feat(recipes): match civitai image hash sections against local hash cache

This commit is contained in:
Will Miao
2026-08-08 22:12:47 +08:00
parent cfc95ee02a
commit 196c8ffc3e
2 changed files with 435 additions and 46 deletions

View File

@@ -4,7 +4,7 @@ import json
import logging
from typing import Dict, Any, Union
from ..base import RecipeMetadataParser
from ..constants import GEN_PARAM_KEYS
from ..constants import GEN_PARAM_KEYS, VALID_LORA_TYPES
from ...services.metadata_service import get_default_metadata_provider
from ...config import config
@@ -216,7 +216,8 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
# Try to look up base model from the checkpoint hash
cp_hash = checkpoint_entry.get("hash")
if cp_hash and metadata_provider:
local_cached = local_cache.get(cp_hash) if local_cache else None
# local_cache keys are stored lowercase
local_cached = local_cache.get(cp_hash.lower()) if local_cache else None
if local_cached:
self._populate_entry_from_cache(
checkpoint_entry, local_cached
@@ -294,8 +295,15 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
# Try to get info from Civitai if hash is available
if lora_hash and metadata_provider:
local_cached = local_cache.get(lora_hash) if local_cache else None
# local_cache keys are stored lowercase
local_cached = local_cache.get(lora_hash.lower()) if local_cache else None
if local_cached:
cached_type = self._cache_item_model_type(local_cached)
if cached_type and cached_type not in VALID_LORA_TYPES:
logger.debug(
f"Skipping non-LoRA cache item for hash {lora_hash}"
)
continue
self._populate_entry_from_cache(
lora_entry, local_cached
)
@@ -304,6 +312,12 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
added_loras[str(lora_entry["id"])] = len(
result["loras"]
)
# Mirror base.py:150-151 counts for API-path loras
bm = local_cached.get("base_model") or ""
if bm:
base_model_counts[bm] = base_model_counts.get(
bm, 0
) + 1
else:
try:
civitai_info = (
@@ -649,6 +663,23 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
}
if metadata_provider:
# local_cache keys are stored lowercase
local_cached = local_cache.get(lora_hash.lower()) if local_cache else None
if local_cached:
cached_type = self._cache_item_model_type(local_cached)
if cached_type and cached_type not in VALID_LORA_TYPES:
logger.debug(
f"Skipping non-LoRA cache item for hash {lora_hash}"
)
continue
self._populate_entry_from_cache(lora_entry, local_cached)
# Mirror base.py:150-151 counts for API-path loras
bm = local_cached.get("base_model") or ""
if bm:
base_model_counts[bm] = base_model_counts.get(bm, 0) + 1
if "id" in lora_entry and lora_entry["id"]:
added_loras[str(lora_entry["id"])] = len(result["loras"])
else:
try:
civitai_info = await metadata_provider.get_model_by_hash(
lora_hash
@@ -711,6 +742,25 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
# Try to get info from Civitai if hash is available
if lora_entry["hash"] and metadata_provider:
# local_cache keys are stored lowercase
local_cached = local_cache.get(lora_hash.lower()) if local_cache else None
if local_cached:
cached_type = self._cache_item_model_type(local_cached)
if cached_type and cached_type not in VALID_LORA_TYPES:
logger.debug(
f"Skipping non-LoRA cache item for hash {lora_hash}"
)
lora_index += 1
continue # Skip non-LoRA cache items
self._populate_entry_from_cache(lora_entry, local_cached)
# Mirror base.py:150-151 counts for API-path loras
bm = local_cached.get("base_model") or ""
if bm:
base_model_counts[bm] = base_model_counts.get(bm, 0) + 1
# If we have a version ID from Civitai, track it for deduplication
if "id" in lora_entry and lora_entry["id"]:
added_loras[str(lora_entry["id"])] = len(result["loras"])
else:
try:
civitai_info = await metadata_provider.get_model_by_hash(
lora_hash
@@ -795,3 +845,14 @@ class CivitaiApiMetadataParser(RecipeMetadataParser):
base_model = cache_item.get("base_model", "")
if base_model:
entry["baseModel"] = base_model
@staticmethod
def _cache_item_model_type(cache_item: dict[str, Any]) -> str:
"""Lowercased civitai.model.type of a cache item, or '' when unknown."""
