Spaces:
Running
Running
Shoufa Zhang
Deploy DocMorph & SeekMorph interactive Studio and 5-layer production benchmark
66227f0 Download seekmorph/cli.py from zhangshoufa/DocMorph: direct link, hf CLI and curl.
- Browser
- Download file 9.02 kB
-
https://huggingface.co/spaces/zhangshoufa/DocMorph/resolve/main/seekmorph/cli.py
- Command line
-
hf download hf://spaces/zhangshoufa/DocMorph/seekmorph/cli.py
-
curl -L -o cli.py https://huggingface.co/spaces/zhangshoufa/DocMorph/resolve/main/seekmorph/cli.py
9.02 kB
| """Command-line interface for SeekMorph.""" | |
| from __future__ import annotations | |
| import argparse | |
| import json | |
| import sys | |
| from seekmorph.agent import PersonalMemoryStore, ProactiveTicketRuntime, TriggerEvent | |
| from seekmorph.engine import KnowledgeItem, SeekMorphEngine, UserProfile | |
| from seekmorph.metrics import evaluate_benchmark | |
| def build_parser() -> argparse.ArgumentParser: | |
| parser = argparse.ArgumentParser( | |
| prog="seekmorph", | |
| description="Hybrid AI search, personalized recommendation reranking, and ACL-scoped knowledge base engine.", | |
| ) | |
| sub = parser.add_subparsers(dest="command", required=True) | |
| p_search = sub.add_parser("search", help="Search and personalize results from a JSONL knowledge base.") | |
| p_search.add_argument("index_path", help="Path to JSONL knowledge base index.") | |
| p_search.add_argument("query", help="Search query.") | |
| p_search.add_argument("--tenant", default="default", help="Tenant ID (default: default).") | |
| p_search.add_argument("--role", action="append", default=[], help="User role(s) for ACL filtering.") | |
| p_search.add_argument("--interest", action="append", default=[], help="User interest tag(s) (e.g. infra:1.0).") | |
| p_search.add_argument("-k", "--top-k", type=int, default=5, help="Top-K results (default: 5).") | |
| p_search.add_argument("--json", action="store_true", help="Output JSON.") | |
| sub.add_parser("demo", help="Run an interactive showcase of personalized search & ACL isolation.") | |
| sub.add_parser("eval", help="Run offline IR & RecSys benchmark evaluation (Recall@K, MRR@K, NDCG@K).") | |
| return parser | |
| def _run_demo() -> int: | |
| engine = SeekMorphEngine() | |
| engine.add_many([ | |
| KnowledgeItem( | |
| item_id="doc_infra_1", | |
| title="Production Deployment Guide (Kubernetes & vLLM)", | |
| content="Deploy DeepSeek-V4 with DSpark speculative decoding on 8x MI308X GPU clusters.", | |
| tags=("infra", "vllm", "deployment"), | |
| tenant_id="acme", | |
| allowed_roles=("sre", "backend"), | |
| age_days=5.0, | |
| impressions=100, | |
| clicks=32, | |
| ), | |
| KnowledgeItem( | |
| item_id="doc_fe_1", | |
| title="Frontend Deployment Guide (CDN & Static Assets)", | |
| content="Deploy the web console static bundle to edge nodes.", | |
| tags=("frontend", "cdn", "deployment"), | |
| tenant_id="acme", | |
| allowed_roles=("frontend", "sre"), | |
| age_days=12.0, | |
| impressions=100, | |
| clicks=18, | |
| ), | |
| KnowledgeItem( | |
| item_id="doc_exec_secret", | |
| title="M&A Executive Compensation Table", | |
| content="Restricted board-only deployment budget and executive packages.", | |
| tags=("finance", "deployment"), | |
| tenant_id="acme", | |
| allowed_roles=("board",), | |
| age_days=1.0, | |
| impressions=20, | |
| clicks=15, | |
| ), | |
| ]) | |
| sre_user = UserProfile( | |
| user_id="u_sre", | |
| tenant_id="acme", | |
| roles=("sre",), | |
| interest_tags={"infra": 1.2, "vllm": 1.0}, | |
| ) | |
| fe_user = UserProfile( | |
| user_id="u_fe", | |
| tenant_id="acme", | |
| roles=("frontend",), | |
| interest_tags={"frontend": 1.2, "cdn": 1.0}, | |
| ) | |
| print("=== Query: 'deployment guide' (SRE Engineer Profile) ===") | |
| for i, r in enumerate(engine.search_and_recommend("deployment guide", user=sre_user, top_k=3), 1): | |
| print( | |
| f" [{i}] {r.item.item_id}: {r.item.title} " | |
| f"(final={r.final_score:.4f}, personal={r.personal_score:.3f}, intent={r.query_intent})" | |
| ) | |
| print("\n=== Query: 'deployment guide' (Frontend Engineer Profile) ===") | |
| for i, r in enumerate(engine.search_and_recommend("deployment guide", user=fe_user, top_k=3), 1): | |
| print( | |
| f" [{i}] {r.item.item_id}: {r.item.title} " | |
| f"(final={r.final_score:.4f}, personal={r.personal_score:.3f}, intent={r.query_intent})" | |
| ) | |
