hermes_bot_models.py 6.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119
  1. """Read-only Bot model catalogue and runtime resolution for one Hermes home. Never writes config.
  2. Ported from the former yaoyao-bot-bridge plugin; the service process already runs in the
  3. selected profile's home, so no per-request profile scope is needed.
  4. """
  5. import re
  6. class ModelSettingsError(Exception):
  7. pass
  8. def defaults(ctx, config):
  9. agent = config.get("agent") or {}
  10. effort = agent.get("reasoning_effort", "medium")
  11. tier = agent.get("service_tier") or "normal"
  12. return {"provider": ctx.current_provider, "model": ctx.current_model,
  13. "reasoningEffort": "none" if effort is False else str(effort or "medium"),
  14. "fastMode": {"priority": "fast"}.get(tier, tier)}
  15. def route_url(ctx, provider):
  16. if provider == ctx.current_provider:
  17. return ctx.current_base_url
  18. entry = (ctx.user_providers or {}).get(provider)
  19. if isinstance(entry, dict):
  20. return str(entry.get("base_url") or "")
  21. for entry in ctx.custom_providers or []:
  22. if provider in (entry.get("name"), entry.get("id")):
  23. return str(entry.get("base_url") or "")
  24. return ""
  25. def capability(provider, model, base_url="", hint=None):
  26. from hermes_constants import VALID_REASONING_EFFORTS
  27. from hermes_cli.models import resolve_fast_mode_overrides
  28. hint = hint or {}
  29. known = hint.get("reasoning") is False
  30. reasoning = hint.get("reasoning", True)
  31. try:
  32. from agent.models_dev import get_model_capabilities
  33. meta = get_model_capabilities(provider, model)
  34. if meta is not None:
  35. known = not meta.supports_reasoning or known
  36. reasoning = bool(meta.supports_reasoning) and reasoning
  37. except Exception:
  38. pass
  39. efforts = (["none"] if hint.get("can_disable_reasoning", True) else []) + list(VALID_REASONING_EFFORTS) if reasoning else []
  40. if provider == "copilot":
  41. from hermes_cli.models import github_model_reasoning_efforts
  42. declared = github_model_reasoning_efforts(model)
  43. if declared:
  44. efforts, known = list(declared), True
  45. fast = resolve_fast_mode_overrides(model, provider=provider, base_url=base_url) is not None
  46. return {"provider": provider, "model": model, "name": model,
  47. "reasoningEfforts": efforts, "reasoningKnown": known,
  48. "fastModes": ["normal", "fast", "auto", "cold"] if fast else ["normal"]}
  49. def options():
  50. from hermes_cli.config import load_config
  51. from hermes_cli.inventory import load_picker_context, build_model_options_payload
  52. if True:
  53. ctx = load_picker_context()
  54. payload = build_model_options_payload(ctx, explicit_only=True)
  55. rows = []
  56. for provider in payload.get("providers", []):
  57. slug = provider["slug"]
  58. for model in provider.get("models", []):
  59. if isinstance(model, str):
  60. rows.append(capability(slug, model, route_url(ctx, slug), (provider.get("capabilities") or {}).get(model)))
  61. return {"version": 1, "defaults": defaults(ctx, load_config()), "models": rows}
  62. def resolve(settings):
  63. from hermes_cli.config import load_config
  64. from hermes_cli.inventory import load_picker_context, _reasoning_catalog_reader
  65. from hermes_cli.model_switch import switch_model
  66. from hermes_cli.model_selection_guards import combined_selection_warning
  67. from hermes_constants import parse_reasoning_effort
  68. if settings is None:
  69. settings = {}
  70. if not isinstance(settings, dict) or set(settings) - {"provider", "model", "reasoningEffort", "fastMode"}:
  71. raise ModelSettingsError("模型配置格式无效")
  72. if True:
  73. ctx, config = load_picker_context(), load_config()
  74. value = defaults(ctx, config)
  75. value.update({key: item for key, item in settings.items() if item is not None})
  76. if bool(settings.get("provider")) != bool(settings.get("model")):
  77. raise ModelSettingsError("模型与 Provider 必须一起选择")
  78. for key in ("model", "provider"):
  79. item = value[key]
  80. if not isinstance(item, str) or not item or len(item) > 512 or re.search(r"[\s\"'\\]", item) or item.startswith("-"):
  81. raise ModelSettingsError("基础机器人尚未配置有效的模型与 Provider")
  82. model_config = config.get("model") or {}
  83. current_key = model_config.get("api_key", "") if isinstance(model_config, dict) else ""
  84. result = switch_model(raw_input=value["model"], current_provider=ctx.current_provider,
  85. current_model=ctx.current_model, current_base_url=ctx.current_base_url,
  86. current_api_key=current_key, is_global=False, explicit_provider=value["provider"],
  87. user_providers=ctx.user_providers, custom_providers=ctx.custom_providers)
  88. if not result.success:
  89. # Provider errors may include secrets/URLs; keep this boundary deliberately generic.
  90. raise ModelSettingsError("所选模型或 Provider 当前不可用,请检查基础机器人的模型服务")
  91. value.update(provider=result.target_provider, model=result.new_model)
  92. hint = {}
  93. reader = _reasoning_catalog_reader(result.target_provider)
  94. detail = reader(result.new_model) if reader else None
  95. if detail:
  96. hint = {"reasoning": detail.get("supports_reasoning", True), "can_disable_reasoning": not detail.get("mandatory")}
  97. caps = capability(value["provider"], value["model"], result.base_url or route_url(ctx, value["provider"]), hint)
  98. if not isinstance(value["reasoningEffort"], str) or parse_reasoning_effort(value["reasoningEffort"]) is None:
  99. raise ModelSettingsError("思考等级无效")
  100. if settings.get("reasoningEffort") is not None and value["reasoningEffort"] not in caps["reasoningEfforts"]:
  101. raise ModelSettingsError("该模型不支持所选思考等级,请重新选择或继承基础机器人")
  102. if value["fastMode"] not in caps["fastModes"]:
  103. raise ModelSettingsError("该模型或 Provider 不支持所选速度,请重新选择普通速度")
  104. warning = combined_selection_warning(value["model"], provider=value["provider"],
  105. base_url=result.base_url, api_key=result.api_key, model_info=result.model_info)
  106. return {"version": 1, "effective": value, "confirmationMessage": warning.message if warning else None}