Text Generation
PEFT
Chinese
English
preference-learning
qlora
agent
personalization
association-engine
Instructions to use feiertu/hermes-association-engine with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use feiertu/hermes-association-engine with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
File size: 25,472 Bytes
98105a3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 | """Hermes CLI — click 壳."""
import json
import os
import signal
import subprocess
import sys
import textwrap
import time
from pathlib import Path
# Fix Unicode display on Windows (GBK terminal)
if sys.platform == "win32":
sys.stdout.reconfigure(encoding="utf-8", errors="replace")
sys.stderr.reconfigure(encoding="utf-8", errors="replace")
import click
from hermes_core.querier import HermesClient
from hermes_core.refiner import refine_scene
from hermes_core.db import init_db, get_record, update_record_state, get_active_records
from hermes_core.types import RecordState, HERMES_DATA_DIR
from hermes_core.config import get_config, save_config, ensure_config_exists, HermesConfig, CONFIG_FILE
# ═══════════════════════════════════════════════════════════════
# 工具函数
# ═══════════════════════════════════════════════════════════════
def _echo_json(data: dict) -> None:
click.echo(json.dumps(data, ensure_ascii=False, indent=2))
def _check_model_cached() -> bool:
"""检查 embedding 模型是否已下载。"""
from hermes_core.embedder import Embedder
return Embedder.is_available()
# ═══════════════════════════════════════════════════════════════
# 主命令组
# ═══════════════════════════════════════════════════════════════
@click.group()
def cli():
"""Hermes 联想引擎 — Agent 偏好学习与共享."""
# ═══════════════════════════════════════════════════════════════
# hermes init — 一键初始化
# ═══════════════════════════════════════════════════════════════
@cli.command()
@click.option("--force", "-f", is_flag=True, help="强制重新初始化")
def init(force):
"""初始化 Hermes:创建配置 → 下载模型 → 验证环境。
首次使用只需运行一次。幂等操作,可安全重复执行。"""
config = ensure_config_exists()
click.echo("╔══════════════════════════════════════════╗")
click.echo("║ Hermes 一键初始化向导 ║")
click.echo("╚══════════════════════════════════════════╝\n")
# Step 1: 配置
click.echo(f"[1/4] 配置文件: {CONFIG_FILE}")
click.echo(f" 数据目录: {config.data_dir}")
click.echo(f" 扫描间隔: {config.scan_interval_seconds}s")
click.echo(f" 匹配阈值: {config.match_threshold}")
# Step 2: 下载模型
click.echo(f"\n[2/4] Embedding 模型: {config.embedding_model}")
if _check_model_cached() and not force:
click.echo(" ✓ 已缓存")
else:
click.echo(" 正在下载(约 420MB,首次约 2-5 分钟)...")
try:
from hermes_core.embedder import Embedder
e = Embedder(config.embedding_model)
_ = e.encode("test") # 触发加载
click.echo(" ✓ 下载完成")
except Exception as exc:
click.echo(f" ✗ 下载失败: {exc}")
click.echo(" 可稍后手动运行: hermes download-models")
# Step 3: 检查训练依赖
click.echo("\n[3/4] QLoRA 训练依赖 (torch + transformers)")
try:
import torch
import transformers
import peft
click.echo(f" ✓ torch {torch.__version__}, transformers {transformers.__version__}")
except ImportError:
click.echo(" ! 未安装(跳过训练的推理模式仍可用)")
click.echo(" 安装训练依赖: pip install -e '.[train]'")
# Step 4: 启动 daemon
click.echo(f"\n[4/4] 下一步")
click.echo(f" hermes start # 后台启动 daemon")
click.echo(f" hermes status # 查看运行状态")
click.echo(f" hermes demo # 运行交互演示")
click.echo(f"\n✅ 初始化完成!")
# ═══════════════════════════════════════════════════════════════
# hermes download-models — 预下载模型
# ═══════════════════════════════════════════════════════════════
@cli.command("download-models")
def download_models():
"""预下载 embedding 模型(避免首次使用时等待)。"""
config = get_config()
click.echo(f"下载 embedding 模型: {config.embedding_model}")
click.echo("(约 420MB,可能需要 2-5 分钟)\n")
try:
from sentence_transformers import SentenceTransformer
# SentenceTransformer 自带进度条
model = SentenceTransformer(config.embedding_model)
_ = model.encode("test")
click.echo("\n✅ 模型下载完成")
except Exception as e:
click.echo(f"\n❌ 下载失败: {e}")
click.echo("请检查网络连接,或设置 HF_ENDPOINT=https://hf-mirror.com 使用镜像")
sys.exit(1)
# ═══════════════════════════════════════════════════════════════
# hermes start / stop / status — daemon 生命周期
# ═══════════════════════════════════════════════════════════════
@cli.command()
def start():
"""后台启动 Hermes Daemon。"""
config = get_config()
pid_file = Path(config.pid_file)
log_file = Path(config.log_file)
if pid_file.exists():
pid = int(pid_file.read_text().strip())
try:
os.kill(pid, 0)
click.echo(f"Daemon 已在运行 (PID: {pid})")
return
except OSError:
pid_file.unlink()
pid_file.parent.mkdir(parents=True, exist_ok=True)
log_file.parent.mkdir(parents=True, exist_ok=True)
log_f = open(str(log_file), "a")
proc = subprocess.Popen(
[sys.executable, "-m", "daemon.daemon"],
stdout=log_f, stderr=log_f,
start_new_session=True,
)
pid_file.write_text(str(proc.pid))
time.sleep(1)
if proc.poll() is None:
click.echo(f"✅ Daemon 已启动 (PID: {proc.pid})")
click.echo(f" 日志: {log_file}")
click.echo(f" 查看状态: hermes status")
else:
click.echo(f"❌ Daemon 启动失败,查看日志: {log_file}")
sys.exit(1)
@cli.command()
def stop():
"""停止 Hermes Daemon。"""
config = get_config()
pid_file = Path(config.pid_file)
if not pid_file.exists():
click.echo("Daemon 未在运行")
return
pid = int(pid_file.read_text().strip())
try:
os.kill(pid, signal.SIGTERM)
for _ in range(10):
try:
os.kill(pid, 0)
time.sleep(0.3)
except OSError:
break
else:
os.kill(pid, signal.SIGKILL)
pid_file.unlink()
click.echo(f"✅ Daemon 已停止 (PID: {pid})")
except OSError:
pid_file.unlink()
click.echo("Daemon 已不在运行")
@cli.command()
@click.option("--user", "-u", default=None, help="按用户过滤")
def status(user):
"""查看 Hermes 运行状态。"""
config = get_config()
pid_file = Path(config.pid_file)
output = {"daemon": "stopped", "pid": None}
if pid_file.exists():
try:
pid = int(pid_file.read_text().strip())
os.kill(pid, 0)
output["daemon"] = "running"
output["pid"] = pid
except OSError:
pid_file.unlink()
# 汇总统计
try:
db_path = Path(config.data_dir) / "users"
if db_path.exists():
users = [d.name for d in db_path.iterdir() if d.is_dir()]
total_records = 0
total_scopes = 0
for uid in users:
conn = init_db(uid)
scopes = conn.execute("SELECT COUNT(*) FROM scopes WHERE status='active'").fetchone()[0]
records = conn.execute("SELECT COUNT(*) FROM records WHERE state='active'").fetchone()[0]
conn.close()
total_scopes += scopes
total_records += records
output["users"] = len(users)
output["active_scopes"] = total_scopes
output["active_records"] = total_records
except Exception:
pass
if user:
conn = init_db(user)
rows = conn.execute(
"SELECT id, label, record_count, coherence, status FROM scopes ORDER BY last_activity DESC"
).fetchall()
conn.close()
output["scopes"] = [dict(r) for r in rows]
_echo_json(output)
# ═══════════════════════════════════════════════════════════════
# hermes demo — 交互体验
# ═══════════════════════════════════════════════════════════════
DEMO_SCENARIOS = [
{
"desc": "后端API开发",
"dims": [{"key": "language", "value": "TypeScript", "context": "未指定语言时默认"},
{"key": "framework", "value": "Express", "context": "默认后端框架"}],
},
{
"desc": "后端API开发",
"dims": [{"key": "language", "value": "TypeScript", "context": "未指定语言时默认"},
{"key": "framework", "value": "Express", "context": "默认后端框架"},
{"key": "database", "value": "PostgreSQL", "context": "默认数据库"}],
},
{
"desc": "前端开发",
"dims": [{"key": "framework", "value": "React", "context": "默认前端框架"},
{"key": "language", "value": "TypeScript", "context": "默认语言"}],
},
{
"desc": "数据分析",
"dims": [{"key": "language", "value": "Python", "context": "默认"},
{"key": "lib", "value": "Pandas", "context": "数据处理"},
{"key": "style", "value": "functional", "context": "函数式风格"}],
},
{
"desc": "周末活动",
"dims": [{"key": "activity", "value": "户外徒步", "context": "周末偏好"}],
},
]
@cli.command()
@click.option("--user", "-u", default="demo_user", help="演示用户 ID")
def demo(user):
"""运行交互演示:模拟 Agent 记录偏好 → 查询偏好。
这是一个沙箱演示,数据写入临时目录,不影响真实数据。"""
import tempfile
config = get_config()
tmp_dir = tempfile.mkdtemp(prefix="hermes_demo_")
# 用临时目录隔离演示数据
os.environ["HERMES_DATA_DIR"] = str(tmp_dir)
from hermes_core.types import HERMES_DATA_DIR as _H
import hermes_core.db as db_m
import hermes_core.trainer as tr_m
db_m.HERMES_DATA_DIR = Path(tmp_dir)
tr_m.HERMES_DATA_DIR = Path(tmp_dir)
click.echo("╔══════════════════════════════════════════════════════╗")
click.echo("║ Hermes 联想引擎 — 交互演示 ║")
click.echo("╚══════════════════════════════════════════════════════╝\n")
click.echo("模拟场景:一个 AI Agent 在与用户对话中记录偏好,")
click.echo("并在后续对话中自动联想这些偏好。\n")
click.echo(f"数据目录: {tmp_dir}\n")
client = HermesClient(user_id=user, agent_id="demo-agent")
# 阶段 1:记录
click.echo("━" * 50)
click.echo("阶段 1:Agent 记录用户偏好 (record_detail)")
click.echo("━" * 50)
for i, s in enumerate(DEMO_SCENARIOS, 1):
click.echo(f"\n 对话 #{i}: 用户提到「{s['desc']}」相关需求")
r = client.record(s["desc"], s["dims"])
dims_str = ", ".join(f"{d['key']}={d['value']}" for d in s["dims"])
click.echo(f" → Agent 记录: [{dims_str}]")
if r["status"] == "recorded":
click.echo(f" 状态: ✓ 已记录 (scope: {r['scope_id'][:12]}...)")
