Model Deployment

Name: Model Deployment
Author: awslabs

awslabs/agent-plugins

121 installs
850 repo stars
Updated August 3, 2026
awslabs/agent-plugins

model-deployment is a Claude skill that generates code to deploy LoRA fine-tuned Nova or OSS models from SageMaker Serverless Model Customization to a SageMaker endpoint or Bedrock.

About

This skill generates code to deploy fine-tuned models from SageMaker Serverless Model Customization to a SageMaker endpoint or Bedrock. It identifies the model type (Nova vs OSS), lets the user choose a deployment target, displays the license, and generates the deployment code. A developer uses it when they want to make a LoRA fine-tuned model available for inference.

Deploys fine-tuned Nova and OSS models from SageMaker Serverless Model Customization
Identifies the correct pathway and targets: SageMaker endpoint or Bedrock
Generates deployment code and handles endpoint configuration and license display

Model Deployment by the numbers

121 all-time installs (skills.sh)
Ranked #765 of 2,064 Data Science & ML skills by installs in the Skillselion catalog
Data as of Aug 4, 2026 (Skillselion catalog sync)

At a glance

model-deployment capabilities & compatibility

Capabilities: model evaluation · finetuning technique · model selection
Works with: aws
Use cases: devops · orchestration

From the docs

What model-deployment says it does

Generates code that deploys fine-tuned models from SageMaker Serverless Model Customization to SageMaker endpoints or Bedrock.

SKILL.md

Full Fine-Tuning (FFT) — only LoRA fine-tuned models are supported

SKILL.md

npx skills add https://github.com/awslabs/agent-plugins --skill model-deployment

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

[![Listed on Skillselion](https://skillselion.com/badge/skills/awslabs/agent-plugins/model-deployment.svg)](https://skillselion.com/skills/awslabs/agent-plugins/model-deployment)

Installs	121
repo stars	★ 850
Last updated	August 3, 2026
Repository	awslabs/agent-plugins ↗

What it does

Deploy a LoRA fine-tuned Nova or OSS model to a SageMaker endpoint or Bedrock.

Who is it for?

Developers deploying a LoRA fine-tuned Nova or OSS model for inference

Skip if: Base models, non-SageMaker fine-tuning, or full fine-tuned (FFT) models

When should I use this skill?

User says deploy my model, create an endpoint, or asks about deployment options

What you get

Generated deployment code plus a running SageMaker endpoint or Bedrock deployment for the model.

Deployment code
A configured SageMaker endpoint or Bedrock deployment
Post-deployment summary with invoke code and cost

By the numbers

6-step deployment workflow
2 deployment targets (SageMaker, Bedrock)

Files

SKILL.mdMarkdownGitHub ↗

Model Deployment

Identifies the correct deployment pathway based on model characteristics and generates deployment code.

Scope

This skill supports deploying Nova and OSS models that were fine-tuned through SageMaker Serverless Model Customization only.

Not supported:

Base models (not fine-tuned)
Models fine-tuned through other processes
Full Fine-Tuning (FFT) — only LoRA fine-tuned models are supported

Prerequisites

The SDK environment has been verified (SDK version, region, execution role). If not done, activate the sdk-getting-started skill first.

---

Principles

1. One thing at a time. Each response advances exactly one decision. 2. Confirm before proceeding. Wait for the user to agree before moving on. But don't re-ask questions already answered in the conversation — use what you know. 3. Don't read files until you need them. Only read pathway references after the pathway is confirmed. 4. Use what you know. If conversation history or artifacts already answer a question, confirm your understanding instead of asking again.

Workflow

Step 1: Identify the Training Job

You need the training job name or ARN. Check the conversation history first — the user may have already mentioned it, or it may be available from earlier steps in the workflow (e.g., fine-tuning). If not, ask the user.

Once you have the training job name or ARN, use the AWS MCP tool to look it up:

1. Use the AWS MCP tool describe-training-job and extract:

S3 output path (from ModelArtifacts.S3ModelArtifacts or OutputDataConfig.S3OutputPath)
IAM role ARN (from RoleArn)
Region

2. Use the AWS MCP tool list-tags on the training job ARN and extract:

Model ID from the sagemaker-studio:jumpstart-model-id tag

3. Determine the model type from the model ID:

Contains "nova" (nova-micro, nova-lite, nova-pro) → Nova
Llama, Mistral, Qwen, GPT-OSS, DeepSeek, etc. → OSS

Unsupported models: This skill only supports OSS and Nova models that were LoRA fine-tuned through SageMaker Serverless Model Customization. If the model doesn't match, tell the user this skill can't help and suggest the finetuning skill.

Step 2: Determine Eligible Deployment Targets

Use the following table:

Model Type	Eligible Targets
OSS	SageMaker, Bedrock
Nova	SageMaker, Bedrock

If only one target is eligible, confirm it with the user. Use details from Step 5.

If multiple targets are eligible, help the user decide. Use details from Step 5.

If no targets are eligible, tell the user and explain why.

Step 3: Let the User Choose a Deployment Target

Present the eligible options to the user. Present these details to help them decide between SageMaker and Bedrock, if both are available options:

SageMaker Endpoint:

Dedicated compute resources for consistent performance
Control instance types and scaling
Best for predictable workloads with specific latency requirements

Bedrock:

Fully managed serverless inference
Auto-scales instantly with no capacity planning
Pay per request
Best for variable workloads with fluctuating demand

Do NOT make a recommendation. Let the user choose.

Do NOT mention technical details like merged/unmerged weights, reference files, or APIs, unless the user asks.

⏸ Wait for user to select a deployment option.

Step 4: Display License Agreement

Before proceeding to deployment, display the model's license or service terms to the user.

