Files
ComfyUI-Lora-Manager/py/services/model_sources/base.py
T
Will Miao 35b291ab19 feat(modelscope): read the model-detail API for card extras
ModelScope's model card is not just README.md: the author's summary
(Description), the site-curated tags (OfficialTags), the internal
architecture enums (VisionFoundation/SubVisionFoundation) and — per
published version — the model filenames with that file's example images
(coverImages) and trigger words all live in the model-detail API.
AIGC repositories there frequently ship an auto-generated boilerplate
README and put the only useful text in Description, so reading just the
README yielded almost nothing.

Add `ModelSource.fetch_model_card_context()` returning a new
`ModelCardContext`, implemented by ModelScopeSource against the public
(no API key) detail endpoint. Example images are matched to the model's
basename through each version's `stats.fileList`, so every checkpoint in
a collection repository gets its own images rather than a sibling's.

Consume the context in the post-processor:

* example images seed `civitai.images` and, being per-file, take priority
  in the preview fallback chain
* the author summary becomes a paragraph in `modelDescription` and fills
  `civitai.description` when the LLM returns no short description
* site-curated tags are always merged in, which also fixes the official
  `character-enhancement` being dropped by the prompt's no-hyphen rule
* per-file trigger words are used before the repo-wide YAML
  `instance_prompt`
* an explicitly stated strength range is recovered by regex so
  `usage_tips` is populated even without an LLM

The prompt gains a Site-Provided Metadata section so the LLM can prefer
the site's first-hand data over its own guesses.
2026-09-14 20:38:56 +08:00

