Files
ss-tools/backend/src/services/llm_prompt_templates.py
2026-05-26 09:30:41 +03:00

215 lines
8.3 KiB
Python

# #region llm_prompt_templates [C:2] [TYPE Module] [SEMANTICS llm, prompt, template, normalization]
# @BRIEF Provide default LLM prompt templates and normalization helpers for runtime usage.
# @LAYER Domain
# @RELATION DEPENDS_ON -> [ConfigManager]
# @INVARIANT All required prompt template keys are always present after normalization.
from __future__ import annotations
from copy import deepcopy
from typing import Any
import warnings
from ..core.logger import logger
# #region DEFAULT_LLM_PROMPTS [C:2] [TYPE Constant]
# @BRIEF Default prompt templates used by documentation, dashboard validation, and git commit generation.
DEFAULT_LLM_PROMPTS: dict[str, str] = {
"dashboard_validation_prompt": (
"Analyze the attached dashboard screenshot and the following execution logs for health and visual issues.\n\n"
"Logs:\n"
"{logs}\n\n"
"Provide the analysis in JSON format with the following structure:\n"
"{\n"
' "status": "PASS" | "WARN" | "FAIL",\n'
' "summary": "Short summary of findings",\n'
' "issues": [\n'
" {\n"
' "severity": "WARN" | "FAIL",\n'
' "message": "Description of the issue",\n'
' "location": "Optional location info (e.g. chart name)"\n'
" }\n"
" ]\n"
"}"
),
"documentation_prompt": (
"Generate professional documentation for the following dataset and its columns.\n"
"Dataset: {dataset_name}\n"
"Columns: {columns_json}\n\n"
"Provide the documentation in JSON format:\n"
"{\n"
' "dataset_description": "General description of the dataset",\n'
' "column_descriptions": [\n'
" {\n"
' "name": "column_name",\n'
' "description": "Generated description"\n'
" }\n"
" ]\n"
"}"
),
"git_commit_prompt": (
"Generate a concise and professional git commit message based on the following diff and recent history.\n"
"Use Conventional Commits format (e.g., feat: ..., fix: ..., docs: ...).\n\n"
"Recent History:\n"
"{history}\n\n"
"Diff:\n"
"{diff}\n\n"
"Commit Message:"
