How do I install the Train Sft skill?

Run npx skillmds@latest add mybigday/train-sft in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Train Sft safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Train Sft?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Train Sft free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Train Sft?

mybigday (@mybigday) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Train Sft by mybigday — Agent Skill

Name: Train Sft
Author: mybigday

Train Sft

Supervised fine-tune a causal LM with TRL `SFTTrainer`. Triggered when the user wants to fine-tune / SFT / instruct-tune / chat-tune a model on conversational, prompt-completion, or text-formatted data. Enforces the literature-first → audit → smoke-test → scale workflow.

mybigday bc1721c 2.6 KB Updated Sep 17, 2026

File contents

train-sft — SFT recipe

Use the working template at scripts/train_sft.py. Do not write the trainer from scratch — TRL renames things every release.

Sequence (no shortcuts)

Research — fire research-recipe if you don't already have a paper- backed recipe. Capture: dataset, method, hyperparameters, reference URL.
Audit — python scripts/inspect_dataset.py <dataset>. Confirm:
- Columns match SFT format (messages, text, or prompt+completion).
- p95 sequence length informs max_seq_length.
Config — copy configs/sft_default.yaml to configs/<run-name>.yaml and override model.name_or_path, dataset.name, hyperparameters, and (when ready to ship) train.push_to_hub: true + train.hub_model_id.
Smoke test locally:
```
python scripts/train_sft.py --config configs/<run-name>.yaml \
    --max-steps 20 --max-samples 256
```
Loss must come down. If push_to_hub is on, the smoke test creates the repo — that's fine.

Pre-flight block — print this to the user before scaling:

- Reference implementation: <paper / GH URL>
- Dataset format verified : <yes — columns: …>
- Model verified          : <repo_id, arch, tokenizer>
- push_to_hub             : True, hub_model_id=<…>
- Timeout                 : <≥2h>
- Monitoring              : Trackio dashboard URL

Scale — either:
- Local: accelerate launch scripts/train_sft.py --config configs/<run>.yaml
- HF Jobs: python scripts/launch_hf_job.py --script scripts/train_sft.py --config configs/<run>.yaml --hardware <flavor> --timeout <≥2h> --hub-model-id <user/repo>

Logging discipline (don't change)

In every TrainingArguments / SFTConfig:

disable_tqdm=True
logging_strategy="steps"
logging_first_step=True
logging_steps=10
report_to=["trackio"]   # or ["wandb"] if WANDB_API_KEY is set

Loss prints as plain text lines you can grep.

OOM

Apply in this order, never changing the user's request:

Halve per_device_train_batch_size, double gradient_accumulation_steps.
gradient_checkpointing=True.
optim="adamw_8bit" (bitsandbytes).
bf16=True (Ampere+).
Larger GPU tier.

Do not silently switch full SFT to LoRA, drop max_seq_length, or remove monitoring. Those change what the user gets.

---name: train-sftdescription: Supervised fine-tune a causal LM with TRL `SFTTrainer`. Triggered when the user wants to fine-tune / SFT / instruct-tune / chat-tune a model on conversational, prompt-completion, or text-formatted data. Enforces the literature-first → audit → smoke-test → scale workflow.---# train-sft — SFT recipeUse the working template at `scripts/train_sft.py`. Do **not** write thetrainer from scratch — TRL renames things every release.## Sequence (no shortcuts)1. **Research** — fire `research-recipe` if you don't already have a paper- backed recipe. Capture: dataset, method, hyperparameters, reference URL.2. **Audit** — `python scripts/inspect_dataset.py <dataset>`. Confirm: - Columns match SFT format (`messages`, `text`, or `prompt`+`completion`). - p95 sequence length informs `max_seq_length`.3. **Config** — copy `configs/sft_default.yaml` to `configs/<run-name>.yaml` and override `model.name_or_path`, `dataset.name`, hyperparameters, and (when ready to ship) `train.push_to_hub: true` + `train.hub_model_id`.4. **Smoke test** locally: ```bash python scripts/train_sft.py --config configs/<run-name>.yaml \ --max-steps 20 --max-samples 256 ``` Loss must come down. If `push_to_hub` is on, the smoke test creates the repo — that's fine.5. **Pre-flight block** — print this to the user before scaling: ``` - Reference implementation: <paper / GH URL> - Dataset format verified : <yes — columns: …> - Model verified : <repo_id, arch, tokenizer> - push_to_hub : True, hub_model_id=<…> - Timeout : <≥2h> - Monitoring : Trackio dashboard URL ```6. **Scale** — either: - Local: `accelerate launch scripts/train_sft.py --config configs/<run>.yaml` - HF Jobs: `python scripts/launch_hf_job.py --script scripts/train_sft.py --config configs/<run>.yaml --hardware <flavor> --timeout <≥2h> --hub-model-id <user/repo>`## Logging discipline (don't change)In every TrainingArguments / SFTConfig:```pythondisable_tqdm=Truelogging_strategy="steps"logging_first_step=Truelogging_steps=10report_to=["trackio"] # or ["wandb"] if WANDB_API_KEY is set```Loss prints as plain text lines you can `grep`.## OOMApply in this order, **never** changing the user's request:1. Halve `per_device_train_batch_size`, double `gradient_accumulation_steps`.2. `gradient_checkpointing=True`.3. `optim="adamw_8bit"` (bitsandbytes).4. `bf16=True` (Ampere+).5. Larger GPU tier.Do **not** silently switch full SFT to LoRA, drop `max_seq_length`, or removemonitoring. Those change what the user gets.

Frequently asked questions