1. Read references/model-licenses.md and look up the model by its model ID (determined in Step 1). 2. Follow the instructions in the Notes column — use the exact phrasing provided. 3. If the model ID is not found in the table, warn the user that you could not find license information for their model and recommend they verify the license independently before proceeding.

⏸ Wait for the user to confirm before proceeding.

Step 5: Follow Pathway Workflow

Read the reference file for the selected pathway and follow its instructions.

Model Type	Deployment Target	Reference
OSS	SageMaker	`references/deploy-oss-sagemaker.md`
OSS	Bedrock	`references/deploy-oss-bedrock.md`
Nova	SageMaker	`references/deploy-nova-sagemaker.md`
Nova	Bedrock	`references/deploy-nova-bedrock.md`

Step 6: Post-Deployment Summary

After deployment completes, provide the user with a summary. Cover these topics, using details from the pathway reference doc you followed in Step 5:

What was deployed — endpoint or model name, ARN, status
How to use it — sample invoke code for the specific deployment target
Cost — billing model (instance-based vs. pay-per-request) and what to expect
Cleanup — how to delete the endpoint or model when done

Troubleshooting

How to check if a model was LoRA or FFT fine-tuned

If deployment fails unexpectedly, the model may have been full fine-tuned (FFT) rather than LoRA. To check, download the training job's hydra config from its S3 output path at .hydra/config.yaml:

peft_config populated (r, alpha, dropout, etc.) → LoRA (supported)
peft_config: null → FFT (not supported by this skill)

# Cell 0 [markdown]: Model Deployment — Bedrock

# Cell 1: Setup

%pip install --upgrade sagemaker>=3.7.1 --quiet  # NOTEBOOK_ONLY

# Cell 2: Configuration

import os
import json
import boto3

os.environ["AWS_DEFAULT_REGION"] = "[REGION]"

from sagemaker.core.resources import TrainingJob
from sagemaker.serve.bedrock_model_builder import BedrockModelBuilder
from sagemaker.core import Attribution, set_attribution
from pprint import pprint

set_attribution(Attribution.SAGEMAKER_AGENT_PLUGIN)

REGION = "[REGION]"
TRAINING_JOB_NAME = "[TRAINING_JOB_NAME]"
ROLE_ARN = "[ROLE_ARN]"
CUSTOM_MODEL_NAME = "[CUSTOM_MODEL_NAME]"

# Cell 3: Build and Deploy to Bedrock

training_job = TrainingJob.get(training_job_name=TRAINING_JOB_NAME)
print(f"Training job status: {training_job.training_job_status}")

bedrock_builder = BedrockModelBuilder(model=training_job)

deployment_result = bedrock_builder.deploy(
    role_arn=ROLE_ARN,
    custom_model_name=CUSTOM_MODEL_NAME,
)

deployment_arn = deployment_result["customModelDeploymentArn"]
pprint(f"Deployment Result: {deployment_result}")

# Cell 4: Test Inference

bedrock_runtime = boto3.client("bedrock-runtime", region_name=REGION)
message = "What is the capital of France?"
print(f"Model Inference Message: {message}")
resp = bedrock_runtime.converse(
    modelId=deployment_arn,
    messages=[{"role": "user", "content": [{"text": message}]}],
    inferenceConfig={"maxTokens": 100, "temperature": 0.7},
)

response_str = resp["output"]["message"]["content"][0]["text"]
print(f"Model Response: {response_str}")

# Save manifest
from pathlib import Path
manifest_dir = Path("[PROJECT_DIR]") / "manifests"
manifest_dir.mkdir(parents=True, exist_ok=True)
manifest_path = manifest_dir / f"deploy-{CUSTOM_MODEL_NAME}.json"
manifest_path.write_text(json.dumps({
    "custom_model_name": CUSTOM_MODEL_NAME,
}, indent=2))
print(f"Manifest saved: {manifest_path}")

# Cell 0 [markdown]: Model Deployment — SageMaker

# Cell 1: Setup

%pip install --upgrade sagemaker>=3.7.1 --quiet  # NOTEBOOK_ONLY

# Cell 2: Configuration

import os
import json

os.environ["AWS_DEFAULT_REGION"] = "[REGION]"

from sagemaker.core.resources import TrainingJob
from sagemaker.serve import ModelBuilder
from sagemaker.core import Attribution, set_attribution

set_attribution(Attribution.SAGEMAKER_AGENT_PLUGIN)

TRAINING_JOB_NAME = "[TRAINING_JOB_NAME]"
ROLE_ARN = "[ROLE_ARN]"
INSTANCE_TYPE = "[INSTANCE_TYPE]"
ENDPOINT_NAME = "[ENDPOINT_NAME]"

# Cell 3: Build Model

training_job = TrainingJob.get(training_job_name=TRAINING_JOB_NAME)
print(f"Training job: {training_job.training_job_name}")

model_builder = ModelBuilder(
    model=training_job,
    role_arn=ROLE_ARN,
    instance_type=INSTANCE_TYPE,
)
model = model_builder.build()
print(f"Model: {model.model_name}")
print(f"Image: {model_builder.image_uri}")
print(f"Env vars: {model_builder.env_vars}")

# Cell 4: Deploy Endpoint

endpoint = model_builder.deploy(endpoint_name=ENDPOINT_NAME)
print(f"Endpoint: {endpoint.endpoint_name}")
print(f"Status: {endpoint.endpoint_status}")

# Cell 5: Test Inference

output = endpoint.invoke(
    body=json.dumps({
        "messages": [{"role": "user", "content": "What is the capital of France?"}],
        "max_tokens": 50,
    }),
    content_type="application/json",
)
print(f"Response: {json.loads(output.body.read())}")

# Save manifest
from pathlib import Path
manifest_dir = Path("[PROJECT_DIR]") / "manifests"
manifest_dir.mkdir(parents=True, exist_ok=True)
manifest_path = manifest_dir / f"deploy-{ENDPOINT_NAME}.json"
manifest_path.write_text(json.dumps({
    "endpoint_name": ENDPOINT_NAME,
}, indent=2))
print(f"Manifest saved: {manifest_path}")

# Cell 0 [markdown]: Model Deployment — Bedrock

# Cell 1: Setup

%pip install --upgrade sagemaker>=3.7.1 --quiet  # NOTEBOOK_ONLY

# Cell 2: Configuration

import boto3
import json
import time
from sagemaker.serve.bedrock_model_builder import BedrockModelBuilder
from sagemaker.core.resources import TrainingJob
from sagemaker.core import Attribution, set_attribution

set_attribution(Attribution.SAGEMAKER_AGENT_PLUGIN)

REGION = "[REGION]"
TRAINING_JOB_NAME = "[TRAINING_JOB_NAME]"
ROLE_ARN = "[ROLE_ARN]"
MODEL_NAME = "[MODEL_NAME]"

sm = boto3.client("sagemaker", region_name=REGION)
s3 = boto3.client("s3", region_name=REGION)

# Cell 3: Flatten S3 Structure and Start Import

# BedrockModelBuilder passes the root model artifacts path to Bedrock CMI,
# but Bedrock expects config.json at the root of the URI. This cell copies
# files from checkpoints/hf_merged/ to the model artifacts root (server-side).

tj = sm.describe_training_job(TrainingJobName=TRAINING_JOB_NAME)
root = tj["ModelArtifacts"]["S3ModelArtifacts"]
parts = root.replace("s3://", "").split("/", 1)
bucket, root_prefix = parts[0], parts[1].rstrip("/") + "/"
hf_prefix = root_prefix + "checkpoints/hf_merged/"

resp = s3.list_objects_v2(Bucket=bucket, Prefix=root_prefix + "config.json", MaxKeys=1)
if resp.get("KeyCount", 0) > 0:
    print("Files already at root, skipping copy")
else:
    paginator = s3.get_paginator("list_objects_v2")
    copied = 0
    for page in paginator.paginate(Bucket=bucket, Prefix=hf_prefix):
        for obj in page.get("Contents", []):
            filename = obj["Key"].replace(hf_prefix, "")
            if not filename or filename.endswith("/"):
                continue
            s3.copy_object(
                Bucket=bucket,
                CopySource={"Bucket": bucket, "Key": obj["Key"]},
                Key=root_prefix + filename,
            )
            copied += 1
    print(f"Copied {copied} files to root")

training_job = TrainingJob.get(training_job_name=TRAINING_JOB_NAME, region=REGION)
builder = BedrockModelBuilder(model=training_job)

result = builder.deploy(
    job_name=MODEL_NAME,
    imported_model_name=MODEL_NAME,
    role_arn=ROLE_ARN,
)

job_arn = result["jobArn"]
print(f"Import job created: {job_arn}")

# Cell 4: Wait for Import to Complete

bedrock = boto3.client("bedrock", region_name=REGION)

while True:
    resp = bedrock.get_model_import_job(jobIdentifier=job_arn)
    status = resp["status"]
    print(f"Status: {status}")

    if status == "Completed":
        model_arn = resp["importedModelArn"]
        print(f"\nModel imported successfully!")
        print(f"Model ARN: {model_arn}")
        break
    elif status in ("Failed", "Stopped"):
        raise RuntimeError(f"Import {status}: {resp.get('failureMessage', 'Unknown error')}")

    time.sleep(30)

# Cell 5: Test Inference

print("Testing inference (model may need a few minutes to warm up)...")
bedrock_runtime = boto3.client("bedrock-runtime", region_name=REGION)

for attempt in range(1, 25):
    try:
        response = bedrock_runtime.invoke_model(
            modelId=model_arn,
            body=json.dumps({
                "prompt": "What is the capital of France?",
                "max_gen_len": 50,
                "temperature": 0.7,
            }),
        )
        result = json.loads(response["body"].read())
        print(f"Response: {json.dumps(result)[:300]}")
        break
    except bedrock_runtime.exceptions.ModelNotReadyException:
        print(f"  Attempt {attempt}: Model not ready, waiting 30s...")
        time.sleep(30)
else:
    print("Model did not become ready after 12 minutes.")

# Save manifest
from pathlib import Path
manifest_dir = Path("[PROJECT_DIR]") / "manifests"
manifest_dir.mkdir(parents=True, exist_ok=True)
manifest_path = manifest_dir / f"deploy-{TRAINING_JOB_NAME}.json"
manifest_path.write_text(json.dumps({
    "model_id": model_arn,
}, indent=2))
print(f"Manifest saved: {manifest_path}")

# Cell 0 [markdown]: Model Deployment — SageMaker

# Cell 1: Setup

%pip install --upgrade sagemaker>=3.7.1 --quiet  # NOTEBOOK_ONLY

# Cell 2: Configuration

import os
import json

os.environ["AWS_DEFAULT_REGION"] = "[REGION]"

from sagemaker.core.resources import TrainingJob
from sagemaker.serve import ModelBuilder
from sagemaker.core import Attribution, set_attribution

set_attribution(Attribution.SAGEMAKER_AGENT_PLUGIN)

TRAINING_JOB_NAME = "[TRAINING_JOB_NAME]"
ROLE_ARN = "[ROLE_ARN]"
INSTANCE_TYPE = "[INSTANCE_TYPE]"
ENDPOINT_NAME = "[ENDPOINT_NAME]"
ADAPTER_IC_NAME = f"{ENDPOINT_NAME}-adapter"
ACCEPT_EULA = [ACCEPT_EULA]  # True if user accepted the license in Step 4, False otherwise

# Cell 3: Build Model

training_job = TrainingJob.get(training_job_name=TRAINING_JOB_NAME)
print(f"Training job: {training_job.training_job_name}")
print(f"Model package: {training_job.output_model_package_arn}")

model_builder = ModelBuilder(
    model=training_job,
    role_arn=ROLE_ARN,
    instance_type=INSTANCE_TYPE,
)
model_builder.accept_eula = ACCEPT_EULA
model = model_builder.build(model_name=ENDPOINT_NAME)
print(f"Model: {model.model_arn}")

# Cell 4: Deploy Endpoint

endpoint = model_builder.deploy(
    endpoint_name=ENDPOINT_NAME,
    inference_component_name=ADAPTER_IC_NAME,
)
print(f"Endpoint: {endpoint.endpoint_name}")

# Cell 5: Test Inference

output = endpoint.invoke(
    body=json.dumps({
        "inputs": "What is the capital of France?",
        "parameters": {"max_new_tokens": 50},
    }),
    inference_component_name=ADAPTER_IC_NAME,
)
print(f"Response: {output.body.read()}")

# Save manifest
from pathlib import Path
manifest_dir = Path("[PROJECT_DIR]") / "manifests"
manifest_dir.mkdir(parents=True, exist_ok=True)
manifest_path = manifest_dir / f"deploy-{ENDPOINT_NAME}.json"
manifest_path.write_text(json.dumps({
    "endpoint_name": ENDPOINT_NAME,
}, indent=2))
print(f"Manifest saved: {manifest_path}")

Code Output Guide

Mode Selection

Ask the user once before generating code: "Would you like me to generate a Jupyter notebook or a Python script?"

If the output format has already been decided in the conversation context, keep consistent — do not re-ask.

Shared Rules (Both Modes)

Use EXACTLY the imports shown in each code template — do not add extras
Replace [PLACEHOLDER] values with user-specific configuration
Include set_attribution(Attribution.SAGEMAKER_AGENT_PLUGIN) in the setup cell/section

Reading Code Templates

Templates use # Cell N: Label markers to delimit sections. # NOTEBOOK_ONLY skips a line in script mode; # NOTEBOOK_ONLY_SECTION on a # Cell N: line skips the entire section.

Notebook Mode

Write a .ipynb file in <project-dir>/notebooks/.

Naming and appending:

Notebook path: <project-dir>/notebooks/<project-name>.ipynb
If the notebook already exists → ask: _"Would you like me to append cells to the existing notebook, or create a new one?"_
If it doesn't exist → create it
When appending, use the template's # Cell 0 [markdown]: cell as the section divider before the new cells

Formatting:

Use your file write tool to create the complete notebook JSON, OR use notebook MCP tools (create_notebook, add_cell) if available
Do NOT use bash commands, shell scripts, or echo/cat piping
2-space JSON indentation
Each source line is a separate string ending with \n (except the last)
Escape quotes: \"
No trailing commas

Structure:

Wrap cells in {"cells": [...], "metadata": {...}, "nbformat": 4, "nbformat_minor": 4}
Code cells: cell_type, execution_count: null, metadata: {}, outputs: [], source: [...]
Markdown cells: cell_type: "markdown", no execution_count or outputs
# Cell 0 [markdown]: becomes a markdown cell; all others become code cells

Execution:

If notebook execution tools are available (e.g., run_cell MCP), offer to run cells for the user. If not available, tell the user to run cells themselves.
Do NOT use bash commands or inline scripts to execute notebook cells.

Script Mode

Write a numbered .py file in <project-dir>/scripts/.

Naming:

Format: NN_<descriptive_name>.py (e.g., 01_sft_finetuning.py) — use the next available number in <project-dir>/scripts/

Formatting:

Plain Python file, standard text
Use # %% cell markers to preserve logical sections (IDE-compatible)
Include a docstring at the top describing what the script does
# Cell 0 [markdown]: → a comment block or docstring

Dependencies:

Install any required pip packages directly (e.g., pip install sagemaker>=3.7.1) before writing or running the script. Do not embed install commands in the script itself.

Execution:

Run the script using standard Python execution (python3 <script>.py).

Resumption After Interruption

If the conversation was interrupted while a job was running (e.g., context compaction, user stopped and restarted, connection drop), do NOT re-run the script. Instead, check for an existing job by name or ARN from the conversation context or PLAN.md, and monitor its status rather than launching a duplicate.

Deploy Nova LoRA to Bedrock (PySDK BedrockModelBuilder)

Scenario

Model Type: Nova
Fine-tuning Method: LoRA
Deployment Target: Bedrock Custom Model
Approach: SageMaker PySdk BedrockModelBuilder

Overview

Uses the SageMaker PySdk BedrockModelBuilder to deploy a Nova fine-tuned LoRA model to Bedrock as a Custom Model. The builder auto-detects Nova models and calls CreateCustomModel.

Required inputs (collected in the steps below):

Training job name
Custom model name
IAM role ARN
AWS region

Prerequisites

Requires SageMaker Python SDK >= 3.7.0 with BedrockModelBuilder Nova support (installed by Cell 1).

Workflow

Important Instructions

Make sure to use dedicated tools instead of bash commands whenever possible

Step 1: Gather Training Job Name

The training job name was identified in Step 1 of the main workflow. Confirm you have it.

Step 2: Gather Custom Model Name

For this step, you need: a name for the deployed custom model.

Suggest a name based on the training job or use case, e.g., nova-micro-bedrock-<timestamp>. Ask the user to confirm or provide their own.

⏸ Wait for user before moving on.

Step 3: Verify IAM Role

Use the IAM role from the training job (extracted in Step 1 of the main workflow via describe-training-job). We'll assume this role has the necessary permissions for Bedrock deployment.

Step 4: Confirm Region

The region was identified in Step 1 of the main workflow. Nova → Bedrock deployment is currently only supported in us-east-1. If the training job is in a different region, tell the user that Bedrock deployment is not supported for this model in this region.

Step 5: Confirm Configuration

"Here's the deployment setup:

- Deployment target: Bedrock (Custom Model)

- Training Job: [job-name]

- Custom Model Name: [name]

- IAM Role: [arn]

- Region: us-east-1

Does this look right?"

⏸ Wait for user approval.

Step 6: Generate Code

Read ../references/code_output_guide.md for output format rules.

If a project directory already exists (from earlier in the workflow), use it. Otherwise, activate the directory-management skill to set one up.

⏸ Wait for user.

Code Structure

Markdown Header

{
  "cell_type": "markdown",
  "metadata": {},
  "source": [
    "# Deploy Nova to Bedrock"
  ]
}

Cells

Each cell's content comes from ../code_templates/deploy-nova-bedrock.py, split on the # Cell N: comments. Each marker starts a new notebook cell — everything between one marker and the next becomes that cell's content.

Cell 1: Setup (pip install)
Cell 2: Configuration (env vars, imports, placeholders)
Cell 3: Build and Deploy to Bedrock (blocks until deployment is Active)
Cell 4: Test Inference

Placeholders

Cell 2:

[REGION] → AWS region (us-east-1)
[TRAINING_JOB_NAME] → SageMaker training job name
[ROLE_ARN] → IAM role ARN
[CUSTOM_MODEL_NAME] → Name for the custom model

All other cells have no placeholders.

Step 7: Provide Run Instructions

To run:
1. Cell 1 — install SDK packages
2. Cell 2 — set configuration values
3. Cell 3 — creates custom model via BedrockModelBuilder and deploys (blocks until Active)
4. Cell 4 — test inference with a sample prompt via Converse API

Common Issues

"ServiceQuotaExceededException: The number of custom models in Creating status has reached the quota limit": Too many concurrent model creations. Wait for in-progress models to finish, or delete old custom models.
"No module named 'sagemaker.serve.bedrock_model_builder'": Re-run Cell 1 to install the required packages, then restart the kernel.
"Access denied to S3": Add S3 read permissions to the IAM role for the model artifacts bucket.
"Provided IAM role could not be assumed": Ensure role has trust policy for bedrock.amazonaws.com.
Deployment status "Failed": Check CloudTrail for the CreateCustomModel event to see the failure reason.

Deploy Nova LoRA to SageMaker

Scenario

Model Type: Nova
Fine-tuning Method: LoRA
Deployment Target: SageMaker Single Model Endpoint
Approach: SageMaker ModelBuilder

Overview

Deploys a Nova fine-tuned model to a SageMaker endpoint using ModelBuilder.

Nova deploys as a model-on-variant (no inference components), so you invoke the endpoint directly without specifying an InferenceComponentName.

Required inputs (collected in the steps below):

Training job name
Instance type
IAM execution role ARN
AWS region
Endpoint name

Prerequisites

Requires SageMaker Python SDK >= 3.7.0 (installed by Cell 1).

Workflow

Important Instructions

Make sure to use dedicated tools instead of bash commands whenever possible

Step 1: Gather Training Job Name

The training job name was identified in Step 1 of the main workflow. Confirm you have it.

Step 2: Determine Instance Type

For this step, you need: the instance type.

First, determine the Nova variant from the training job's model package. Use your AWS tool to run sagemaker describe-training-job for the training job name and extract the OutputModelPackageArn from the response. Then inspect the model package to find the hub_content_name (e.g., nova-textgeneration-micro).

Supported instances by Nova variant (smallest to largest). Larger instances support longer context lengths.

Nova Micro (nova-textgeneration-micro): ml.g5.12xlarge, ml.g5.24xlarge, ml.g6.12xlarge, ml.g6.24xlarge, ml.g6.48xlarge, ml.p5.48xlarge

Nova Lite (nova-textgeneration-lite): ml.g6.48xlarge, ml.p5.48xlarge

Nova Lite v2 (nova-textgeneration-lite-v2): ml.p5.48xlarge

Nova Pro (nova-textgeneration-pro): ml.g6.48xlarge, ml.p5.48xlarge

Present the supported instance types and ask which one the user would like to use. The larger instances will be more expensive, but have larger context windows.

⏸ Wait for user to confirm before moving on.

Step 3: Verify IAM Role

Use the IAM role from the training job (extracted in Step 1 of the main workflow via describe-training-job). This role should already have the necessary SageMaker and S3 permissions. Confirm with the user.

Step 4: Confirm Region

The region was identified in Step 1 of the main workflow. Nova deployment is only supported in: us-east-1, us-west-2, eu-west-2, ap-northeast-1. If the region isn't supported, tell the user that SageMaker deployment is not supported for this model in this region.

Step 5: Choose Endpoint Name

Suggest a name based on the model, e.g., nova-micro-deploy-<timestamp>. Ask the user to confirm or provide their own.

⏸ Wait for user before moving on.

Step 6: Confirm Configuration

"Here's the deployment setup:

- Model: [base-model-name] fine-tuned with LoRA (e.g., "Nova Micro fine-tuned with LoRA")

- Deployment target: SageMaker Endpoint

- Training Job: [name]

- Instance Type: [type]

- IAM Role: [arn]

- Region: [region]

- Endpoint Name: [name]

Does this look right?"

⏸ Wait for user approval.

Step 7: Generate Code

Read ../references/code_output_guide.md for output format rules.

If a project directory already exists (from earlier in the workflow), use it. Otherwise, activate the directory-management skill to set one up.

⏸ Wait for user.

Code Structure

Markdown Header

{
  "cell_type": "markdown",
  "metadata": {},
  "source": [
    "# Deploy Nova Fine-Tuned Model to SageMaker"
  ]
}

Cells

Each cell's content comes from ../code_templates/deploy-nova-sagemaker.py, split on the # Cell N: comments. Each marker starts a new notebook cell — everything between one marker and the next becomes that cell's content.

