trl

trl is a skill for Claude Code, Codex from benchflow-ai/skillsbench. It costs 100 tokens per session (966 once invoked), scanned A, original, Apache-2.0.

A reference guide to TRL, a Python library for training language models with methods such as supervised fine-tuning and preference optimization. It explains how the library's trainers are organized and what their shared contracts are.

In plain words
What is it for?
Use it when working on TRL trainers, training configuration, model utilities, or the relationships between SFT, DPO, GRPO, KTO, and related methods.
Why use it?
It gives you the intended behavior of the codebase before you change or debug trainer implementations.

Skill for Claude CodeCodex

About the project

SkillsBench is a benchmark for measuring how effectively AI agents use modular skills—folders containing instructions, scripts, and resources—to complete specialized tasks. It helps researchers and developers evaluate both skill quality and agent behavior, including tasks that require combining multiple skills. The catalogue’s skills and instructions are evaluated as part of this workflow.

benchflow-ai/skillsbench · 1,747 stars · on GitHub

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/benchflow-ai/skillsbench/trl
Any agent
npx skills add benchflow-ai/skillsbench --skill trl
Clone the repo
git clone --depth 1 https://github.com/benchflow-ai/skillsbench

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for trl

README.md
[![agentmods](https://agentmods.dev/badge/skills/benchflow-ai/skillsbench/trl.svg)](https://agentmods.dev/skills/benchflow-ai/skillsbench/trl)
Your own site
<a href="https://agentmods.dev/skills/benchflow-ai/skillsbench/trl"><img src="https://agentmods.dev/badge/skills/benchflow-ai/skillsbench/trl.svg" alt="Measured on agentmods" height="20"></a>
Per session 100 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 966 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00100 $0.00966
Opus 5 $0.00050 $0.00483
Sonnet 5 $0.00020 $0.00193
Haiku 4.5 $0.00010 $0.00097

Measured yesterday against content hash 96e11df1379f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

trl scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

tasks/debug-trl-grpo/environment/skills/trl/SKILL.md · 85 lines

How it starts

The opening of the file, as written. The whole thing — 85 lines — stays where its author put it; the contents beside it link to each section on GitHub.

TRL Library Reference

Package Structure

TRL is organized around a trainer hierarchy that extends Hugging Face transformers.Trainer.

trl/
├── trainer/
│   ├── grpo_trainer.py       # GRPOTrainer
│   ├── grpo_config.py        # GRPOConfig
│   ├── sft_trainer.py        # SFTTrainer (supervised fine-tuning)
│   ├── dpo_trainer.py        # DPOTrainer (direct preference optimization)
│   ├── kto_trainer.py        # KTOTrainer (Kahneman-Tversky optimization)
│   ├── online_dpo_trainer.py # OnlineDPOTrainer
│   ├── utils.py              # Shared utilities (log probs, decoding, padding)
│   └── ...
├── models/
│   └── modeling_value_head.py  # Value head for PPO-style trainers
├── data_utils.py
├── commands/                   # CLI entry points
└── ...

Trainer Hierarchy

All TRL trainers extend transformers.Trainer:

transformers.Trainer
├── SFTTrainer          # Supervised fine-tuning
├── DPOTrainer          # Direct preference optimization
├── GRPOTrainer         # Group relative policy optimization
├── KTOTrainer          # Kahneman-Tversky optimization
└── OnlineDPOTrainer    # Online DPO

Each trainer overrides compute_loss with its specific objective, and RL-based trainers (GRPO, OnlineDPO) additionally override training_step to add a generation phase before the optimization step.

Shared Utility Functions (trainer/utils.py)

These utilities are used across multiple trainers. Read the source before modifying; the contracts below are what callers rely on.

selective_log_softmax(logits, index)

Memory-efficient per-token log-probability. Equivalent in value to F.log_softmax(logits, dim=-1).gather(...) at the selected token positions, but avoids materializing the full vocab-sized tensor.

Contract:

  • Input: logits [B, T, V], index [B, T]
  • Output: log_probs [B, T], each entry a valid log-probability (i.e. non-positive)
  • Must agree with F.log_softmax to within numerical tolerance on the same inputs

decode_and_strip_padding(input_ids, tokenizer)

Converts a batch of token ID tensors into the cleaned text strings that the reward function will score.

Read the full file on GitHub · 85 lines

Files

What ships with it

1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 85 lines · 100 tokens per session scan A 96e11df1379f

Subscribe to this mod's changes

trl is a skill published in the GitHub repository benchflow-ai/skillsbench (1,747 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 100 tokens to every session and 966 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.