bedrock-fine-tuning
Amazon Bedrock Model Customization with fine-tuning, continued pre-training, reinforcement fine-tuning (NEW 2025 - 66% accuracy gains), and distillation. Create customization jobs, monitor training, deploy custom models, and evaluate performance. Use when customizing Claude, Titan, or other Bedrock models for domain-specific tasks, adapting to proprietary data, improving accuracy on specialized workflows, or distilling large models to smaller ones.
npx skills add majiayu000/claude-skill-registry --skill bedrock-fine-tuning --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
Guides Amazon Bedrock Model Customization across four approaches: fine-tuning (supervised), continued pre-training (domain adaptation), reinforcement fine-tuning (RLHF/RLAIF), and distillation (teacher-student). Provides model support matrix, data format requirements, and step-by-step procedures to create and monitor customization jobs, upload data to S3, deploy, and run tests with a custom model. Includes quick-start tasks for preparing data and invoking models.
How it works
Describes distinct customization methods with their use cases and typical gains:
- Fine-Tuning: train with labeled JSONL examples to optimize for specific tasks and formats. Requires JSONL with prompt/completion and supports 32K token max per example.
- Continued Pre-Training: unlabeled domain text to build domain knowledge. JSONL unlabeled text; typical gains 15-30% domain accuracy.
- Reinforcement Fine-Tuning: RLHF/RLAIF using prompts and preferred outputs, with ranking option. Requires prompt, chosen, rejected responses (JSONL); typical gains 40-66% accuracy.
- Distillation: transfer from larger to smaller models; no training data needed for configuration. Includes teacher and student model IDs. The guide provides example code blocks for creating customization jobs in Python with AWS Bedrock SDK, including function definitions for create_fine_tuning_job, create_continued_pretraining_job, create_reinforcement_finetuning_job, and create_distillation_job, along with sample usage and S3 data upload steps.
When to use it
Use when you need domain-specific adaptation of Claude, Titan, or other Bedrock models, leveraging proprietary data to improve accuracy on specialized workflows, or when aiming to distill large models to smaller, cheaper ones.
What it can touch
All operations rely on AWS Bedrock and S3 interactions. The procedures reference:
- AWS Bedrock API calls like create_model_customization_job and get_model_customization_job
- S3 URIs for training, validation, and outputs
- boto3 Python client interactions: bedrock, bedrock-runtime
- Example ARNs for roles and base models
Caveats
Contains typical data format requirements for each customization type (e.g., JSONL structures, max tokens, and example counts). Notes that availability varies by region. Mentions industry-standard considerations such as epochs, batch sizes, learning rates, and RL algorithm choices, but prices, quotas, and region-specific limits are not provided here. License: MIT.
# Amazon Bedrock Model Customization Complete guide to customizing Amazon Bedrock foundation models through fine-tuning, continued pre-training, reinforcement fine-tuning, and distillation. ## Overview Amazon Bedrock Model Customization allows you to adapt foundation models to your specific use cases without managing infrastructure. Four customization approaches are available: ### 1. Fine-Tuning (Supervised Learning) Adapt models to specific tasks using labeled examples (input-output pairs). Best for: - Task-specific optimization (classification, extraction, generation) - Improving responses for domain terminology - Teaching specific output formats - **Typical gains**: 20-40% accuracy improvement ### 2. Continued Pre-Training (Domain Adaptation) Continue training on unlabeled domain-specific text to build domain knowledge. Best for: - Medical, legal, financial, technical domains - Proprietary knowledge bases - Industry-specific language - **Typical gains**: 15-30% domain accuracy improvement ### 3. Reinforcement Fine-Tuning (NEW 2025) Use reinforcement learning with human feedback (RLHF) or AI feedback (RLAIF) for alignment. Best for: - Improving response quality and safety - Align
- Overview
- 1. Fine-Tuning (Supervised Learning)
- 2. Continued Pre-Training (Domain Adaptation)
- 3. Reinforcement Fine-Tuning (NEW 2025)
- 4. Distillation (Teacher-Student)
- Supported Models
- Training Data Formats
- Fine-Tuning Format (JSONL)
- Continued Pre-Training Format (JSONL)
- Reinforcement Fine-Tuning Format (JSONL)
- Distillation Format (No Training Data Required)
- Quick Start
- 1. Prepare Training Data
- 2. Upload to S3
What does the bedrock-fine-tuning skill do?
Amazon Bedrock Model Customization with fine-tuning, continued pre-training, reinforcement fine-tuning (NEW 2025 - 66% accuracy gains), and distillation. Create customization jobs, monitor training, deploy custom models, and evaluate performance. Use when customizing Claude, Titan, or other Bedrock models for domain-specific tasks, adapting to proprietary data, improving accuracy on specialized workflows, or distilling large models to smaller ones.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill bedrock-fine-tuning --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.