civ = cache_item.get("civitai")
if not isinstance(civ, dict):
return ""
model_info = civ.get("model")
if not isinstance(model_info, dict):
return ""
return (model_info.get("type") or "").lower()

View File

@@ -570,3 +570,331 @@ async def test_backfill_lora_cache_item_without_sha256_does_not_crash():
)
assert result["thumbnailUrl"].startswith("https://image.civitai.com/")
class _RecordingProvider:
"""Metadata provider stub that records get_model_by_hash calls.
With raise_on_call=True any hash lookup fails the test loudly — used to
prove that local_cache hits skip the CivitAI API. Otherwise the result
is returned for the cache-miss path.
"""
def __init__(self, result=None, raise_on_call=False):
self.hash_calls = []
self._result = result
self._raise_on_call = raise_on_call
async def get_model_by_hash(self, model_hash):
self.hash_calls.append(model_hash)
if self._raise_on_call:
raise AssertionError(
f"get_model_by_hash should not be called on local_cache hit, got {model_hash}"
)
return self._result
async def get_model_version_info(self, version_id):
return None, "Model not found"
def _cache_item(name="Local Style", base_model="SDXL 1.0", model_type="LORA"):
"""Build a scanner cache item shaped like the local hash cache values."""
return {
"file_path": f"/loras/{name.lower().replace(' ', '_')}.safetensors",
"file_name": name,
"sha256": "aabbccddeeff00112233445566778899aabbccddeeff00112233445566778899",
"autov3": "",
"preview_url": "/previews/style.png",
"base_model": base_model,
"civitai": {
"id": 300,
"modelId": 400,
"name": "v1",
"model": {"name": name, "type": model_type},
},
}
def _make_lora_info(base_model="SDXL 1.0"):
"""Civitai response for a lora hash lookup on the cache-miss path."""
return {
"id": 300,
"modelId": 400,
"model": {"name": "Style LoRA", "type": "lora"},
"name": "v1",
"images": [{"url": "https://image.civitai.com/lora/original=true"}],
"baseModel": base_model,
"downloadUrl": "https://civitai.com/api/download/300",
"files": [
{
"type": "Model",
"primary": True,
"sizeKB": 512,
"name": "style.safetensors",
"hashes": {"SHA256": "ff00112233445566778899aabbccddeeff00112233445566778899aabbccddee"},
}
],
}
async def _parse_with_cache(monkeypatch, provider, metadata, local_cache=None):
"""Run parse_metadata with a fixed metadata provider and optional local_cache."""
async def fake_metadata_provider():
return provider
monkeypatch.setattr(
"py.recipes.parsers.civitai_image.get_default_metadata_provider",
fake_metadata_provider,
)
parser = CivitaiApiMetadataParser()
return await parser.parse_metadata(metadata, local_cache=local_cache)
@pytest.mark.asyncio
async def test_local_cache_hashes_section_populates_from_cache_and_skips_api(monkeypatch):
"""Hashes-section lora whose hash is a local_cache key is populated from
the cache and the metadata provider is never consulted."""
provider = _RecordingProvider(raise_on_call=True)
item = _cache_item(name="Local Style", base_model="SDXL 1.0")
local_cache = {"a1b2c3d4e5f6": item}
metadata = {"hashes": {"LORA:Local Style": "A1B2C3D4E5F6"}}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
lora = result["loras"][0]
assert lora["existsLocally"] is True
assert lora["localPath"] == item["file_path"]
assert lora["hash"] == item["sha256"]
assert lora["name"] == "Local Style"
@pytest.mark.asyncio
async def test_local_cache_lora_n_section_populates_from_cache_and_skips_api(monkeypatch):
"""Lora_N section lora whose hash is a local_cache key is populated from
the cache and the metadata provider is never consulted."""
