kali20gakki/msAgent / skills/quantizer/quant-tuning-history-append-dit
quant-tuning-history-append-dit
VerifiedAppend a DiT-tuning round to history.yaml. DiT-side counterpart of the LLM-path accuracy_append.py; records practice.md5, inference_outputs, fp_baseline_outputs, and the scoring fields (scores / overall_score / loss_vs_baseline / is_satisfied) populated by quant-tuning-score-dit. Idempotent on practice_id.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.0 KB |
| scripts/README.md | markdown | 1.8 KB |
| scripts/append.py | script | 7.5 KB |
scripts/append.py
View on GitHub# -------------------------------------------------------------------------
# This file is part of the MindStudio project.
# Copyright (c) 2026 Huawei Technologies Co.,Ltd.
#
# MindStudio is licensed under Mulan PSL v2.
# You can use this software according to the terms and conditions of the Mulan PSL v2.
# You may obtain a copy of Mulan PSL v2 at:
#
# http://license.coscl.org.cn/MulanPSL2
#
# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
# EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
# MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
# See the Mulan PSL v2 for more details.
# -------------------------------------------------------------------------
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""Append a single DiT-tuning round to ``{workdir}/history/history.yaml``.
Mirrors the LLM-path ``accuracy_append.py`` shape, but writes DiT-specific
fields (``inference_outputs``, ``fp_baseline_outputs``, and the scoring
fields ``scores`` / ``overall_score`` / ``loss_vs_baseline`` /
``is_satisfied`` populated by ``quant-tuning-score-dit``) as
documented in
``msagent/skills/quantization-accuracy-tuning-orchestrator/references/output_format.md §3``.
Behaviour:
* Idempotent on identical ``practice_id``: re-running with the same id
overwrites the previous entry instead of duplicating it.
* Auto-creates ``history/`` directory if missing.
* Writes UTF-8 YAML with stable ordering so human diffs stay readable.
* Does not touch LLM/VLM ``records`` — appends a separate DiT section when
the file already contains LLM data (see ``--append-as`` flag).
"""
from __future__ import annotations
import argparse
import json
import sys
from pathlib import Path
from typing import Any, Dict, List, Optional
# Bootstrap: common helpers + sibling yaml_utils.
_HERE = Path(__file__).resolve().parent
_REPO_SKILLS = _HERE.resolve().parents[1] # .../quantizer/
_COMMON = _REPO_SKILLS / "msmodelslim-tools-common" / "scripts"
# DiT v3: yaml_utils now lives under the experience library scripts/ folder.
_PRACTICE = _REPO_SKILLS / "quantization-expert-experience-tuning-rules" / "scripts"
for _p in (_COMMON, _PRACTICE):
if str(_p) not in sys.path:
sys.path.insert(0, str(_p))
from script_utils import file_md5, now_iso, upsert_yaml_section # noqa: E402
from yaml_utils import dump_yaml, load_yaml # noqa: E402
# ---------------------------------------------------------------------------
# Core
# ---------------------------------------------------------------------------
def append_dit_record(
history_path: str | Path,
*,
practice_id: str,
practice_path: str | Path,
inference_outputs: List[str],
fp_baseline_outputs: Optional[List[str]] = None,
scores: Optional[Dict[str, float]] = None,
overall_score: Optional[float] = None,
loss_vs_baseline: Optional[float] = None,
is_satisfied: Optional[bool] = None,
append_as: str = "dit_records",
) -> Dict[str, Any]:
"""Append (or upsert) one DiT record into ``history.yaml``.
Args:
history_path: ``{workdir}/history/history.yaml``.
practice_id: Unique ID for this round (``dit-round-2``).
practice_path: ``{workdir}/round_{N}/practice.yaml`` (md5 auto-computed).
inference_outputs: List of video / image paths produced this round.
fp_baseline_outputs: List of FP baseline paths (when ``quant-tuning-evaluate`` DiT workflow ran in FP baseline mode with --ckpt_dir → FP weights).
scores: Per-dimension score dict from ``quant-tuning-score-dit``.
overall_score: Aggregated VBench overall_score (Quality/Semantic/Total).
loss_vs_baseline: Quant vs FP overall loss; only populated when baseline-outputs is enabled.
is_satisfied: ``True`` iff ``loss_vs_baseline >= -tolerance``; orchestrator uses this as
the loop exit signal.
append_as: Section name to use in the YAML. Defaults to ``dit_records``.
Returns:
The full record dict that was written, including ``time`` and ``quant_config_md5``.
"""
history_p = Path(history_path).expanduser().resolve()
history_p.parent.mkdir(parents=True, exist_ok=True)
practice_p = Path(practice_path).expanduser().resolve()
if not practice_p.is_file():
raise FileNotFoundError(f"practice_path not found: {practice_p}")
record: Dict[str, Any] = {
"practice_id": practice_id,
"quant_config_md5": file_md5(practice_p),
"time": now_iso(),
"practice_path": str(practice_p),
"inference_outputs": list(inference_outputs),
"fp_baseline_outputs": list(fp_baseline_outputs) if fp_baseline_outputs else None,
"scores": scores,
"overall_score": overall_score,
"loss_vs_baseline": loss_vs_baseline,
"is_satisfied": is_satisfied,
}
existing = load_yaml(history_p) if history_p.is_file() else {}
if not isinstance(existing, dict):
existing = {}
upsert_yaml_section(existing, append_as, record, key_field="practice_id")
dump_yaml(existing, history_p, sort_keys=True)
return record
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def _build_cli_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(
description="Append a DiT-tuning round to history.yaml.",
)
parser.add_argument("--history-path", required=True,
help="Path to history.yaml (will be created if missing).")
parser.add_argument("--practice-id", required=True,
help="Unique ID, e.g. 'dit-round-2'.")
parser.add_argument("--practice-path", required=True,
help="Path to the round's practice.yaml (md5 computed).")
parser.add_argument("--inference-outputs", required=True,
help="Comma-separated list of video/image paths.")
parser.add_argument("--fp-baseline-outputs", default=None,
help="Comma-separated list of FP baseline paths (optional).")
parser.add_argument("--scores-json", default=None,
help="JSON object of per-dimension scores (optional).")
parser.add_argument("--overall-score", type=float, default=None)
parser.add_argument("--loss-vs-baseline", type=float, default=None)
parser.add_argument("--is-satisfied", choices=["true", "false"], default=None)
parser.add_argument("--append-as", default="dit_records",
help="YAML section name (default: dit_records).")
return parser
def _parse_csv(value: Optional[str]) -> Optional[List[str]]:
if value is None:
return None
return [v.strip() for v in value.split(",") if v.strip()]
def main() -> int:
args = _build_cli_parser().parse_args()
scores = None
if args.scores_json:
scores = json.loads(args.scores_json)
is_satisfied = None
if args.is_satisfied is not None:
is_satisfied = args.is_satisfied == "true"
record = append_dit_record(
history_path=args.history_path,
practice_id=args.practice_id,
practice_path=args.practice_path,
inference_outputs=_parse_csv(args.inference_outputs) or [],
fp_baseline_outputs=_parse_csv(args.fp_baseline_outputs),
scores=scores,
overall_score=args.overall_score,
loss_vs_baseline=args.loss_vs_baseline,
is_satisfied=is_satisfied,
append_as=args.append_as,
)
print(json.dumps({"ok": True, "record": record}, ensure_ascii=False))
return 0
if __name__ == "__main__":
sys.exit(main())
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |