nemotron-environment-setup
Set up or audit the dfw NeMo-RL/Nemotron diffusion training environment, including the Nemo-RL fork/worktree, Git submodules for Megatron-Bridge and…
Skill by sajadn
Chalk Labs
Need help deploying nemotron-environment-setup?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/sajadn-nemotron-environment-setup
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: security
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/nemotron-environment-setup
What's inside
- Skill name
nemotron-environment-setup- Description
- Set up or audit the dfw NeMo-RL/Nemotron diffusion training environment, including the Nemo-RL fork/worktree, Git submodules for Megatron-Bridge and Megatron-LM, local editable SGLang fork, UV driver and worker environments, and sanity checks before training or evaluation.
- License
- Apache-2.0
- Instructions
- 179 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Nemotron Environment Setup
Use this skill when setting up a fresh dfw environment, auditing dependency paths, or debugging why a run is using the wrong NeMo-RL, Megatron-Bridge, Megatron-LM, or SGLang code.
## New User Variables
The concrete paths in this skill are parameterized through the variables below. Use this mapping for a fresh setup:
```bash
export USER_NAME=<your_dfw_user>
export GITHUB_USER=<your_github_user>
export HOME_DIR=/home/${USER_NAME}
export REPO_DIR=${HOME_DIR}/diffusion_RL/RL
export LUSTRE_USER_ROOT=/lustre/fsw/portfolios/coreai/users/${USER_NAME}
export DRIVER_ENV_ROOT=${LUSTRE_USER_ROOT}/nemorl_uv_driver_envs
export WORKER_ENV_ROOT=${LUSTRE_USER_ROOT}/nemo_rl_worker_venvs
export UV_CACHE_ROOT=${LUSTRE_USER_ROOT}/uv_cache
```
Then create the base directories:
```bash
mkdir -p "${HOME_DIR}/diffusion_RL" "${LUSTRE_USER_ROOT}"
```
If your workspace uses a different Lustre portfolio or project path, set `LUSTRE_USER_ROOT` to that location and keep the driver envs, worker envs, and UV cache underneath it.
## Canonical Working Tree
For the current reference deployment, the main repo is:
```bash
cd "${REPO_DIR}"
```
For a new user, clone the NeMo-RL fork used by this workflow into `REPO_DIR` and check out `dllm_clean`:
```bash
git clone --branch dllm_clean git@github.com:sajadn/RL.git "${REPO_DIR}"
cd "${REPO_DIR}"
```
Check the current branch and commit before changing dependencies:
```bash
git branch --show-current
git rev-parse HEAD
git status --short
```
## Forks Submodules (Megatron-Bridge and Megatron-LM)
The repo uses local editable proxy packages in `pyproject.toml`, backed by Git submodules:
```toml
megatron-bridge = { path = "3rdparty/Megatron-Bridge-workspace", editable = true }
megatron-core = { workspace = true }
```
The source repos live one level deeper:
```text
3rdparty/Megatron-Bridge-workspace/Megatron-Bridge
3rdparty/Megatron-LM-workspace/Megatron-LM
```
Initialize or repair submodules with:
```bash
git submodule sync --recursive
git submodule update --init --recursive
```
Expected `.gitmodules` intent:
```text
Megatron-Bridge: git@github.com:sajadn/Megatron-Bridge.git, branch nemotron-diffusion
Megatron-LM: https://github.com/NVIDIA/Megatron-LM.git
```
Verify the actual checked-out sources:
```bash
git submodule status --recursive
git -C 3rdparty/Megatron-Bridge-workspace/Megatron-Bridge remote -v
git -C 3rdparty/Megatron-Bridge-workspace/Megatron-Bridge branch --show-current
git -C 3rdparty/Megatron-Bridge-workspace/Megatron-Bridge rev-parse HEAD
git -C 3rdparty/Megatron-LM-workspace/Megatron-LM remote -v
git -C 3rdparty/Megatron-LM-workspace/Megatron-LM rev-parse --abbrev-ref HEAD
git -C 3rdparty/Megatron-LM-workspace/Megatron-LM rev-parse HEAD
```
If committing submodule changes in NeMo-RL, first commit/push the nested repo change in the submodule repo, then commit the updated gitlink in the NeMo-RL repo. Git records only the submodule commit hash, not the whole nested repository contents.