provider = _RecordingProvider(raise_on_call=True)
item = _cache_item(name="Lora N Style", base_model="SDXL 1.0")
local_cache = {"abc123def456": item}
metadata = {
"Lora_0 Model hash": "ABC123DEF456",
"Lora_0 Model name": "Lora N Style",
"Lora_0 Strength model": 0.7,
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
lora = result["loras"][0]
assert lora["existsLocally"] is True
assert lora["localPath"] == item["file_path"]
assert lora["weight"] == 0.7
assert lora["hash"] == item["sha256"]
@pytest.mark.asyncio
async def test_local_cache_uppercase_hash_matches_lowercase_key(monkeypatch):
"""Resources lora with an UPPERCASE hash still matches the lowercase key."""
provider = _RecordingProvider(raise_on_call=True)
sha256 = "aabbccddeeff00112233445566778899aabbccddeeff00112233445566778899"
item = _cache_item(name="Upper Case", base_model="SDXL 1.0")
local_cache = {sha256: item}
metadata = {
"resources": [
{"hash": sha256.upper(), "name": "Upper Case", "type": "lora", "weight": 0.5},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
assert result["loras"][0]["existsLocally"] is True
assert result["loras"][0]["hash"] == sha256
@pytest.mark.asyncio
async def test_local_cache_lora_hits_increment_base_model_counts_for_fallback(monkeypatch):
"""When ALL loras hit the cache, base_model_counts is populated so the
max(counts) fallback still resolves the base model (parity with the API path)."""
provider = _RecordingProvider(raise_on_call=True)
item1 = _cache_item(name="Lora One", base_model="SDXL 1.0")
item2 = _cache_item(name="Lora Two", base_model="SDXL 1.0")
local_cache = {"hash1111111111": item1, "hash2222222222": item2}
metadata = {
"resources": [
{"hash": "hash1111111111", "name": "Lora One", "type": "lora"},
{"hash": "hash2222222222", "name": "Lora Two", "type": "lora"},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 2
assert result["base_model"] == "SDXL 1.0"
@pytest.mark.asyncio
async def test_local_cache_checkpoint_hit_does_not_increment_base_model_counts(monkeypatch):
"""Parity pin: a checkpoint cache hit never contributes to base_model_counts.
Scenario 1 sets result["base_model"] directly (like the API path); scenario 2
proves a base_model-less checkpoint added nothing to the counts fallback."""
provider = _RecordingProvider(raise_on_call=True)
cp_item = _cache_item(name="My Checkpoint", base_model="CP Base", model_type="Checkpoint")
lora_item = _cache_item(name="Style LoRA", base_model="Lora Base", model_type="LORA")
metadata1 = {
"resources": [
{"hash": "cp1234567890", "name": "My Checkpoint", "type": "model"},
{"hash": "lora123456789", "name": "Style LoRA", "type": "lora"},
],
}
result1 = await _parse_with_cache(
monkeypatch,
provider,
metadata1,
local_cache={"cp1234567890": cp_item, "lora123456789": lora_item},
)
assert result1["base_model"] == "CP Base"
cp_no_bm = _cache_item(name="No Bm Checkpoint", base_model="", model_type="Checkpoint")
metadata2 = {
"resources": [
{"hash": "cp9999999999", "name": "No Bm Checkpoint", "type": "model"},
{"hash": "lora123456789", "name": "Style LoRA", "type": "lora"},
],
}
result2 = await _parse_with_cache(
monkeypatch,
provider,
metadata2,
local_cache={"cp9999999999": cp_no_bm, "lora123456789": lora_item},
)
assert result2["base_model"] == "Lora Base"
@pytest.mark.asyncio
async def test_local_cache_type_gate_skips_checkpoint_cache_item_in_lora_section(monkeypatch):
"""A cache item whose civitai.model.type is a checkpoint is skipped in a
lora section — no entry is appended for it."""