elif r["status"] == "rejected":
click.echo(f" 状态: ✗ 被拒绝 ({r['reason']})")
else:
click.echo(f" 状态: ↻ {r['status']}")
# 阶段 2:查询
click.echo("\n" + "━" * 50)
click.echo("阶段 2:Agent 推理前查询偏好 (query)")
click.echo("━" * 50)
queries = [
"帮我写一个用户登录的REST API",
"做一个数据可视化的Dashboard",
"这周末想出去玩",
]
for q in queries:
click.echo(f"\n 用户: 「{q}」")
result = client.query(q)
if result.matched_scope:
click.echo(f" → 匹配场景: {result.matched_scope.scope_label} "
f"(置信度 {result.matched_scope.confidence:.0%})")
for pref in result.related_preferences:
click.echo(f" - {pref.key}: {pref.value}")
else:
click.echo(f" → 未匹配到已知场景(将使用默认行为)")
if result.alternative_scopes:
alt = result.alternative_scopes[0]
click.echo(f" 最接近: {alt.scope_label} (置信度 {alt.confidence:.0%})")
# 阶段 3:查看数据
click.echo("\n" + "━" * 50)
click.echo("阶段 3:查看训练集状态")
click.echo("━" * 50)
conn = init_db(user)
scopes = conn.execute("SELECT id, label, record_count, coherence FROM scopes WHERE status='active'").fetchall()
conn.close()
click.echo(f"\n 共有 {len(scopes)} 个动态场景:")
for s_c in scopes:
click.echo(f" {s_c['id']} — {s_c['label']} "
f"({s_c['record_count']} 条记录, 内聚度 {s_c['coherence']:.2f})")
click.echo(f"\n 📊 数据保存在: {tmp_dir}(可手动删除)")
click.echo(f" 💡 下次演示: hermes demo --user {user}")
import shutil
shutil.rmtree(tmp_dir, ignore_errors=True)
# ═══════════════════════════════════════════════════════════════
# hermes record — 写入训练样本
# ═══════════════════════════════════════════════════════════════
@cli.command()
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--scope-desc", required=True, help="场景描述")
@click.option("--dimensions", required=True, help='JSON: [{"key":"language","value":"TS"}]')
@click.option("--source-conv", default="", help="来源会话 ID")
@click.option("--source-agent", default="cli", help="来源 agent ID")
def record(user, scope_desc, dimensions, source_conv, source_agent):
"""写入一条训练样本。"""
try:
dims = json.loads(dimensions)
except json.JSONDecodeError:
_echo_json({"status": "error", "reason": "Invalid dimensions JSON"})
sys.exit(1)
client = HermesClient(user_id=user, agent_id=source_agent)
result = client.record(scope_desc, dims, source_conv=source_conv)
_echo_json(result)
# ═══════════════════════════════════════════════════════════════
# hermes refine — 调整场景粒度
# ═══════════════════════════════════════════════════════════════
@cli.command()
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--id", "record_id", required=True, help="记录 ID")
@click.option("--scope-desc", required=True, help="新的场景描述")
@click.option("--direction", type=click.Choice(["narrow", "broaden"]), required=True)
def refine(user, record_id, scope_desc, direction):
"""调整记录的场景粒度。"""
from hermes_core.embedder import Embedder
embedder = Embedder()
result = refine_scene(user, record_id, scope_desc, direction, embedder)
_echo_json(result)
# ═══════════════════════════════════════════════════════════════
# hermes query — 场景识别 + 偏好查询
# ═══════════════════════════════════════════════════════════════
@cli.command()
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--text", "-t", required=True, help="用户输入文本")
def query(user, text):
"""场景识别 + 偏好查询。"""
client = HermesClient(user_id=user, agent_id="cli")
result = client.query(text)
output = {
"matched_scope": None,
"active_loras": [],
"related_preferences": [],
"training_outdated": result.training_outdated,
}
if result.matched_scope:
output["matched_scope"] = {
"scope_id": result.matched_scope.scope_id,
"scope_label": result.matched_scope.scope_label,
"confidence": round(result.matched_scope.confidence, 4),
}
if result.alternative_scopes:
output["alternative_scopes"] = [
{"scope_id": s.scope_id, "scope_label": s.scope_label,
"confidence": round(s.confidence, 4)}
for s in result.alternative_scopes[:3]
]
output["active_loras"] = [
{"scope_id": l.scope_id, "version": l.version, "priority": l.priority}
for l in result.active_loras
]
output["related_preferences"] = [
{"key": p.key, "value": p.value, "source": p.source}
for p in result.related_preferences
]
_echo_json(output)
# ═══════════════════════════════════════════════════════════════
# hermes review — 训练集审核
# ═══════════════════════════════════════════════════════════════
@cli.group()
def review():
"""训练集审核。"""
@review.command("pending")
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--scope", "scope_id", default=None, help="按 scope 过滤")
def review_pending(user, scope_id):
"""列出待审核记录。"""
conn = init_db(user)
if scope_id:
records = get_active_records(conn, scope_id)
else:
from hermes_core.db import get_active_scopes
scopes = get_active_scopes(conn)
records = []
for s in scopes:
records.extend(get_active_records(conn, s.id))
conn.close()
click.echo(json.dumps([
{
"id": r.id, "scope_id": r.scope_id, "scope_label": r.scope_label,
"dimensions": [{"key": d.key, "value": d.value} for d in r.dimensions],
"confidence": r.confidence, "occurrences": r.occurrences,
}
for r in records
], ensure_ascii=False, indent=2))
@review.command("accept")
@click.option("--user", "-u", required=True, help="用户 ID")
@click.argument("record_ids", nargs=-1)
def review_accept(user, record_ids):
"""批量通过记录。"""
conn = init_db(user)
for rid in record_ids:
update_record_state(conn, rid, RecordState.active)
conn.close()
_echo_json({"status": "accepted", "count": len(record_ids)})
@review.command("reject")
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--reason", default="", help="拒绝原因")
@click.argument("record_ids", nargs=-1)
def review_reject(user, reason, record_ids):
"""批量拒绝记录。"""
conn = init_db(user)
for rid in record_ids:
update_record_state(conn, rid, RecordState.rejected)
conn.close()
_echo_json({"status": "rejected", "count": len(record_ids), "reason": reason})
# ═══════════════════════════════════════════════════════════════
# hermes train-status / train — 训练管理
# ═══════════════════════════════════════════════════════════════
@cli.command("train-status")
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--scope", "scope_id", default=None, help="scope ID")
def train_status(user, scope_id):
"""查询训练任务状态。"""
conn = init_db(user)
if scope_id:
rows = conn.execute(
"SELECT * FROM training_runs WHERE scope_id=? ORDER BY version DESC LIMIT 3",
(scope_id,)
).fetchall()
else:
rows = conn.execute(
"SELECT * FROM training_runs ORDER BY started_at DESC LIMIT 10"
).fetchall()
conn.close()
click.echo(json.dumps([
{"id": r["id"], "scope_id": r["scope_id"], "version": r["version"],
"status": r["status"], "started_at": r["started_at"] or "N/A",
"finished_at": r["finished_at"] or "N/A"}
for r in rows
], ensure_ascii=False, indent=2))
@cli.command("train")
@click.option("--user", "-u", required=True, help="用户 ID")
@click.option("--scope", "scope_id", required=True, help="scope ID")
def train(user, scope_id):
"""手动触发训练。"""
_echo_json({
"status": "queued",
"message": f"Training for {scope_id} queued. Use 'train-status' to check progress."
})
# ═══════════════════════════════════════════════════════════════
# hermes config — 查看/修改配置
# ═══════════════════════════════════════════════════════════════
@cli.group()
def config_cmd():
"""查看和修改配置。"""
pass
@config_cmd.command("show")
def config_show():
"""显示当前配置。"""
config = get_config()
_echo_json(config.to_dict())
@config_cmd.command("set")
@click.argument("key")
@click.argument("value")
def config_set(key, value):
"""修改配置项。例如: hermes config set match_threshold 0.65"""
config = get_config()
if key not in HermesConfig.__dataclass_fields__:
click.echo(f"未知配置项: {key}")
click.echo(f"可用配置项: {', '.join(HermesConfig.__dataclass_fields__.keys())}")
sys.exit(1)
field_type = type(getattr(config, key))
try:
setattr(config, key, field_type(value))
except (ValueError, TypeError) as e:
click.echo(f"值类型错误: {e}")
sys.exit(1)
save_config(config)
click.echo(f"✅ {key} = {getattr(config, key)}")
# ═══════════════════════════════════════════════════════════════
# hermes daemon — 启动/停止(兼容旧接口)
# ═══════════════════════════════════════════════════════════════
@cli.command("daemon")
@click.option("--foreground", "-f", is_flag=True, help="前台运行模式")
def daemon(foreground):
"""启动 Daemon(前台),建议用 hermes start 后台运行。"""
if foreground:
from daemon.daemon import main as daemon_main
click.echo("Hermes Daemon 前台运行中 (Ctrl+C 停止)")
daemon_main()
else:
click.echo("请用 hermes start 后台启动,或 hermes daemon --foreground 前台运行")
if __name__ == "__main__":
cli()
|