Cell 1: Setup (pip install)
Cell 2: Configuration
Cell 3: Build Model
Cell 4: Deploy Endpoint
Cell 5: Test Inference

Placeholders

Cell 2:

[REGION] → AWS region
[TRAINING_JOB_NAME] → Training job name
[ROLE_ARN] → IAM execution role ARN
[INSTANCE_TYPE] → SageMaker instance type (e.g., ml.g5.12xlarge)
[ENDPOINT_NAME] → Endpoint name

Step 8: Provide Run Instructions

To run:
1. Cell 1 — install SDK packages, then restart the kernel before continuing
2. Cell 2 — set configuration values
3. Cell 3 — build model via ModelBuilder (~30s, creates SageMaker Model resource)
4. Cell 4 — deploy endpoint (waits for InService, ~10-15 min)
5. Cell 5 — test inference with a sample prompt

Common Issues

"No module named 'sagemaker.core'" or "No module named 'sagemaker.train'": Re-run Cell 1 to install the required packages, then restart the kernel.
"Must setup local AWS configuration with a region": Set AWS_DEFAULT_REGION env var or configure ~/.aws/config
"Cannot create already existing endpoint configuration": An endpoint with that name already exists. Use a different name or delete the existing one first.
Endpoint fails to reach InService: Check CloudWatch logs for the endpoint. Common causes: wrong instance type for the model size, or IAM role missing permissions.

Deploy OSS Merged LoRA to Bedrock CMI

Scenario

Model Type: OSS (Open Source)
Fine-tuning Method: LoRA
Merge Status: Merged (merge_weights: true)
Deployment Target: Bedrock Custom Model Import (CMI)
Approach: SageMaker PySdk BedrockModelBuilder

Overview

Uses the SageMaker PySdk BedrockModelBuilder to import a fine-tuned model into Bedrock as a Custom Model Import (CMI). The builder auto-resolves model artifacts from a training job.

Required inputs (collected in the steps below):

Training job name
Model name
IAM role ARN (with Bedrock trust policy and S3 read access)
AWS region (must be us-east-1, us-east-2, us-west-2, or eu-central-1)

Prerequisites

Model Size Limit

Bedrock CMI has a 200GB limit for text models (100GB for multimodal). Check before proceeding using the AWS MCP tool list-objects-v2 (S3 service) with the bucket and prefix <prefix>/checkpoints/hf_merged/. Sum the Size field from all returned objects to determine total size. If the model exceeds 200GB, this pathway cannot be used.

Required Files

The hf_merged/ folder must contain: .safetensors files, config.json, tokenizer.json, tokenizer_config.json.

SDK Version

Requires sagemaker>=3.7.0 with BedrockModelBuilder support.

Workflow

Step 1: Gather Training Job Name

The training job name was identified in Step 1 of the main workflow. Confirm you have it.

Step 2: Gather Model Name

Suggest a name for the deployed model based on the training job or use case. Format: lowercase, alphanumeric with hyphens. Confirm with the user.

Step 3: Verify IAM Role

Use the IAM role from the training job (extracted in Step 1 of the main workflow via describe-training-job). We'll assume this role has the necessary permissions for Bedrock deployment.

Step 4: Confirm Region

Bedrock CMI is available in: us-east-1, us-east-2, us-west-2, eu-central-1.

The region was identified in Step 1. Confirm it's in the supported list. If not, tell the user that Bedrock deployment is not supported for this model in this region.

Step 5: Confirm Configuration

"Here's the deployment setup:

- Deployment target: Bedrock

- Training Job: [job-name]

- Model Name: [name]

- IAM Role: [arn]

- Region: [region]

Does this look right?"

⏸ Wait for user approval.

Step 6: Generate Code

Read ../references/code_output_guide.md for output format rules.

If a project directory already exists (from earlier in the workflow), use it. Otherwise, activate the directory-management skill to set one up.

⏸ Wait for user.

Code Structure

Markdown Header

{
  "cell_type": "markdown",
  "metadata": {},
  "source": [
    "# Deploy to Bedrock"
  ]
}

Cells

Each cell's content comes from ../code_templates/deploy-oss-bedrock.py, split on the # Cell N: comments. Each marker starts a new notebook cell — everything between one marker and the next becomes that cell's content.

Cell 1: Setup (pip install)
Cell 2: Configuration
Cell 3: Flatten S3 Structure and Start Import
Cell 4: Wait for Import to Complete
Cell 5: Test Inference

Placeholders

Cell 2:

[REGION] → AWS region
[TRAINING_JOB_NAME] → SageMaker training job name
[ROLE_ARN] → IAM role ARN with Bedrock trust policy and S3 read permissions
[MODEL_NAME] → Name for the imported model

All other cells have no placeholders.

Step 7: Provide Run Instructions

To run:
1. Cell 1 — install/upgrade SageMaker SDK
2. Cell 2 — configuration and imports
3. Cell 3 — flattens S3 structure and starts import job via BedrockModelBuilder
4. Cell 4 — waits for import to complete (typically a few minutes)
5. Cell 5 — test inference with a sample prompt

Common Issues

"Model weights are larger than 200GB": Cannot use this pathway.
"No module named 'sagemaker.serve.bedrock_model_builder'": Upgrade SDK: pip install --upgrade sagemaker>=3.7.1
Import starts but uses wrong region: Known issue — BedrockModelBuilder defaults to us-east-1. The notebook code overrides this.
"Access denied to S3": Add S3 read permissions to the IAM role for the model bucket.
"Provided IAM role could not be assumed": Ensure role has trust policy for bedrock.amazonaws.com.

