kali20gakki/msAgent / skills/quantizer/quantization-expert-experience-tuning-rules
quantization-expert-experience-tuning-rules
Verified量化专家经验调优库:按「L1 通用专家调优意见 → L2 结构化专家调优意见 → L3 模型专属专家调优意见」三级递进, 输出量化精度调优的完整手段(离群值抑制 / 量化方法·粒度·对称性选型 / 校准集调整 / 敏感层回退 / 模型专属策略), 每条结论附专家原因分析与专家意见可信度等级。 本 Skill 是专家经验库,负责说明可选离群值抑制算法并询问用户选择哪些算法,以及回答「怎么调、哪些层需要回退及为什么」;不运行离群值抑制 processor,不修改 YAML、不执行量化、不做 EP 检查 / 服务化 / 任务评测。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.8 KB |
| cross-model-pitfalls.md | markdown | 8.3 KB |
| models/README.md | markdown | 1.3 KB |
| models/deepseek/deepseek-v3.md | markdown | 2.9 KB |
| models/deepseek/deepseek-v4.md | markdown | 2.5 KB |
| models/glm/glm-4-moe.md | markdown | 1.3 KB |
| models/glm/glm-5.md | markdown | 1.4 KB |
| models/minimax/minimax-m.md | markdown | 1.6 KB |
| scripts/README.md | markdown | 3.1 KB |
| scripts/apply_rollback.py | script | 10.1 KB |
| scripts/inspect_model_structure.py | script | 7.1 KB |
| scripts/yaml_utils.py | script | 4.2 KB |
| structure-family-pitfalls.md | markdown | 14.2 KB |
scripts/yaml_utils.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 -*-
"""Thin YAML helpers used by the practice-dit skill.
Practice YAMLs are *small* (one ``spec`` block, one ``process`` list, one
``exclude`` list) — we deliberately avoid pulling in msmodelslim's full
yaml_validation_helpers stack so the skill has no transitive dependency on
the quant framework being installed. ``PyYAML`` is the only requirement.
"""
from __future__ import annotations
import sys
from pathlib import Path
from typing import Any, Dict
def _yaml():
try:
import yaml # type: ignore
return yaml
except ImportError as exc: # pragma: no cover — env-mismatch guard
raise RuntimeError(
"PyYAML is required by the quantization experience library. "
"Install it via `pip install pyyaml`."
) from exc
def load_yaml(path: str | Path) -> Dict[str, Any]:
"""Load a YAML file into a Python dict. Always returns ``{}`` for empty files."""
p = Path(path).expanduser().resolve()
if not p.is_file():
raise FileNotFoundError(f"YAML not found: {p}")
text = p.read_text(encoding="utf-8")
if not text.strip():
return {}
parsed = _yaml().safe_load(text)
return parsed or {}
def dump_yaml(obj: Dict[str, Any], path: str | Path, *, sort_keys: bool = False) -> None:
"""Dump a Python dict to YAML with a stable style (block, indent=2)."""
p = Path(path).expanduser().resolve()
p.parent.mkdir(parents=True, exist_ok=True)
text = _yaml().safe_dump(
obj,
sort_keys=sort_keys,
allow_unicode=True,
default_flow_style=False,
indent=2,
width=120,
)
p.write_text(text, encoding="utf-8")
def find_exclude_target(process: list) -> int | None:
"""Return index of the first ``process`` step whose ``exclude`` is a list, or ``None``.
Shared by ``merge_exclude`` (read-only) and ``apply_rollback`` (needs the same
index to write back the merged list without re-scanning).
"""
for idx, step in enumerate(process):
if isinstance(step, dict) and isinstance(step.get("exclude"), list):
return idx
return None
def merge_exclude(
base_practice: Dict[str, Any],
extras: list[str],
) -> tuple[list[str], list[str]]:
"""Append ``extras`` to the first ``spec.process[*].exclude`` list.
Returns ``(merged, original)`` so callers that need to compute the
``appended`` delta (e.g. ``apply_rollback``) don't have to re-locate the
target step. Honors the existing ``exclude`` list — duplicates are dropped,
and the *original order* of the base list is preserved so YAML diffs stay
readable. Does **not** mutate the input dict.
"""
spec = base_practice.get("spec")
if not isinstance(spec, dict):
raise ValueError("base_practice has no `spec` mapping")
process = spec.get("process")
if not isinstance(process, list) or not process:
raise ValueError("base_practice.spec.process must be a non-empty list")
target_idx = find_exclude_target(process)
if target_idx is None:
raise ValueError(
"No process step with `exclude` list found in base_practice; "
"expert_rollback requires an existing exclude list to append to."
)
base_exclude = list(process[target_idx]["exclude"])
seen = set(base_exclude)
merged: list[str] = list(base_exclude)
for entry in extras:
if entry and entry not in seen:
merged.append(entry)
seen.add(entry)
return merged, base_exclude
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 470k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |