" in mcp_ids
has_advanced_search = "advanced-search" in mcp_ids
base_caps: Dict[str, bool] = {
"tool_use": False,
"vision": has_vision,
}
variants: List[tuple] = []
# --- 1. Base variant (always generated) ---
variants.append((
name,
{**base_caps, "web_search": False, "thinking": False},
{}, # no scene defaults for base
[], # no MCP servers for base
))
# --- 2. Thinking variant ---
if has_think and "thinking" not in name_lower:
thinking_name = f"{name}-Thinking"
variants.append((
thinking_name,
{**base_caps, "web_search": False, "thinking": True},
{"enable_thinking": True},
[],
))
# --- 3. Search variant ---
if (has_web_search or has_deep_web_search) and "search" not in name_lower:
search_name = f"{name}-Search"
variants.append((
search_name,
{**base_caps, "web_search": True, "thinking": False},
{"web_search": True, "auto_web_search": True},
[],
))
# --- 4. Agent variant ---
if has_agent_mode and "agent" not in name_lower:
agent_name = f"{name}-Agent"
variants.append((
agent_name,
{**base_caps, "web_search": False, "thinking": True, "agent": True},
{"flags": ["general_agent"], "enable_thinking": True},
[],
))
# --- 5. advanced-search variant ---
if has_advanced_search and "advanced-search" not in name_lower:
adv_search_name = f"{name}-advanced-search"
variants.append((
adv_search_name,
{**base_caps, "web_search": True, "thinking": True},
{"enable_thinking": True, "web_search": True, "auto_web_search": True},
["advanced-search"],
))
return variants
# ------------------------------------------------------------------
# Alias mapping
# ------------------------------------------------------------------
def apply_aliases(self, aliases_str: Optional[str] = None) -> None:
"""Apply model aliases from a comma-separated ``alias=target`` string.
Each alias inherits the target model's upstream mapping, capabilities,
MCP servers, and scene defaults. Unknown targets are silently skipped.
Args:
aliases_str: Alias definitions. Falls back to
``settings.MODEL_ALIASES`` when *None*.
"""
if aliases_str is None:
aliases_str = getattr(settings, "MODEL_ALIASES", "")
if not aliases_str or not aliases_str.strip():
self._rebuild_casefold_index()
return
applied = 0
for pair in aliases_str.split(","):
pair = pair.strip()
if "=" not in pair:
continue
alias, target = pair.split("=", 1)
alias = alias.strip()
target = target.strip()
if not alias or not target:
continue
# Target must exist in current mapping
if target not in self.model_mapping:
logger.warning(
"Alias target %r not found in model mapping, "
"skipping alias %r",
target, alias,
)
continue
# Copy all properties from target
self.model_mapping[alias] = self.model_mapping[target]
if target in self.model_capabilities:
self.model_capabilities[alias] = dict(self.model_capabilities[target])
if target in self.model_mcp_servers:
self.model_mcp_servers[alias] = list(self.model_mcp_servers[target])
if target in self.model_scene_defaults:
self.model_scene_defaults[alias] = dict(self.model_scene_defaults[target])
# Add to supported models list
if self._parsed and alias not in self._dynamic_supported_models:
self._dynamic_supported_models.append(alias)
elif not self._parsed and alias not in self._hardcoded_supported_models:
self._hardcoded_supported_models.append(alias)
applied += 1
if applied:
logger.debug("Applied %d model alias(es)", applied)
self._rebuild_casefold_index()
# ------------------------------------------------------------------
# Case-insensitive model name resolution
# ------------------------------------------------------------------
def _rebuild_casefold_index(self) -> None:
"""Rebuild case-insensitive lookup index from model_mapping keys."""
self._casefold_index = {k.casefold(): k for k in self.model_mapping}
def _resolve_name(self, model_name: str) -> Optional[str]:
"""Resolve model_name to its canonical key (case-insensitive).
Tries exact match first, then falls back to case-insensitive lookup.
"""
if model_name in self.model_mapping:
return model_name
return self._casefold_index.get(model_name.casefold())
# ------------------------------------------------------------------
# Public query API (signatures unchanged)
# ------------------------------------------------------------------
def get_upstream_model_id(self, model_name: str) -> Optional[str]:
"""Return the upstream model ID for *model_name*, or None."""
canonical = self._resolve_name(model_name)
return self.model_mapping[canonical] if canonical else None
def get_mcp_servers(self, model_name: str) -> List[str]:
"""Return the default MCP server list for *model_name*."""
return self.model_mcp_servers.get(
self._resolve_name(model_name) or model_name, []
)
def get_scene_defaults(self, model_name: str) -> Dict[str, Any]:
"""Return scene default parameters for *model_name*."""
return self.model_scene_defaults.get(
self._resolve_name(model_name) or model_name, {}
)
def get_model_capabilities(self, model_name: str) -> Dict[str, bool]:
"""Return capability declarations for *model_name*.
Unregistered models receive a minimal base capability set.
"""
return self.model_capabilities.get(
self._resolve_name(model_name) or model_name, {
"tool_use": False,
"vision": False,
"web_search": False,
"thinking": False,
}
)
def get_supported_models(self) -> List[str]:
"""Return the list of externally exposed supported models.
Uses dynamically generated list when online parsing has succeeded,
otherwise falls back to the hardcoded list.
"""
if self._parsed and self._dynamic_supported_models:
return list(self._dynamic_supported_models)
return list(self._hardcoded_supported_models)
# ------------------------------------------------------------------
# Feature resolution (signature unchanged)
# ------------------------------------------------------------------
def resolve_model_features(self, request: OpenAIRequest) -> Dict[str, Any]:
"""Resolve model feature parameters from request context.
Priority order: explicit client parameters > scene defaults >
model-suffix inference.
Args:
request: OpenAI-compatible request object.
Returns:
Dict with keys: upstream_model_id, enable_thinking, web_search,
auto_web_search, mcp_servers, extra, flags.
"""
requested_model = request.model
is_thinking_model = "-thinking" in requested_model.casefold()
is_search_model = "-search" in requested_model.casefold()
scene_defaults = self.get_scene_defaults(requested_model)
# enable_thinking: client > scene default > suffix inference
enable_thinking = request.enable_thinking
if enable_thinking is None:
enable_thinking = scene_defaults.get("enable_thinking", is_thinking_model)
# web_search: client > scene default > suffix inference
web_search = request.web_search
if web_search is None:
web_search = scene_defaults.get("web_search", is_search_model)
# auto_web_search: scene default > False
auto_web_search = scene_defaults.get("auto_web_search", False)
# Upstream model ID
upstream_model_id = self.get_upstream_model_id(requested_model)
if upstream_model_id is None:
raise ValueError(f"Unsupported model: {requested_model}")
# mcp_servers: client pass-through takes priority
mcp_servers = (
request.mcp_servers
if request.mcp_servers is not None
else self.get_mcp_servers(requested_model)
)
# extra and flags: client > scene default > empty
extra = request.extra if request.extra is not None else scene_defaults.get("extra", {})
flags = request.flags if request.flags is not None else scene_defaults.get("flags", [])
return {
"upstream_model_id": upstream_model_id,
"enable_thinking": enable_thinking,
"web_search": web_search,
"auto_web_search": auto_web_search,
"mcp_servers": mcp_servers,
"extra": extra,
"flags": flags,
}
|