|
Download docs_draft/TEAM_COORDINATION.md from hjerpe/sql_env: direct link, hf CLI and curl.
- Browser
- Download file 4.9 kB
-
https://huggingface.co/spaces/hjerpe/sql_env/resolve/main/docs_draft/TEAM_COORDINATION.md
- Command line
-
hf download hf://spaces/hjerpe/sql_env/docs_draft/TEAM_COORDINATION.md
-
curl -L -o TEAM_COORDINATION.md https://huggingface.co/spaces/hjerpe/sql_env/resolve/main/docs_draft/TEAM_COORDINATION.md
4.9 kB
| # SQLEnv Team Coordination | |
| ## Team | |
| | Person | Role | Contact | | |
| |--------|------|---------| | |
| | You (Hjerp) | Coordinator | - | | |
| | Kevlar | Contributor | - | | |
| | Jindal | Contributor | - | | |
| --- | |
| ## Kickoff Meeting - Feb 9 | |
| ### 1. Personal Wins | |
| > "What would make this project a win for you personally?" | |
| | Person | Personal Win | | |
| |--------|--------------| | |
| | You | | | |
| | Kevlar | | | |
| | Jindal | | | |
| ### 2. Availability | |
| | Person | Hours/Week | Best Times | Known Conflicts | | |
| |--------|------------|------------|-----------------| | |
| | You | | | | | |
| | Kevlar | | | | | |
| | Jindal | | | | | |
| **Total available:** _____ hours/week | |
| **Estimated need:** ~40-60 hours total over 16 days (~15-20 hrs/week combined) | |
| **Gap?** If yes, scope discussion needed. | |
| ### 3. Ownership Areas | |
| From project brief, 5 natural ownership areas. Let people **claim** - don't assign. | |
| | Area | Owner | Notes | | |
| |------|-------|-------| | |
| | Environment Engineering | | OpenEnv integration, WebSocket, Docker, action handlers | | |
| | Reward Design | | 3-layer rewards, progress metrics, anti-gaming | | |
| | Dataset Curation | | Spider questions, answer verification, difficulty balance | | |
| | Training Pipeline | | GRPO setup, prompts, evaluation, Green Agent | | |
| | Storytelling/Blog | | Blog post, demos, results visualization | | |
| **Note:** Some areas can be shared or split. Training Pipeline depends on Environment + Reward being done first. | |
| --- | |
| ## First Deliverables (Full Tickets) | |
| Use **full format** for first deliverables (commitment test). | |
| ### Ticket 1 | |
| **WHO:** | |
| **WHAT:** | |
| **WHY:** | |
| **CHALLENGE:** | |
| **DUE:** | |
| **DELIVERED:** | |
| ### Ticket 2 | |
| **WHO:** | |
| **WHAT:** | |
| **WHY:** | |
| **CHALLENGE:** | |
| **DUE:** | |
| **DELIVERED:** | |
| ### Ticket 3 | |
| **WHO:** | |
| **WHAT:** | |
| **WHY:** | |
| **CHALLENGE:** | |
| **DUE:** | |
| **DELIVERED:** | |
| --- | |
| ## Suggested First Deliverables (If Needed) | |
| Based on Phase 1 requirements, here are ready-to-use tickets: | |
| ### Environment Scaffold | |
| **WHO:** [Claim during meeting] | |
| **WHAT:** Run `openenv init sql_env`, customize Pydantic models (SQLAction, SQLObservation), get `openenv validate` passing | |
| **WHY:** Unblocks all environment work; proves Docker/WebSocket setup works | |
| **CHALLENGE:** Docker configuration; WebSocket timeout settings; understanding OpenEnv API | |
| **DUE:** [Wednesday EOD?] | |
| **DELIVERED:** | |
| ### Initial Question Set | |
| **WHO:** [Claim during meeting] | |
| **WHAT:** Select 30 questions from Spider dev set (12 easy, 12 medium, 6 hard) with gold answers in JSON format | |
| **WHY:** Enables manual testing of environment in Phase 2; needed for reward computation | |
| **CHALLENGE:** Balancing difficulty; avoiding questions needing unsupported SQL features | |
| **DUE:** [Wednesday EOD?] | |
| **DELIVERED:** | |
| ### OpenEnv Tutorial Review | |
| **WHO:** [Claim during meeting] | |
| **WHAT:** Complete OpenEnv tutorial notebook, document key learnings and gotchas for team | |
| **WHY:** Reduces ramp-up time for others; surfaces unknowns early | |
| **CHALLENGE:** Tutorial may have gaps; need to map to our SQL use case | |
| **DUE:** [Tuesday EOD?] | |
| **DELIVERED:** | |
| --- | |
| ## Coordination Rituals | |
| ### Daily Async Standup | |
| Post in shared doc/channel (30 seconds): | |
| ``` | |
| [Date] [Name] | |
| Did: [What you accomplished] | |
| Blocked: [Nothing / specific blocker] | |
| ``` | |
| **Where:** [TBD - decide in kickoff] | |
| ### Weekly Sync | |
| **When:** [TBD - decide in kickoff] | |
| **Duration:** 30 min | |
| **Agenda:** | |
| 1. Blockers (10 min) - resolve or escalate | |
| 2. Decisions (10 min) - use decision format below | |
| 3. Next deliverables (10 min) - create tickets for next week | |
| --- | |
| ## Decisions Log | |
| ### Decision: [Topic] | |
| **Context:** [Why needed now] | |
| **Options:** | |
| 1. [Option A] | |
| - Pro: | |
| - Con: | |
| 2. [Option B] | |
| - Pro: | |
| - Con: | |
| **Recommendation:** | |
| **Decided:** [Date] - [Choice] - [Who consulted] | |
| --- | |
| ## Open Questions for Kickoff | |
| From project brief - need team input: | |
| 1. **Reward components**: Expose as separate rewards to TRL, or sum into single scalar? | |
| 2. **Question selection**: Hand-pick for diversity, or random sample by difficulty? | |
| 3. **HINT action**: Add a hint mechanism, or keep it pure exploration? | |
| --- | |
| ## Timeline Summary | |
| | Phase | Days | Key Milestone | Dependencies | | |
| |-------|------|---------------|--------------| | |
| | 1. Scaffold | 1-2 | `openenv validate` passes | None | | |
| | 2. Core Loop | 3-5 | Full episode works manually | Phase 1 | | |
| | 3. Dense Reward | 6-8 | Reward varies meaningfully | Phase 2 | | |
| | 4. Training | 9-13 | Trained model beats random | Phases 1-3 | | |
| | 5. Polish | 14-16 | All artifacts submitted | Phase 4 | | |
| **Submission deadline:** ~16 days from kickoff | |
| --- | |
| ## Communication Channels | |
| | Channel | Purpose | | |
| |---------|---------| | |
| | [TBD] | Daily standups | | |
| | [TBD] | Quick questions / blockers | | |
| | [TBD] | Code (GitHub repo) | | |
| | Google Drive | Shared docs (this doc, project brief) | | |
| --- | |
| ## Next Sync | |
| **When:** [Fill in after kickoff] | |
| **Where:** [Fill in after kickoff] | |
| --- | |
| ## Completed Tickets Archive | |
| Move completed tickets here with DELIVERED filled in. | |
| *(None yet)* | |