## SGLang Fork
Current setup uses a local editable SGLang checkout in `pyproject.toml`:
```toml
sglang = { path = "${HOME_DIR}/diffusion_RL/sglang-nemotron-dllm-a652eb48/python", editable = true }
```
That local checkout should track the fork:
```text
git@github.com:sajadn/slgang-nemotron-diffusion.git
```
Set up or refresh it with:
```bash
git clone git@github.com:sajadn/slgang-nemotron-diffusion.git \
"${HOME_DIR}/diffusion_RL/sglang-nemotron-dllm-a652eb48"
cd "${HOME_DIR}/diffusion_RL/sglang-nemotron-dllm-a652eb48"
git checkout restore/a652eb48-good-runFrom sajadn/diffusion_RL, Apache-2.0.
More from sajadn/diffusion_RL
- 0.517score
benchmark-eval
sajadn
Run, monitor, or troubleshoot IFBench, LiveCodeBench, SciCode, and AA-LCR checkpoint evaluations on DFW using submit_benchmark_eval.sh.
SkillBFreshness: BClaude Code +9 - 0.504score
nemotron-experiment-submit
sajadn
Submit and monitor NeMo-RL NemotronLabsDiffusion GRPO-family experiments on dfw, including AR GRPO, JustGRPO, DiffuGRPO, and NeMo-Gym/NeMo-Skills tool…
SkillBFreshness: BClaude Code +9 - 0.468score
gsm8k-checkpoint-eval
sajadn
Evaluate native Nemotron Diffusion NeMo-RL Megatron checkpoints on math benchmarks such as GSM8K and AIME by converting them to Hugging Face format, then…
SkillCFreshness: CClaude Code +9 - 0.406score
contributing
sajadn
Contribution conventions for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.406score
copyright
sajadn
NVIDIA copyright header requirements for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.406score
error-handling
sajadn
Error handling guidelines for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.406score
code-style
sajadn
Code style guidelines for NeMo-RL (Python and shell).
SkillDFreshness: DClaude Code +9 - 0.406score
config-conventions
sajadn
Configuration conventions for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.402score
docs
sajadn
Documentation conventions for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.251score
dfw-cluster
sajadn
No description from the publisher yet.
SkillCFreshness: CClaude Code +9 - 0.226score
testing
sajadn
Testing conventions for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.211score
review-pr
sajadn
No description from the publisher yet.
SkillDFreshness: DClaude Code +9
Signals
- Stars
- 0
- Forks
- 1
- Open issues
- 0
- Last commit (this folder)
- Jun 7, 2026 (4 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/sajadn/diffusion_RL
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from sajadn
- 0.517score
benchmark-eval
sajadn
Run, monitor, or troubleshoot IFBench, LiveCodeBench, SciCode, and AA-LCR checkpoint evaluations on DFW using submit_benchmark_eval.sh.
SkillBFreshness: BClaude Code +9 - 0.504score
nemotron-experiment-submit
sajadn
Submit and monitor NeMo-RL NemotronLabsDiffusion GRPO-family experiments on dfw, including AR GRPO, JustGRPO, DiffuGRPO, and NeMo-Gym/NeMo-Skills tool…
SkillBFreshness: BClaude Code +9 - 0.468score
gsm8k-checkpoint-eval
sajadn
Evaluate native Nemotron Diffusion NeMo-RL Megatron checkpoints on math benchmarks such as GSM8K and AIME by converting them to Hugging Face format, then…
SkillCFreshness: CClaude Code +9 - 0.406score
code-style
sajadn
Code style guidelines for NeMo-RL (Python and shell).
SkillDFreshness: DClaude Code +9 - 0.406score
contributing
sajadn
Contribution conventions for NeMo-RL.
SkillDFreshness: DClaude Code +9 - 0.406score
copyright
sajadn
NVIDIA copyright header requirements for NeMo-RL.
SkillDFreshness: DClaude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get nemotron-environment-setup running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
nemotron-environment-setup skill. Agent Mart. https://agentmart.dev/skills/sajadn-nemotron-environment-setup (last updated 2026-10-24)