- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add awslabs/agent-plugins --skill finetuning -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add awslabs/agent-plugins --skill finetuning -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add awslabs/agent-plugins --skill finetuning -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add awslabs/agent-plugins --skill finetuning -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/awslabs/agent-plugins/tree/HEAD/plugins/sagemaker-ai/skills/finetuning來自發布者
Before starting this workflow, verify:
A use_case_spec.md file exists
use-case-specification skill first, then resumeA fine-tuning technique (SFT, DPO, RLVR, RLAIF, or CPT/RFT (for Nova)) and base model have already been selected
model-selection and/or finetuning-technique skills to collect what's missing, then resumeA base model name available on SageMakerHub has been identified
model-selection skill to get itmodel-selection retrieves, as it may differ from other commonly used names for the same modelThe SDK environment has been verified (SDK version, region, execution role)
sdk-getting-started skill first, then resumeA training dataset uploaded to a bucket in the environment's default region.
run_cell is available, offer to run it. Otherwise, tell them to run cells one by one (mention ipykernel requirement).python3 <script>.py⏸ Wait for user.
Read references/code_output_guide.md for output format rules, then read the code template matching the finetuning strategy:
code_templates/sft.pycode_templates/dpo.pycode_templates/rlvr.pycode_templates/rlaif_builtin.pycode_templates/rlaif_custom_prompt.pyThe template is a Python file where each # Cell N: Label comment marks the start of a new section. Split on these markers — everything between one marker and the next becomes one unit of output.
code_output_guide.mdmeta-):
ACCEPT_EULA = False line from the config cellaccept_eula=ACCEPT_EULA, line from the trainer callmax_epochs or lr_warmup_steps_ratio from the Configure Trainer section and the Hyperparameter Overrides sectionIn the 'Setup & Credentials' cell, populate:
BASE_MODEL
MODEL_PACKAGE_GROUP_NAME
use_case_spec.md if needed)[a-zA-Z0-9](-*[a-zA-Z0-9]){0,62}customer-support-chatbot-v1Save notebook
references/rlvr_reward_function.md section "Helping Users Create Custom Reward Functions"CUSTOM_REWARD_FUNCTION in the Notebook with the ARN of the reward function (either given directly by the user, or from the function generation code as evaluator.arn).Read references/rlaif_guide.md and follow its instructions.
meta-)
ACCEPT_EULA = True and uncomment accept_eula=ACCEPT_EULA in the generated notebook. If the user declines, leave ACCEPT_EULA = False and warn that training will fail without acceptance.ACCEPT_EULA variable and accept_eula parameter should already be omitted from the notebook (see Step 1.3).After generating the code, offer to run it. Training can take hours depending on your dataset and model.
Notebook mode: If run_cell is available, offer to run the cells. Otherwise tell the user to run cells themselves.
Script mode: Present the user with options:
"Would you like me to:
- Leave it to you — run with
python scripts/[script_name]- Run it and wait until it's done
- Start it but don't wait — we can check status later"
trainer.train(wait=True) blocks until complete. Report final status.wait=True to wait=False in the script, execute, report the training job name.Checking status:
describe-training-job --training-job-name NAME → TrainingJobStatus, FailureReason, SecondaryStatusTransitionslist-model-packages --model-package-group-name GROUP_NAME --sort-by CreationTime --sort-order Descending --max-results 1Showing results after completion:
scripts/mlflow_reference.py as the pattern to query MLflow metricsCRITICAL:
If the user wants to finetune a model they had already customized, follow the instructions in references/continuous_customization.md
rlvr_reward_function.md - Lambda reward function creation guide (RLVR only)templates/rlvr_reward_function_source_template.py - Lambda reward function source template for open-weights models (RLVR only)templates/nova_rlvr_reward_function_source_template.py - Lambda reward function source template for Nova 2.0 Lite (RLVR only)code_templates/sft.py - Complete notebook template for Supervised Fine-Tuning (OSS path)code_templates/dpo.py - Complete notebook template for Direct Preference Optimization (OSS path)code_templates/rlvr.py - Complete notebook template for Reinforcement Learning from Verifiable Rewards (OSS path)references/continuous_customization.md - Instructions on fine-tuning an already fine-tuned model.rlaif_guide.md - instructions on RLAIF finetuning optionsrlaif_builtin.py - Code template for RLAIF with built-in judge promptrlaif_custom_prompt.py - Code template for RLAIF with custom judge promptGenerates code that fine-tunes a base model using SageMaker serverless training jobs. Use when the user says "start training", "fine-tune my model", "I'm ready to train", or when the plan reaches the
---name: finetuningdescription: Generates code that fine-tunes a base model using SageMaker serverless training jobs. Use when the user says…---# plugins/sagemaker-ai/skills/finetuning/SKILL.md總覽
Generates code that fine-tunes a base model using SageMaker serverless training jobs. Use when the user says "start training", "fine-tune my model", "I'm ready to train", or when the plan reaches the finetuning step. Supports SFT, DPO, RLVR, and RLAIF trainers, including RLVR Lambda reward function and RLAIF custom prompt creation.
常見問題
從發布者下載 plugins/sagemaker-ai/skills/finetuning/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/awslabs/agent-plugins。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。