Run any Skill in Manus with one click

train-sft

Stars0

Forks0

UpdatedApril 28, 2026 at 08:03

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.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

mybigday

mybigday/ml-intern-kit

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	train-sft
description	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 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.

More from this repository

same repository

rocm-strix-halo

mybigday/ml-intern-kit

Set up training and inference on AMD Ryzen AI Max+ 395 / Strix Halo (gfx1151, RDNA 3.5) with TheRock nightly ROCm wheels. Triggered when the host has gfx1151, when `rocminfo` shows Strix Halo, or when the user mentions Strix Halo / Ryzen AI Max / gfx1151 / 128GB unified memory.

2026-04-280

env-bootstrap

mybigday/ml-intern-kit

Recreate the ml-intern-kit Python environment on a new machine (laptop, rented GPU box, Docker, fresh checkout). Triggered when the user is on a new host, sees ImportError on a core dep (torch/transformers/trl/peft/accelerate), or wants to install flash-attn / unsloth / bitsandbytes after the fact.

2026-04-280

eval-model

mybigday/ml-intern-kit

Evaluate a trained or downloaded language model with `lm-eval-harness` standard tasks (arc, hellaswag, gsm8k, mmlu, truthfulqa, ifeval, ...). Triggered when the user wants to benchmark, eval, or compare a model — pre- or post-training.

2026-04-280

inspect-dataset

mybigday/ml-intern-kit

Audit a Hugging Face dataset before training to confirm splits, columns, format, sample rows, distributions, and duplicates. Triggered before any training/fine-tuning script runs, when a user mentions a new dataset, or when you hit a KeyError / format mismatch in a training job.

2026-04-280

launch-hf-job

mybigday/ml-intern-kit

Submit a training/inference script to Hugging Face Jobs (`hf jobs run`). Triggered when the user wants to run training in the cloud, scale beyond local hardware, or kick off a multi-hour fine-tune. Enforces pre-flight: hub_model_id required, ≥2h timeout for training, single-job validation before batch.

2026-04-280

research-recipe

mybigday/ml-intern-kit

Run a literature-first crawl before writing ANY ML training/fine-tuning/inference code. Spawns an Explore sub-agent that mines papers, citation graphs, methodology sections, and matched HF datasets to produce a ranked list of training recipes attributed to specific published results. Triggered when the user asks to fine-tune, train, or improve a model, or when the user names a task/benchmark and you need a recipe before coding.

2026-04-280

name	train-sft
description	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 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.