DemonGatanjieuandClaude Opus 5.5
cb523930da
- An AI app calling by an older tool list (tool_help or use_tool not offered, use_tool for a tool offered directly, a tool the list lacks) is told so, and the task header shows how to reconnect until the app fetches tools/list again. - add_lora / set_model compare the base models on the model line (model_advice.py; the page reports the line) and add base_model_note. - get_model gives author_examples (sampler, steps, CFG, size, a LoRA's weight from the Civitai record's example pictures) and the version name. - run_workflow says ComfyUI's refusal as which node, what is wrong and the tool that fixes it (run_refusal.py; mcp_run.js passes the refusal as data). - Gates "When I don't answer": wait for me, or go with the recommendation once the waiting time runs out (choices with a recommended option only), handed to the AI in the same call and shown on the stage and under Decisions. - A tool budget test (count, size, description length, required arguments) and the exam tasks with their scorecard (docs/architecture/tool-exam.md, tools/tool_exam.mjs). The tool set's text recommends Full for Antigravity. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
125 lines
6.7 KiB
Python
125 lines
6.7 KiB
Python
"""What Anomalous tells the AI about using a model well (api/model_advice.py): the settings of the
|
||
author's example pictures in get_model, and the note when add_lora or set_model puts a LoRA on a
|
||
main model of another base model."""
|
||
import json
|
||
import os
|
||
import sys
|
||
import tempfile
|
||
import unittest
|
||
from pathlib import Path
|
||
from unittest import mock
|
||
|
||
PLUGIN = Path(__file__).resolve().parents[1]
|
||
sys.path[:0] = [str(PLUGIN), str(PLUGIN.parents[1])]
|
||
|
||
from api import mcp_actions, model_advice # noqa: E402
|
||
|
||
SHA = "abcdef0123" + "0" * 54
|
||
|
||
|
||
def pictures(*metas):
|
||
return {"id": 77, "baseModel": "Illustrious", "images": [{"url": "x", "meta": meta} for meta in metas] + [{"url": "y", "meta": None}]}
|
||
|
||
|
||
class ExampleTests(unittest.TestCase):
|
||
def setUp(self):
|
||
self.folder = tempfile.mkdtemp(prefix="amb_advice_")
|
||
self.model = os.path.join(self.folder, "glow_style.safetensors")
|
||
Path(self.model).write_bytes(b"x")
|
||
|
||
def keep(self, record):
|
||
Path(self.folder, "glow_style.civitai.info").write_text(json.dumps(record), encoding="utf-8")
|
||
|
||
def test_the_authors_settings_summed_up(self):
|
||
self.keep(pictures(
|
||
{"sampler": "Euler a", "steps": 28, "cfgScale": 5, "Size": "832x1216", "Clip skip": "2",
|
||
"resources": [{"type": "lora", "name": "glow_style", "hash": "ABCDEF0123", "weight": 0.8}]},
|
||
{"sampler": "Euler a", "steps": "30", "cfgScale": 6, "width": 832, "height": 1216, "clipSkip": 2,
|
||
"additionalResources": [{"type": "lora", "name": "loras/glow_style.safetensors", "strength": 0.6}]},
|
||
{"sampler": "DPM++ 2M", "scheduler": "karras", "steps": 25, "cfgScale": 5, "Size": "1024x1024",
|
||
"civitaiResources": [{"type": "lora", "modelVersionId": 77, "weight": 1}]},
|
||
))
|
||
found = model_advice.author_examples(self.model, "loras", SHA)
|
||
self.assertEqual(found["pictures"], 3, "pictures without settings are left out")
|
||
self.assertEqual(found["sampler"], "Euler a (2 of 3)")
|
||
self.assertEqual(found["steps"], "25–30")
|
||
self.assertEqual(found["cfg"], "5–6")
|
||
self.assertEqual(found["size"], "832x1216 (2 of 3)")
|
||
self.assertEqual(found["clip_skip"], "2")
|
||
self.assertEqual(found["lora_weight"], "0.6–1", "by hash, by file name and by Civitai version")
|
||
self.assertEqual(found["scheduler"], "karras (1 of 3)", "counted against all the pictures")
|
||
|
||
def test_the_lora_most_pictures_name_is_this_one(self):
|
||
lora = lambda name, weight: {"type": "lora", "name": name, "weight": weight}
|
||
self.keep(pictures(
|
||
{"sampler": "Undefined", "resources": [lora("authors_own_name", 0.7), lora("other_style", 1)]},
|
||