),
}
# #endregion DEFAULT_LLM_PROMPTS
# #region DEFAULT_LLM_PROVIDER_BINDINGS [C:2] [TYPE Constant]
# @BRIEF Default provider binding per task domain.
DEFAULT_LLM_PROVIDER_BINDINGS: dict[str, str] = {
"dashboard_validation": "",
"documentation": "",
"git_commit": "",
}
# #endregion DEFAULT_LLM_PROVIDER_BINDINGS
# #region DEFAULT_LLM_ASSISTANT_SETTINGS [C:2] [TYPE Constant]
# @BRIEF Default planner settings for assistant chat intent model/provider resolution.
DEFAULT_LLM_ASSISTANT_SETTINGS: dict[str, str] = {
"assistant_planner_provider": "",
"assistant_planner_model": "",
}
# #endregion DEFAULT_LLM_ASSISTANT_SETTINGS
# #region normalize_llm_settings [C:3] [TYPE Function]
# @BRIEF Ensure llm settings contain stable schema with prompts section and default templates.
# @PRE llm_settings is dictionary-like value or None.
# @POST Returned dict contains prompts with all required template keys.
# @RELATION DEPENDS_ON -> LLMProviderService
def normalize_llm_settings(llm_settings: Any) -> dict[str, Any]:
normalized: dict[str, Any] = {
"providers": [],
"default_provider": "",
"prompts": {},
"provider_bindings": {},
**DEFAULT_LLM_ASSISTANT_SETTINGS,
}
if isinstance(llm_settings, dict):
normalized.update(
{
k: v
for k, v in llm_settings.items()
if k
in (
"providers",
"default_provider",
"prompts",
"provider_bindings",
"assistant_planner_provider",
"assistant_planner_model",
)
}
)
prompts = normalized.get("prompts") if isinstance(normalized.get("prompts"), dict) else {}
merged_prompts = deepcopy(DEFAULT_LLM_PROMPTS)
merged_prompts.update({k: v for k, v in prompts.items() if isinstance(v, str) and v.strip()})
normalized["prompts"] = merged_prompts
bindings = normalized.get("provider_bindings") if isinstance(normalized.get("provider_bindings"), dict) else {}
merged_bindings = deepcopy(DEFAULT_LLM_PROVIDER_BINDINGS)
merged_bindings.update({k: v for k, v in bindings.items() if isinstance(v, str)})
normalized["provider_bindings"] = merged_bindings
for key, default_value in DEFAULT_LLM_ASSISTANT_SETTINGS.items():
value = normalized.get(key, default_value)
normalized[key] = value.strip() if isinstance(value, str) else default_value
return normalized
# #endregion normalize_llm_settings
# #region is_multimodal_model [C:3] [TYPE Function]
# @BRIEF Heuristically determine whether model supports image input required for dashboard validation.
# @DEPRECATED Use the explicit `db_provider.is_multimodal` flag instead (see migration 9f8e7d6c5b4a).
# @RATIONALE Added import warnings + warnings.warn(DeprecationWarning) to is_multimodal_model as a deprecation shim.
# Replaced by an explicit boolean flag on `LLMProvider` that users control via checkbox in UI.
# This function is retained only as a backward-compatibility shim for the Alembic migration
# backfill and must not be imported in new production code.
# @REJECTED Keeping the function as a backward-compatibility shim; do not use for new validation.
# @PRE model_name may be empty or mixed-case.
# @POST Returns True when model likely supports multimodal input.
# @RELATION DEPENDS_ON -> LLMProviderService
def is_multimodal_model(model_name: str, provider_type: str | None = None) -> bool:
warnings.warn(
"is_multimodal_model is deprecated; use LLMProvider.is_multimodal instead",
DeprecationWarning,
stacklevel=2,
)
token = (model_name or "").strip().lower()
if not token:
return False
text_only_markers = (
"text-only",
"embedding",
"rerank",
"whisper",
"tts",
"transcribe",
)
if any(marker in token for marker in text_only_markers):
return False
multimodal_markers = (
"gpt-4o",
"gpt-4.1",
"vision",
"vl",
"gemini",
"claude-3",
"claude-sonnet-4",
"omni",
"multimodal",
"pixtral",
"llava",
"internvl",
"qwen-vl",
"qwen2-vl",
)
return any(marker in token for marker in multimodal_markers)
# #endregion is_multimodal_model
# #region resolve_bound_provider_id [C:3] [TYPE Function]
# @BRIEF Resolve provider id configured for a task binding with fallback to default provider.
# @PRE llm_settings is normalized or raw dict from config.
# @POST Returns configured provider id or fallback id/empty string when not defined.
# @RELATION DEPENDS_ON -> LLMProviderService
def resolve_bound_provider_id(llm_settings: Any, task_key: str) -> str:
normalized = normalize_llm_settings(llm_settings)
bindings = normalized.get("provider_bindings", {})
bound = bindings.get(task_key)
if isinstance(bound, str) and bound.strip():
return bound.strip()
default_provider = normalized.get("default_provider", "")
return default_provider.strip() if isinstance(default_provider, str) else ""
# #endregion resolve_bound_provider_id
# #region render_prompt [C:3] [TYPE Function]
# @BRIEF Render prompt template using deterministic placeholder replacement with graceful fallback.
# @PRE template is a string and variables values are already stringifiable.
# @POST Returns rendered prompt text with known placeholders substituted. Warns about unfilled placeholders.
# @RELATION DEPENDS_ON -> LLMProviderService
def render_prompt(template: str, variables: dict[str, Any]) -> str:
rendered = template
for key, value in variables.items():
rendered = rendered.replace("{" + key + "}", str(value))
# Warn about unfilled placeholders that would be sent to LLM
import re
unfilled = re.findall(r'\{(\w+)\}', rendered)
if unfilled:
logger.warning(
f"[render_prompt] Unfilled placeholders in rendered prompt: {unfilled}"
)
return rendered
# #endregion render_prompt
# #endregion llm_prompt_templates