| memory = PersonalMemoryStore(engine=engine) | |
| memory.remember( | |
| memory_id="mem_sre_pref", | |
| user_id="u_sre", | |
| tenant_id="acme", | |
| title="SRE Alerting Preference", | |
| content="Send critical GPU node OOM alerts to PagerDuty and mute staging warnings.", | |
| kind="preference", | |
| tags=("infra", "alerting"), | |
| ) | |
| recalled = memory.recall(sre_user, "GPU node alerts", top_k=1, kind="preference") | |
| print(f"\n=== Personal Agent Memory Recall (u_sre): {recalled[0].item.title} ===") | |
| runtime = ProactiveTicketRuntime(allowed_actors={"u_sre"}, tenant_id="acme", eval_mode="off") | |
| ticket = runtime.receive( | |
| TriggerEvent( | |
| trigger_id="gh_101", | |
| source="github", | |
| thread_id="issue_42", | |
| actor_id="u_sre", | |
| text="@bot investigate GPU OOM on MI308X node", | |
| addressed=True, | |
| ) | |
| ) | |
| print(f"=== Proactive SuperAgent Ticket Gate: ticket={ticket.ticket_id if ticket else None} state={ticket.state if ticket else None} ===") | |
| return 0 | |
| def _run_eval() -> int: | |
| engine = SeekMorphEngine() | |
| engine.add_many([ | |
| KnowledgeItem( | |
| item_id="infra_vllm", | |
| title="Deployment Guide (vLLM & GPU Cluster)", | |
| content="Standard rollout checklist for DeepSeek-V4 on MI308X GPU nodes.", | |
| tags=("infra", "vllm"), | |
| tenant_id="acme", | |
| age_days=2.0, | |
| impressions=100, | |
| clicks=35, | |
| ), | |
| KnowledgeItem( | |
| item_id="legacy_infra", | |
| title="Deployment Guide (Legacy VM)", | |
| content="Standard rollout checklist for legacy virtual machines.", | |
| tags=("legacy",), | |
| tenant_id="acme", | |
| age_days=400.0, | |
| impressions=100, | |
| clicks=3, | |
| ), | |
| KnowledgeItem( | |
| item_id="fe_cdn", | |
| title="Deployment Guide (Frontend CDN)", | |
| content="Standard rollout checklist for static web bundles.", | |
| tags=("frontend", "cdn"), | |
| tenant_id="acme", | |
| age_days=5.0, | |
| impressions=100, | |
| clicks=28, | |
| ), | |
| ]) | |
| sre_user = UserProfile(user_id="sre", tenant_id="acme", interest_tags={"infra": 1.5, "vllm": 1.0}) | |
| fe_user = UserProfile(user_id="fe", tenant_id="acme", interest_tags={"frontend": 1.5, "cdn": 1.0}) | |
| sre_preds = [r.item.item_id for r in engine.search_and_recommend("rollout checklist", user=sre_user, top_k=3)] | |
| fe_preds = [r.item.item_id for r in engine.search_and_recommend("rollout checklist", user=fe_user, top_k=3)] | |
| report = evaluate_benchmark( | |
| predictions=[sre_preds, fe_preds], | |
| ground_truth_graded=[ | |
| {"infra_vllm": 3.0, "fe_cdn": 1.0, "legacy_infra": 0.0}, | |
| {"fe_cdn": 3.0, "infra_vllm": 1.0, "legacy_infra": 0.0}, | |
| ], | |
| k=2, | |
| ) | |
| print( | |
| f"Benchmark Evaluation (queries={report.num_queries}, K={report.k}): " | |
| f"Recall@{report.k}={report.mean_recall:.4f} | " | |
| f"MRR@{report.k}={report.mean_mrr:.4f} | " | |
| f"NDCG@{report.k}={report.mean_ndcg:.4f}" | |
| ) | |
| return 0 | |
| def main(argv: list[str] | None = None) -> int: | |
| args = build_parser().parse_args(argv) | |
| if args.command == "demo": | |
| return _run_demo() | |
| if args.command == "eval": | |
| return _run_eval() | |
| if args.command == "search": | |
| engine = SeekMorphEngine.load(args.index_path) | |
| interests: dict[str, float] = {} | |
| for spec in args.interest: | |
| if ":" in spec: | |
| k, v = spec.split(":", 1) | |
| interests[k.strip()] = float(v) | |
| else: | |
| interests[spec.strip()] = 1.0 | |
| user = UserProfile( | |
| user_id="cli_user", | |
| tenant_id=args.tenant, | |
| roles=tuple(args.role), | |
| interest_tags=interests, | |
| ) | |
| results = engine.search_and_recommend(args.query, user=user, top_k=args.top_k) | |
| if args.json: | |
| print( | |
| json.dumps( | |
| [ | |
| { | |
| "rank": i + 1, | |
| "item_id": r.item.item_id, | |
| "title": r.item.title, | |
| "final_score": r.final_score, | |
| "lexical_score": r.lexical_score, | |
| "personal_score": r.personal_score, | |
| "recency_score": r.recency_score, | |
| "intent": r.query_intent, | |
| } | |
| for i, r in enumerate(results) | |
| ], | |
| ensure_ascii=False, | |
| indent=2, | |
| ) | |
| ) | |
| else: | |
| for i, r in enumerate(results, 1): | |
| print(f"[{i}] {r.item.title} ({r.item.item_id}) | score={r.final_score:.4f} [intent={r.query_intent}]") | |
| print(f" {r.item.content}") | |
| return 0 | |
| return 1 | |