datagen-standard-launch

name

datagen-standard-launch

description

Launch NON-AGENTIC (standard) data generation — plain vLLM/API completion generation with NO Harbor agent loop or Daytona sandboxes. Two paths: (1) Curator sharded datagen — the multi-node data-parallel run_curator_datagen_sharded.sbatch (one vLLM server per node, disjoint dataset slices, auto-resume, manual afterany restart chain, `curator` conda env, --account=reformo); (2) the declarative `generate.py` / class-based `generate_abstract.py` generators under data/ (data/generation BaseDataGenerator + InferenceEngine for OpenAI/Anthropic/vLLM). Use for bulk completion/synthetic-data generation. The AGENTIC trace-generation path (Harbor + Daytona rollouts, MiniMax/GLM trace sets) is the SEPARATE `datagen-launch` skill.

datagen-standard-launch

Generate completions or synthetic data directly from a model, without Harbor or Daytona. For agentic trace-gen, use datagen-launch.

---

Path 1 — Curator sharded datagen (multi-node DP) — the primary path

data/sbatches/run_curator_datagen_sharded.sbatch runs one vLLM server per node over disjoint input slices and merges the results. Default: 32 nodes (#SBATCH --nodes=32, 4 GPUs/node), --account=reformo on Jupiter, and the curator environment (not otagent).

Auto-resume: stable shard output paths data/sbatches/curator_runs/<model>__<dataset>__<N>shards/shard_<i>/checkpoint_*.parquet let an afterany restart resume after a SLURM timeout.

Pre-req (login node, has internet): cache the input dataset first —

conda activate curator
python -c "from datasets import load_dataset; ds=load_dataset('<dataset>',split='train'); print(len(ds))"

Launch — positional args <model> <input_dataset> <output_repo> [limit] [save_every]:

# Simple (no restarts):
sbatch data/sbatches/run_curator_datagen_sharded.sbatch <model> <input_dataset> <output_repo> [limit] [save_every]

# With restart chain (recommended for long datasets) — build the afterany chain MANUALLY:
FIRST=$(sbatch data/sbatches/run_curator_datagen_sharded.sbatch \
  <model> <input_dataset> <output_repo> [limit] [save_every] | awk '{print $4}')
PREV=$FIRST; for i in $(seq 1 6); do
  PREV=$(sbatch --dependency=afterany:$PREV \
    data/sbatches/run_curator_datagen_sharded.sbatch \
    <model> <input_dataset> <output_repo> [limit] [save_every] | awk '{print $4}')
done

• MAX_RESTARTS is not implemented; build the --dependency=afterany: chain by hand.
• save_every: pass 700, not the default 200 (the fifth positional arg).
• limit (4th arg) caps rows for a smoke run; omit for the full set.

---

Path 2 — declarative / class-based generator scripts (data/)

data/ has named pipeline directories in two styles:

• Declarative generate.py — self-contained scripts for local / one-off runs:

  python data/<dataset>/generate.py [--flags]
  

• Class-based generate_abstract.py — subclass BaseDataGenerator for HPC runs with launcher-managed vLLM endpoints, submitted through the unified launcher:

  python -m hpc.launch --job_type datagen \
    --datagen_script data/<dataset>/generate_abstract.py \
    --datagen_target_repo <org/repo> \
    --datagen_extra_args "--stage both --limit 2000"
  

Core modules in data/generation/: base.py (BaseDataGenerator), schemas.py (GenerationRequest/GenerationResult), engines.py (InferenceEngine for OpenAI / Anthropic / vLLM).

---

Cleanup / verification

Standard datagen pushes merged parquet shards to <output_repo>. Verify the row count and that the repo self-populated; no Daytona/trace-export step applies. Cluster details: .agents/ops/jupiter/; launcher: .agents/projects/ot-agent/ot-agent.md.