Repository navigation
contributing guide updates - #881
Conversation
Signed-off-by: Lawrence Lane <llane@nvidia.com>
Signed-off-by: Lawrence Lane <llane@nvidia.com>
jmabry
left a comment
There was a problem hiding this comment.
Documentation appears in triplicate. Possible to reference a single-source of truth so that is more DRY in any secondary docs?
|
can we update the DCO fix instructions regarding force push too? |
β¦pates Signed-off-by: Lawrence Lane <llane@nvidia.com>
- Single-source the 'Use of AI and LLM Tools' section in the Fern development-setup page; CONTRIBUTING.md and the contribute index now carry the guiding principle plus a link instead of duplicating it (DRY). - Recommend tracking worthwhile drive-by changes as their own issue. - Add a PR template with an AI-assistance disclosure section and point attribution guidance at it (commit trailer kept as an alternative). - Reframe 'What We Will Close' as 'When We Request Changes' (close only if unaddressed). - Document force-push (--force-with-lease) when fixing DCO sign-off on an already-pushed branch. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Lawrence Lane <llane@nvidia.com>
|
Thanks for the reviews β addressed all the feedback in Note on scope: this branch predated the Fern docs migration, so it was editing the old Changes:
|
|
πΏ Preview your docs: https://nvidia-preview-llane-contributing-udpates.docs.buildwithfern.com/nemo/gym Here are the markdown pages you've updated: |
Co-authored-by: JOSHUA MABRY <mabryjn@gmail.com> Signed-off-by: Seph Mard <seph.mard@gmail.com>
Remove PR-template AI disclosure and the attribution section, point agents at AGENTS.md, clarify main.yml nav edits, and separate NVIDIA Slack from external Fern org access requests. Signed-off-by: Seph Mard <smard@nvidia.com>
Signed-off-by: Seph Mard <smard@nvidia.com>
Make DCO -s required and cryptographic signing optional, point the quality bar at AGENTS.md, clarify Fern org access requests and Makefile login for externals, and allow N/A on the PR tests checklist. Signed-off-by: Seph Mard <smard@nvidia.com>
|
/claude review |
|
SHIP β LGTM, no reliability concerns. Docs/metadata-only change (7 files: PR template, The substantive change is a policy shift: cryptographic commit signing ( Note (author's call, non-blocking): this relaxes an existing convention β CLAUDE.md still states "All commits require DCO sign-off ( |
jmabry
left a comment
There was a problem hiding this comment.
AGENTS.md and Claude.md will be out of sync now. Need something akin to
my-ai-library/
βββ AGENTS.md # Unified root instructions (Codex, Claude, Copilot)
βββ CLAUDE.md # Claude-specific rules or symlink to AGENTS.md
...
Merge former CLAUDE.md content into AGENTS.md (DCO -s only; -S optional) and turn CLAUDE.md into a symlink so Claude Code and other tools stay in sync. Signed-off-by: Seph Mard <smard@nvidia.com>
|
@jmabry Addressed the AGENTS.md / CLAUDE.md sync concern:
|
|
|
||
| 1. **Dataset** β JSONL with one task per row. NeMo Gym uses the OpenAI Responses API as its native format because it natively represents multi-turn, tool-calling agentic trajectories without custom serialization. Each row has `responses_create_params.input` (the input messages in Responses API format) and `verifier_metadata` (task-specific data passed to the verifier) | ||
| 2. **Resources Server** β implements verification logic, environment-specific tools, and per-task state isolation | ||
| 3. **Agent Harness** β reuse a built-in agent harness (e.g. OpenHands) or bring your own |
There was a problem hiding this comment.
feels we should mention multi agent somewhere - eg user simulation etc - but fine for another time
| pre-commit install | ||
|
|
||
| # Run servers | ||
| gym env start \ |
There was a problem hiding this comment.
should prob include rollouts, and setting env.yaml
Replacement PR for: #662
Contributors: @lbliii
first pass at guidelines for AI-enabled contributions. Open to all ideas/suggestions.