provider = _RecordingProvider(raise_on_call=True)
cp_item = _cache_item(name="Disguised Checkpoint", base_model="CP Base", model_type="Checkpoint")
lora_item = _cache_item(name="Real Lora", base_model="Lora Base", model_type="LORA")
metadata = {
"resources": [
{"hash": "cp1111111111", "name": "Disguised Checkpoint", "type": "lora"},
{"hash": "lora111111111", "name": "Real Lora", "type": "lora"},
],
}
result = await _parse_with_cache(
monkeypatch,
provider,
metadata,
local_cache={"cp1111111111": cp_item, "lora111111111": lora_item},
)
assert provider.hash_calls == []
assert [l["name"] for l in result["loras"]] == ["Real Lora"]
@pytest.mark.asyncio
async def test_local_cache_type_gate_accepts_uppercase_lora_type(monkeypatch):
"""Cache items storing the type as UPPERCASE 'LORA' must not be skipped
(false-skip guard — stored types are verbatim, VALID_LORA_TYPES is lowercase)."""
provider = _RecordingProvider(raise_on_call=True)
item = _cache_item(name="Uppercase Lora", base_model="SDXL 1.0", model_type="LORA")
local_cache = {"upcasehash12": item}
metadata = {
"resources": [
{"hash": "upcasehash12", "name": "Uppercase Lora", "type": "lora"},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
assert result["loras"][0]["existsLocally"] is True
@pytest.mark.asyncio
async def test_local_cache_type_gate_accepts_item_without_civitai_type(monkeypatch):
"""Local-only cache items without civitai type info are treated as valid."""
provider = _RecordingProvider(raise_on_call=True)
item = {
"file_path": "/loras/local_only.safetensors",
"file_name": "Local Only",
"sha256": "00112233445566778899aabbccddeeff00112233445566778899aabbccddeeff",
"preview_url": "/previews/local_only.png",
"base_model": "SDXL 1.0",
}
local_cache = {"localonly123": item}
metadata = {
"resources": [
{"hash": "localonly123", "name": "Local Only", "type": "lora"},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
assert result["loras"][0]["existsLocally"] is True
@pytest.mark.asyncio
async def test_local_cache_miss_calls_provider(monkeypatch):
"""A hash absent from local_cache falls through to the provider as before."""
lora_info = _make_lora_info(base_model="SDXL 1.0")
provider = _RecordingProvider(result=lora_info)
metadata = {
"resources": [
{"hash": "missedhash123", "name": "Missed LoRA", "type": "lora", "weight": 0.8},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache={})
assert provider.hash_calls == ["missedhash123"]
assert len(result["loras"]) == 1
assert result["loras"][0]["name"] == "Style LoRA"
@pytest.mark.asyncio
async def test_local_cache_dedup_same_hash_produces_one_entry_on_hit(monkeypatch):
"""Repeated same-hash resources produce a single entry on the cache-hit path."""
provider = _RecordingProvider(raise_on_call=True)
item = _cache_item(name="Dedup Lora", base_model="SDXL 1.0")
local_cache = {"deduphash123": item}
metadata = {
"resources": [
{"hash": "deduphash123", "name": "Dedup Lora", "type": "lora"},
{"hash": "deduphash123", "name": "Dedup Lora", "type": "lora"},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache=local_cache)
assert provider.hash_calls == []
assert len(result["loras"]) == 1
@pytest.mark.asyncio
async def test_local_cache_dedup_same_hash_produces_one_entry_on_miss(monkeypatch):
"""Repeated same-hash resources produce a single entry on the cache-miss path."""
provider = _RecordingProvider(result=_make_lora_info(base_model="SDXL 1.0"))
metadata = {
"resources": [
{"hash": "missdedup123", "name": "Dedup Miss", "type": "lora"},
{"hash": "missdedup123", "name": "Dedup Miss", "type": "lora"},
],
}
result = await _parse_with_cache(monkeypatch, provider, metadata, local_cache={})
assert provider.hash_calls == ["missdedup123"]
assert len(result["loras"]) == 1