Post-Deployment Summary

After the notebook runs successfully, tell the user:

Model: [MODEL_NAME] has been imported to Bedrock
How to invoke: Use the Bedrock invoke_model API with the imported model ARN
Billing: Pay per request — no cost while idle
Cleanup: When done, delete the imported model using the AWS MCP tool delete-imported-model (Bedrock service) with the model name.

Deploy OSS LoRA to SageMaker Multi-Adapter Endpoint

Scenario

Model Type: OSS (Open Source)
Fine-tuning Method: LoRA
Merge Status: Unmerged (merge_weights: false)
Deployment Target: SageMaker Multi-Adapter Endpoint
Approach: SageMaker PySdk JumpStartModel

Overview

Uses the SageMaker PySdk JumpStartModel to resolve the base model S3 URI and container image, rather than manually querying describe_hub_content and parsing the hub content document JSON. Requires sagemaker>=3.7.0.

Required inputs (collected in the steps below):

Training job name (to resolve JumpStart model ID from tags)
Instance type
IAM execution role ARN
AWS region
EULA acceptance (from Step 4 of the main workflow)

Prerequisites

SDK Version

Requires sagemaker>=3.7.0 with JumpStartModel support.

Key Gotchas

ArtifactUrl for adapter ICs: An S3 prefix (directory) works despite docs saying it must be .tar.gz. No need to repackage.
Container version: LMI 0.31.0 does NOT have the vllm_async_service entrypoint. Use OPTION_ROLLING_BATCH=lmi-dist instead.
Gated models: Use JumpStart S3 cache via ModelDataSource to avoid needing HF_TOKEN.
Endpoint config: Including ExecutionRoleArn enables inference-component mode. Do NOT include ModelName in ProductionVariants.

Workflow

Step 1: Gather Training Job Name

The training job name was identified in Step 1 of the main workflow. Confirm you have it.

This is needed to look up the JumpStart model ID (from training job tags), which JumpStartModel uses to resolve the base model S3 URI and container image automatically.

Step 2: Determine Instance Type

For this step, you need: the instance type.

Recommend an instance based on model size:

Small models (<3B): ml.g5.2xlarge (1 GPU, ~24GB)
Medium models (<10B): ml.g5.12xlarge (4 GPUs, ~96GB)
Large models (>10B): ml.g6e.48xlarge (8 GPUs, ~1TB)

Give your suggestion to the user with reasoning and ask them to confirm. If they would like a different instance type, accept their choice. If you think it will cause issues (e.g., not enough GPU memory for the model), call that out.

⏸ Wait for user to confirm before moving on.

Step 3: Verify IAM Role

Step 4: Confirm Region

The region was identified in Step 1 of the main workflow. Confirm it with the user.

Step 5: Confirm Configuration

"Here's the deployment setup:

- Target: SageMaker Multi-Adapter Endpoint

- Training Job: [name]

- Instance Type: [type]

- IAM Role: [arn]

- Region: [region]

Does this look right?"

⏸ Wait for user approval.

Step 6: Generate Code

Read ../references/code_output_guide.md for output format rules.

If a project directory already exists (from earlier in the workflow), use it. Otherwise, activate the directory-management skill to set one up.

⏸ Wait for user.

Code Structure

Markdown Header

{
  "cell_type": "markdown",
  "metadata": {},
  "source": [
    "# Deploy to SageMaker Multi-Adapter Endpoint"
  ]
}

Cells

Each cell's content comes from ../code_templates/deploy-oss-sagemaker.py, split on the # Cell N: comments. Each marker starts a new notebook cell — everything between one marker and the next becomes that cell's content.

Cell 1: Setup (pip install)
Cell 2: Configuration
Cell 3: Create Model and Endpoint
Cell 4: Create Base Model and Adapter Inference Components
Cell 5: Test Inference

Placeholders

Cell 2:

[REGION] → AWS region
[INSTANCE_TYPE] → SageMaker instance type (e.g., ml.g5.2xlarge)
[TRAINING_JOB_NAME] → Training job name (used to look up JumpStart model ID from tags)
[ROLE_ARN] → IAM execution role ARN
[ENDPOINT_NAME] → Name for the endpoint (agent should generate a reasonable default)
[ACCEPT_EULA] → True if the user accepted the license in Step 4 of the main workflow, False otherwise

Step 7: Explicitly State EULA Acceptance

Before running the notebook (either via run_cell or by the user), confirm the EULA acceptance from Step 4 of the main workflow. Tell the user: "Since you accepted the license agreement, I've set EULA acceptance to True in the deployment code." If the user did not accept the license, tell them deployment cannot continue without license acceptance.

Step 8: Provide Run Instructions

To run:
1. Cell 1 — install/upgrade SageMaker SDK
2. Cell 2 — configuration (resolves adapter path and base model metadata via JumpStartModel)
3. Cell 3 — creates model and endpoint (waits for endpoint to be InService, ~5-10 min)
4. Cell 4 — creates base model and adapter inference components (waits for both to be InService, ~5-10 min)
5. Cell 5 — test inference with a sample prompt

Common Issues

"No module named 'sagemaker.jumpstart'": Upgrade SDK: pip install --upgrade sagemaker>=3.7.1
"ModuleNotFoundError" for vllm_async_service: Using LMI 0.31.0 container. Use OPTION_ROLLING_BATCH=lmi-dist instead of OPTION_ENTRYPOINT.
Base IC fails health check: Check MinMemoryRequiredInMb fits within instance memory. Reduce if needed.
"Inference Component Name header is required": Must pass InferenceComponentName when invoking the endpoint.
Console shows "Missing required key 'ModelName'": This is a console UI issue, not a deployment issue. The endpoint works correctly.
Adapter IC fails: Verify adapter weights exist at <model-s3-uri>/checkpoints/hf/. Check that the S3 prefix is accessible.

Post-Deployment Summary

After the notebook runs successfully, tell the user:

Endpoint: [ENDPOINT_NAME] is now InService
How to invoke: Use SageMaker runtime InvokeEndpoint with InferenceComponentName set to the adapter IC name (derived from the endpoint name)
Billing: This endpoint is billed by the hour while running, even when idle. Delete it when you're done testing.
Cleanup: Delete the adapter inference component first, then the base inference component, then the endpoint using the AWS MCP tool:

1. Use delete-inference-component (SageMaker service) with the adapter IC name 2. Wait for deletion to complete, then use delete-inference-component with the base IC name 3. Wait for deletion to complete, then use delete-endpoint with the endpoint name

Model Licenses

License URLs for models supported by the deployment skill. Look up the model by its JumpStart model ID (from the sagemaker-studio:jumpstart-model-id training job tag).

Display the license or service terms to the user and follow any instructions in the Notes column.

Model ID	License	License URL	Notes
`huggingface-reasoning-qwen3-32b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-32B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-reasoning-qwen3-14b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-14B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-reasoning-qwen3-8b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-8B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-reasoning-qwen3-4b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-4B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-reasoning-qwen3-1-7b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-1.7B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-reasoning-qwen3-06b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-llm-qwen2-5-72b-instruct`	Qwen License Agreement	https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE	Say: "This model is licensed under the Qwen License Agreement. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-llm-qwen2-5-32b-instruct`	Apache 2.0	https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-llm-qwen2-5-14b-instruct`	Apache 2.0	https://huggingface.co/Qwen/Qwen2.5-14B-Instruct/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-llm-qwen2-5-7b-instruct`	Apache 2.0	https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-llama-70b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-qwen-32b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-qwen-14b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-llama-8b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-qwen-7b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`deepseek-llm-r1-distill-qwen-1-5b`	MIT	https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/blob/main/LICENSE	Say: "This model is licensed under MIT. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`openai-reasoning-gpt-oss-120b`	Apache 2.0	https://huggingface.co/openai/gpt-oss-120b/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`openai-reasoning-gpt-oss-20b`	Apache 2.0	https://huggingface.co/openai/gpt-oss-20b/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`meta-textgeneration-llama-3-3-70b-instruct`	Llama 3.3 Community	https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct/blob/main/LICENSE	Say: "This model is licensed under Llama 3.3 Community License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`meta-textgeneration-llama-3-2-3b-instruct`	Llama 3.2 Community	https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct/blob/main/LICENSE.txt	Say: "This model is licensed under Llama 3.2 Community License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`meta-textgeneration-llama-3-2-1b-instruct`	Llama 3.2 Community	https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct/blob/main/LICENSE.txt	Say: "This model is licensed under Llama 3.2 Community License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`meta-textgeneration-llama-3-1-8b-instruct`	Llama 3.1 Community	https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/blob/main/LICENSE	Say: "This model is licensed under Llama 3.1 Community License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`nova-textgeneration-pro`	AWS Service Terms	https://aws.amazon.com/service-terms/	Say: "This model is subject to the AWS Service Terms: {URL}. Would you like to proceed?"
`nova-textgeneration-micro`	AWS Service Terms	https://aws.amazon.com/service-terms/	Say: "This model is subject to the AWS Service Terms: {URL}. Would you like to proceed?"
`nova-textgeneration-lite`	AWS Service Terms	https://aws.amazon.com/service-terms/	Say: "This model is subject to the AWS Service Terms: {URL}. Would you like to proceed?"
`nova-textgeneration-lite-v2`	AWS Service Terms	https://aws.amazon.com/service-terms/	Say: "This model is subject to the AWS Service Terms: {URL}. Would you like to proceed?"
`huggingface-reasoning-nvidia-nemotron-3-nano-30b-a3b-bf16`	NVIDIA Open Model License	https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/main/LICENSE	Say: "This model is licensed under the NVIDIA Open Model License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-vlm-qwen3-6-27b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-VL-27B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-vlm-qwen3-5-27b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-VL-27B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-vlm-qwen3-5-9b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-VL-9B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-vlm-qwen3-5-4b`	Apache 2.0	https://huggingface.co/Qwen/Qwen3-VL-4B/blob/main/LICENSE	Say: "This model is licensed under Apache 2.0. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`huggingface-vlm-gemma-4-31b-it`	Gemma Terms of Use	https://huggingface.co/google/gemma-4-31b-it/blob/main/LICENSE	Say: "This model is licensed under Gemma Terms of Use. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."
`meta-vlm-llama-4-scout-17b-16e-instruct`	Llama 4 Community	https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct/blob/main/LICENSE	Say: "This model is licensed under Llama 4 Community License. Please review the license terms here: {URL}. Say 'yes' to accept and proceed."

Related skills

Microsoft FoundryDeploy, evaluate, and continuously improve Microsoft Foundry agents from a single agent interface.490k1.3k

Ai Research ReproductionOrchestrate trustworthy, auditable reproduction of deep learning repositories directly from their READMEs.176k511

Run TrainSafely execute selected deep learning training commands with standardized evidence capture.176k511

Explore RunSafely run isolated exploratory experiments with clear recording and conservative selection before committing changes.176k511

Paper Context ResolverFetch precise reproduction-critical details like dataset splits, preprocessing steps, or evaluation protocols from the original academic paper when the repo README leav141k511

Repo Intake And PlanScan unfamiliar AI research repositories and receive a minimal, trustworthy reproduction target before investing significant time.140k511

FAQ

Which models can it deploy?

Only Nova and OSS models LoRA fine-tuned through SageMaker Serverless Model Customization; base models and full fine-tuned (FFT) models are not supported.

What deployment targets are available?

SageMaker endpoints and Bedrock; the skill presents both and lets the user choose without recommending one.

Data Science & MLllmautomation