Catalog
affaan-m/ito-training

affaan-m

ito-training

Run an ML training job on a completed Itô compute booking through the canonical Itô backend. Use after ito-compute has booked GPU nodes and the user wants pre-training, fine-tuning, or RL on that metal. Chains off a booking record; ECC implements no training stack of its own.

v1.0Latest
New~597Updated Aug 7, 2026

Itô Training

Run training work on rented Itô metal by delegating to the canonical Itô compute backend (Layer 0.3). ECC does not implement a parallel training stack, trainer, or scheduler, and does no browser automation. This skill chains off a completed booking from ito-compute; it never books, reserves, or spends.

Prerequisite

A completed booking from the ito-compute skill (booking id, node IPs, SSH, GPU SKU, node count, fabric) in harness memory. Without one, stop.

Delegation

ECC calls the canonical backend through the ecc ito bridge; it never re-implements training. Authenticate once with ecc ito login, as ito-compute documents. Never put a key or token in arguments, files, logs, or chat.

ecc ito train \
  --booking <booking-id> \
  --model-size <e.g. 8B> \
  --data <data-ref> \
  --target <capability> \
  --budget-usd <ceiling> \
  [--post-training sft|dpo|rlvr]

What the backend does (Layer 0.3)

The desk backend runs a staged, eval-gated pipeline; this skill reports stage gates and never overrides one:

  1. Data prep — manifest, dedup, decontamination against the eval suite; 150M-ladder decision job as the cheap pre-check for custom data.
  2. Parallelism and precision — selected from model size, node count, fabric; wasteful combinations refused.
  3. Checkpointing and fault tolerance — async DCP, torchft; detect < 10 min, resume < 15 min. Loss-spike restart is a proposed, human-gated action.
  4. Curriculum and eval gates — staged pretrain / mid-train / long-context / post-training, each with a fixed eval battery; a failed gate stops the run.
  5. Post-training — SFT → DPO → RLVR (GRPO with DAPO stability fixes), trainer/rollout separation with bounded staleness.

Emits desk telemetry (goodput, interruption rate, checkpoint bandwidth) so the desk prices training blocks honestly.

Unavailable today

Not yet wired: the canonical CLI's run verb and the desk training-run backend are scaffolds. Until they land, this skill reports the missing capability and stops. Never substitute a local trainer or a purchase endpoint.

Files1
1 files · 1.0 KB

Select a file to preview

Overall Score

78/100

Grade

B

Good

Safety

85

Quality

78

Clarity

82

Completeness

68

Summary

A skill that delegates ML training job execution to the canonical Itô compute backend (Layer 0.3) after a completed GPU booking. It chains off an existing booking record from ito-compute and provides a wrapper around the `ecc ito train` CLI with strict guardrails: no booking, reservation, or spending operations; no credential embedding; and explicit refusal to override backend stage gates or substitute local trainers.

Detected Capabilities

Shell command execution (ecc ito train CLI)Authentication delegation (ecc ito login)Booking record reference (from harness memory)Backend telemetry reporting

Trigger Keywords

Phrases that MCP clients use to match this skill to user intent.

run ml training jobfine-tune on gpustart training workloadpost-training pipelinetrain on booked nodes

Risk Signals

INFO

Credential handling discipline: skill explicitly documents never embedding API keys or tokens in arguments, files, logs, or chat; delegates authentication to separate ecc ito login command

Delegation section
INFO

Incomplete feature: canonical CLI's run verb and training-run backend are scaffolds; skill reports missing capability and stops without substitution

Unavailable today section
INFO

Prerequisite enforcement: skill requires completed booking from ito-compute; stops if unavailable

Prerequisite section

Use Cases

  • Execute pre-training runs on rented GPU metal after booking nodes via ito-compute
  • Run fine-tuning or post-training (SFT, DPO, RLVR) workloads on provisioned Itô infrastructure
  • Delegate ML training pipeline execution while respecting eval gates and checkpoint fault-tolerance mechanisms
  • Monitor training telemetry (goodput, interruption rate, checkpoint bandwidth) from the canonical backend

Quality Notes

  • Strong security discipline: explicit guardrails against credential embedding, local trainer substitution, and backend gate override
  • Clear scope boundary: skill is a delegation wrapper, not a training implementation; explicitly documents what ECC does not do
  • Well-documented pipeline stages: 5-stage backend process clearly explained so agent understands when to expect gates/failures
  • Incomplete feature gracefully handled: skill documents that run verb is a scaffold and refuses to proceed, preventing silent failures
  • Architecture clarity: distinguishes ECC's role (CLI wrapper) from Layer 0.3 backend (actual training logic), preventing scope creep
Model: claude-haiku-4-5-20251001Analyzed: Aug 7, 2026

Reviews

Add this skill to your library to leave a review.

No reviews yet

Be the first to share your experience.

Use affaan-m/ito-training in your dev environment

Command Palette

Search for a command to run...