[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-155":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":14,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":16,"rankGlobal":8,"rankLanguage":8,"license":8,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":19,"hasPages":17,"topics":20,"createdAt":8,"pushedAt":8,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":14,"starSnapshotCount":14,"syncStatus":24,"lastSyncTime":25,"discoverSource":26},155,"deepswarm","amanning3390\u002Fdeepswarm","amanning3390",null,"Python",122,4,108,1,0,10,37.1,false,"main",true,[],"2026-06-11 04:00:17","# DeepSwarm 2.0 — Task-Agnostic Parallel Worker Orchestration\n\nSpawn N parallel API workers for **any** batch task. Auto-optimizes worker count + stagger. Tiered model delegation: orchestrator plans (V4 Pro) → workers execute (V4 Flash). 99.95% API success rate.\n\n## Install\n\n```bash\nhermes skills tap add amanning3390\u002Fdeepswarm\n```\n\n## Quick Start\n\n```bash\n# 1. Define your task\ncp task.yaml my_task.yaml\n# Edit: prompt_template, worker_model, max_tokens\n\n# 2. Generate seeds\npython3 scripts\u002Fseed.py --count 1000 --template \"Generate {{seed}}\" > seeds.jsonl\n\n# 3. Launch (auto-optimizes everything)\nexport DEEPSEEK_API_KEY=sk-...\npython3 scripts\u002Fswarm.py --task my_task.yaml --total 1000\n```\n\n## Tiered Delegation\n\n```yaml\norchestrator_model: deepseek-v4-pro   # Plans (few calls, frontier quality)\nworker_model: deepseek-v4-flash       # Executes (many calls, cheaper)\n```\n\nV4 Pro costs ~3× V4 Flash per token. For batch tasks with thousands of calls, tiered delegation saves 60-70%.\n\n## Auto-Optimization\n\n| Call Duration | Workers | Stagger | Success | Throughput |\n|--------------|---------|---------|---------|------------|\n| \u003C10s | 16 | 1s | 99.9% | ~5,760\u002Fhr |\n| 10-30s | 12 | 2s | 99.9% | ~1,440\u002Fhr |\n| 30-60s | 8 | 5s | 99.95% | ~440\u002Fhr |\n| 60-90s | 6 | 10s | 99.9% | ~240\u002Fhr |\n\nOmit `workers` and `stagger` in task.yaml — DeepSwarm runs a calibration call and picks optimal values.\n\n## Task Types\n\nBuilt-in: `generation`, `translation`, `summarization`, `classification`, `custom`\n\nFor multi-turn tasks (tool calling, conversation loops):\n```yaml\nmulti_turn: true\nmax_turns: 20\n```\n\n## Files\n\n```\ndeepswarm\u002F\n├── SKILL.md              # Hermes skill definition\n├── README.md\n├── task.yaml             # Sample task config\n├── architecture.html     # Pipeline diagram\n├── scripts\u002F\n│   ├── seed.py           # Seed generator (simple + rich templates)\n│   ├── swarm.py          # Orchestrator (auto-optimize + launch)\n│   ├── worker.py         # Task-agnostic batch processor\n│   └── filter.py         # Quality filter with JSON repair\n├── templates\u002F\n│   └── prompts.py        # v2 prompt templates\n└── references\u002F\n    ├── api-rate-limits.md\n    └── generation-patterns.md\n```\n\n## Post-Processing\n\nAfter generation, filter the raw output to remove malformed traces:\n\n```bash\npython3 scripts\u002Ffilter.py --input-dir output\u002F --output clean.jsonl --errors errors.jsonl\n```\n\nThe filter applies 3 passes: JSON repair (fixes 17% API error rate), structural validation\n(think blocks, tag balance), and length thresholds. Typically lifts pass rate from ~28% to ~62%.\n\n## Provenance\n\nBuilt from the [DeepSeek Hermes Reasoning Traces](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fning423\u002Fdeepseek-hermes-reasoning-traces) project:\n- 19,331 traces · 192K tool calls\n- 96 workers · 31K API calls\n- 99.95% success rate\n- 8 workers + 5s stagger = the magic formula\n","DeepSwarm 2.0 是一个任务无关的并行工作调度系统，能够为任何批量任务启动多个并行API工作线程。其核心功能包括自动优化工作线程数量和延迟时间、分层模型委派（使用V4 Pro作为规划者，V4 Flash执行任务），以及高达99.95%的API成功率。通过这种方式，对于需要数千次调用的批量任务，成本可以节省60-70%。该工具特别适用于需要大量API调用的数据处理场景，如生成、翻译、摘要、分类等任务，并且支持多轮对话或工具调用。用户只需定义任务配置文件，系统会自动进行校准并选择最优参数运行。此外，还提供了后处理脚本来修复生成数据中的错误，进一步提高输出质量。",2,"2026-06-11 02:31:08","CREATED_QUERY"]