{"resources": [lora("authors_own_name", 0.9)]},
|
||
{"resources": [lora("authors_own_name", 0.8), lora("third", 1)]},
|
||
))
|
||
found = model_advice.author_examples(self.model, "loras")
|
||
self.assertEqual(found["lora_weight"], "0.7–0.9", "named in every picture: the author's name for this file")
|
||
self.assertNotIn("sampler", found, "Civitai's 'Undefined' is no sampler")
|
||
self.keep(pictures({"resources": [lora("a", 0.5), lora("b", 1)]}, {"resources": [lora("a", 0.5), lora("b", 1)]}))
|
||
self.assertNotIn("lora_weight", model_advice.author_examples(self.model, "loras"), "two named as often: no guess")
|
||
|
||
def test_nothing_kept_says_nothing(self):
|
||
self.assertEqual(model_advice.author_examples(self.model, "loras"), {})
|
||
self.keep(pictures())
|
||
self.assertEqual(model_advice.author_examples(self.model, "checkpoints"), {})
|
||
|
||
def test_a_checkpoint_has_no_weight(self):
|
||
self.keep(pictures({"sampler": "Euler", "steps": 20}))
|
||
self.assertNotIn("lora_weight", model_advice.author_examples(self.model, "checkpoints"))
|
||
|
||
|
||
class BaseModelTests(unittest.TestCase):
|
||
def bases(self, **by_path):
|
||
get = mock.patch.object(model_advice.metadata, "get_metadata", side_effect=lambda path: {"baseModel": by_path.get(path, "")})
|
||
full = mock.patch.object(model_advice.folder_paths, "get_full_path",
|
||
side_effect=lambda kind, name: name if kind in ("checkpoints", "loras") and name in by_path else None)
|
||
for patcher in (get, full):
|
||
patcher.start()
|
||
self.addCleanup(patcher.stop)
|
||
|
||
def test_families(self):
|
||
self.assertEqual(model_advice.family("SDXL 1.0"), model_advice.family("Pony"))
|
||
self.assertEqual(model_advice.family("Illustrious"), "SDXL")
|
||
self.assertEqual(model_advice.family("SD 1.5"), "SD 1.x")
|
||
self.assertEqual(model_advice.family("Flux.1 D"), model_advice.family("Chroma"))
|
||
self.assertEqual(model_advice.family("Other"), "")
|
||
|
||
def test_another_family_will_not_work(self):
|
||
self.bases(lora="SD 1.5", main="SDXL 1.0")
|
||
note = model_advice.lora_note("lora", {"model": "main"})
|
||
self.assertIn("clash", note)
|
||
self.assertIn("base_model='SDXL 1.0'", note)
|
||
|
||
def test_a_branch_of_the_same_family_may_look_off(self):
|
||
self.bases(lora="Pony", main="Illustrious")
|
||
self.assertIn("same family", model_advice.lora_note("lora", {"model": "main"}))
|
||
|
||
def test_the_same_base_or_an_unknown_one_says_nothing(self):
|
||
self.bases(lora="SDXL 1.0", main="sdxl 1.0", plain="", flux_d="Flux.1 D", flux_s="Flux.1 S")
|
||
self.assertIsNone(model_advice.lora_note("lora", {"model": "main"}))
|
||
self.assertIsNone(model_advice.lora_note("plain", {"model": "main"}))
|
||
self.assertIsNone(model_advice.lora_note("lora", {"model": "not_there"}))
|
||
self.assertIsNone(model_advice.lora_note("flux_d", {"model": "flux_s"}))
|
||
|
||
def test_add_lora_and_set_model_answer_with_the_note(self):
|
||
self.bases(**{"old.safetensors": "SD 1.5", "main.safetensors": "SDXL 1.0"})
|
||
line = {"model": "main.safetensors", "loras": ["old.safetensors"]}
|
||
with mock.patch.object(mcp_actions, "_resolve_model", return_value=("loras", 0, "", "old.safetensors")), \
|
||
mock.patch.object(mcp_actions, "trains_text_encoder", return_value=False), \
|
||
mock.patch.object(mcp_actions, "_page", return_value={"node_id": 9, "line": line}):
|
||
answer = mcp_actions.add_lora({}, "m")
|
||
self.assertIn("clash", answer["base_model_note"])
|
||
self.assertNotIn("line", answer, "the line is the server's, not the AI's")
|
||
with mock.patch.object(mcp_actions, "_resolve_model", return_value=("checkpoints", 0, "", "main.safetensors")), \
|
||
mock.patch.object(mcp_actions, "_page", return_value={"node_id": 4, "line": dict(line)}):
|
||
answer = mcp_actions.set_model({}, 4, "m")
|
||
self.assertIn("clash", answer["base_model_note"], "a new main model is checked against the LoRAs after it")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
unittest.main()
|