383 lines
13 KiB
Python

"""Base types for the external model-source provider abstraction.
A *model source* is a third-party site that hosts model files and a model
card (README) describing them — Hugging Face, ModelScope, TensorArt, and
whatever gets added later. Everything the rest of the codebase needs to
know about such a site is expressed by :class:`ModelSource`:
* how to recognise one of its URLs (:meth:`ModelSource.parse`)
* the canonical page URL for a source id (:meth:`ModelSource.canonical_url`)
* how to fetch the model card (:meth:`ModelSource.fetch_model_card`)
* how to fetch the extras that live *outside* the README
(:meth:`ModelSource.fetch_model_card_context`)
* how to turn repository-relative asset paths into absolute URLs
(:meth:`ModelSource.asset_base_url`)
* which capabilities the site actually supports
(``supports_enrichment`` / ``supports_download``)
Keeping this in one place means the agent pipeline, the scanners, and the
HTTP handlers never need site-specific branching.
"""
from __future__ import annotations
import logging
import os
import re
from dataclasses import dataclass, field
from typing import Any, Iterable, Optional
import aiohttp
from ...utils.constants import MODEL_FILE_EXTENSIONS
logger = logging.getLogger(__name__)
#: Shared HTTP timeout for model-card fetches.
HTTP_TIMEOUT = 30
#: User agent used for all model-source HTTP requests.
USER_AGENT = "ComfyUI-LoRA-Manager/1.0"
#: Platform → short prefix used when building version-group keys.
#: ``huggingface`` keeps the historical ``hf:`` prefix for backward
#: compatibility with already-cached group keys.
GROUP_PREFIXES: dict[str, str] = {
"huggingface": "hf",
"modelscope": "ms",
"tensorart": "ta",
}
@dataclass(frozen=True)
class SourceRef:
"""A parsed reference to a model hosted on an external site."""
platform: str
"""Canonical platform id, e.g. ``"huggingface"``."""
source_id: str
"""Site-specific identity, e.g. ``"user/repo"`` or ``"827823520299086029"``."""
url: str
"""Canonical URL of the model page."""
@dataclass
class ModelCardContext:
"""Site-specific extras that accompany a model's README model card.
A model card is not always just ``README.md``. ModelScope, for example,
keeps the author's summary, the site-curated tags, and the per-file
example images in its model-detail API rather than in the repository.
Sources with no such extras return an empty context (the default), so
every field here must be treated as optional by callers.
"""
description: str = ""
"""Author-written summary shown on the model page, outside the README."""
base_model: str = ""
"""Base model as reported by the site (possibly a site-local id)."""
base_model_aliases: list[str] = field(default_factory=list)
"""Other names the site uses for the same base model.
Sites often publish both a link-style id (``krea/Krea-2-Turbo``) and an
internal architecture enum (``KREA_2``). The enum usually normalises
cleanly onto this system's canonical vocabulary, so it is the better
resolution hint for :mod:`py.services.agent.base_model_resolver`.
"""
official_tags: list[str] = field(default_factory=list)
"""Content tags curated by the site itself."""
example_images: list[str] = field(default_factory=list)
"""Absolute URLs of example images for the requested model file."""
trigger_words: list[str] = field(default_factory=list)
"""Trigger words the site records for the requested model file."""
def is_empty(self) -> bool:
"""Return ``True`` when the site contributed nothing extra."""
return not any(
(
self.description,
self.base_model,
self.base_model_aliases,
self.official_tags,
self.example_images,
self.trigger_words,
)
)
class ModelSourceError(Exception):
"""Raised when a model source cannot satisfy a request.
Carries the HTTP status the API handler should answer with, so the
handlers stay free of per-site error mapping.
"""
def __init__(self, message: str, status: int = 502) -> None:
super().__init__(message)
self.status = status
#: Repository ids are always exactly ``owner/name``. Components may contain
#: dots (``black-forest-labs/FLUX.1-dev``) but must not be empty, ``.`` / ``..``,
#: or start with a dot - the id is used as a path segment on disk.
_SOURCE_ID_COMPONENT = re.compile(r"^[A-Za-z0-9_][A-Za-z0-9_.\-]*$")
def is_valid_source_id(source_id: str) -> bool:
"""Return ``True`` when *source_id* is a safe ``owner/name`` repository id."""
if not source_id or not isinstance(source_id, str) or source_id.count("/") != 1:
return False
owner, name = source_id.split("/", 1)
return all(
part and part not in (".", "..") and _SOURCE_ID_COMPONENT.match(part)
for part in (owner, name)
)
async def fetch_text(url: str, *, timeout: int = HTTP_TIMEOUT) -> str:
"""Fetch *url* and return its body as text, or ``""`` on any failure.
Network problems are expected (offline installs, rate limits, dead
repos) and must never bubble up into the pipeline, so every error is
logged at debug level and normalised to an empty string.
"""
try:
async with aiohttp.ClientSession(
headers={"User-Agent": USER_AGENT},
timeout=aiohttp.ClientTimeout(total=timeout),
) as session:
async with session.get(url) as resp:
if resp.status == 200:
return await resp.text()
logger.debug("Fetch %s returned HTTP %s", url, resp.status)
except Exception as exc: # pragma: no cover - network dependent
logger.debug("Failed to fetch %s: %s", url, exc)
return ""
async def fetch_json(
url: str, *, timeout: int = HTTP_TIMEOUT
) -> tuple[int, Any]:
"""Fetch *url* and return ``(status, parsed_body)``.
Unlike :func:`fetch_text` this reports the status, because callers such as
the file-listing endpoints need to distinguish "repo not found" (404) from
a transport failure. ``parsed_body`` is ``None`` when the response is not
JSON or the request failed outright (status ``0``).
"""
try:
async with aiohttp.ClientSession(
headers={"User-Agent": USER_AGENT},
timeout=aiohttp.ClientTimeout(total=timeout),
) as session:
async with session.get(url) as resp:
if resp.status != 200:
return resp.status, None
try:
return resp.status, await resp.json(content_type=None)
except Exception:
return resp.status, None
except Exception as exc: # pragma: no cover - network dependent
logger.debug("Failed to fetch %s: %s", url, exc)
return 0, None
class ModelSource:
"""Description and I/O for one external model hosting site."""
#: Canonical platform id stored in metadata.
platform: str = ""
#: Human-readable name used in UI copy and prompts.
label: str = ""
#: Whether the agent skill can fetch a model card and run AI extraction.
supports_enrichment: bool = False
#: Whether models can be downloaded directly from this site.
supports_download: bool = False
#: Branch used when the caller does not pass an explicit revision.
default_revision: str = ""
#: Sub-directory the "use default paths" template places downloads in.
default_subdir: str = ""
#: Lenient pattern used to recognise URLs already stored in metadata.
#: Captures the site-specific source id in group ``id``.
url_pattern: re.Pattern[str] | None = None
#: Strict pattern used to validate user input. Must match the whole URL.
strict_url_pattern: re.Pattern[str] | None = None
# ------------------------------------------------------------------
# Parsing
# ------------------------------------------------------------------
def parse(self, url: str, *, strict: bool = False) -> Optional[str]:
"""Return the source id contained in *url*, or ``None``.
With ``strict=True`` the URL must match this site's canonical shape
exactly (used when validating what a user pasted); with
``strict=False`` sub-paths such as ``/resolve/main/file.bin`` are
tolerated (used when normalising already-stored values).
"""
if not url or not isinstance(url, str):
return None
candidate = url.strip()
if not candidate:
return None
pattern = self.strict_url_pattern if strict else self.url_pattern
if pattern is None:
return None
match = pattern.match(candidate)
return match.group("id") if match else None
def ref(self, url: str, *, strict: bool = False) -> Optional[SourceRef]:
"""Return a :class:`SourceRef` for *url*, or ``None`` if not ours."""
source_id = self.parse(url, strict=strict)
if not source_id:
return None
return SourceRef(
platform=self.platform,
source_id=source_id,
url=self.canonical_url(source_id),
)
# ------------------------------------------------------------------
# URLs and content
# ------------------------------------------------------------------
def canonical_url(self, source_id: str) -> str:
"""Return the canonical model-page URL for *source_id*."""
raise NotImplementedError
def asset_base_url(self, source_id: str, revision: str = "") -> str:
"""Base URL used to resolve repository-relative asset paths."""
return ""
def group_key(self, source_id: str) -> str:
"""Return the version-group key for *source_id*."""
prefix = GROUP_PREFIXES.get(self.platform, self.platform)
return f"{prefix}:{source_id}"
async def fetch_model_card(self, source_id: str) -> str:
"""Fetch the raw model card (README) markdown for *source_id*."""
return ""
async def fetch_model_card_context(
self, source_id: str, filename: str = ""
) -> ModelCardContext:
"""Return the card extras the site keeps outside the README.
*filename* is the model file's basename (no directory) and selects
the right entry when a repository holds several models. Sites whose
model card is fully described by :meth:`fetch_model_card` need no
override and inherit this empty context.
Implementations must never raise: enrichment treats a missing
context as "the site had nothing extra to say".
"""
return ModelCardContext()
# ------------------------------------------------------------------
# Download support
# ------------------------------------------------------------------
async def list_files(
self, source_id: str, revision: str = ""
) -> list[dict[str, Any]]:
"""List downloadable weight files in *source_id*.
Returns ``[{"filename": <repo-relative path>, "size": <bytes>}]``,
largest first, filtered to :data:`MODEL_FILE_EXTENSIONS`. Sites
without download support return an empty list.
Raises :class:`ModelSourceError` when the repository cannot be read,
so the handler can surface "not found" separately from a transport
failure.
"""
return []
def file_download_url(
self, source_id: str, filename: str, revision: str = ""
) -> str:
"""Return the direct (redirecting) download URL for one file."""
raise ModelSourceError(
f"{self.label or self.platform} does not support downloads", status=400
)
def resolve_revision(self, revision: str = "") -> str:
"""Return *revision*, falling back to this site's default branch."""
return revision or self.default_revision
def page_url_for_file(self, source_id: str, filename: str) -> str:
"""Return the human-facing page for *filename* inside *source_id*."""
return self.canonical_url(source_id)
def __repr__(self) -> str: # pragma: no cover - debugging aid
return f"<ModelSource {self.platform}>"
def clean_source_url(url: Any) -> str:
"""Normalise a stored source URL value into a stripped string."""
if not isinstance(url, str):
return ""
return url.strip()
def filter_weight_files(entries: Iterable[tuple[str, int]]) -> list[dict[str, Any]]:
"""Keep model-weight files from ``(path, size)`` pairs, largest first.
Every site lists a lot more than weights (READMEs, configs, tokenizers,
…); the download picker only ever wants the files ComfyUI can load, which
is exactly :data:`MODEL_FILE_EXTENSIONS`.
"""
files = [
{"filename": path, "size": int(size or 0)}
for path, size in entries
if path and os.path.splitext(path)[1].lower() in MODEL_FILE_EXTENSIONS
]
files.sort(key=lambda entry: entry["size"], reverse=True)
return files
__all__ = [
"GROUP_PREFIXES",
"HTTP_TIMEOUT",
"ModelCardContext",
"ModelSource",
"ModelSourceError",
"SourceRef",
"USER_AGENT",
"clean_source_url",
"fetch_json",
"fetch_text",
"filter_weight_files",
"is_valid_source_id",
]