{"slug": "aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization", "title": "AWS SageMaker FFT: Full Fine-Tuning on Serverless Model Customization", "summary": "Amazon SageMaker AI Serverless Model Customization added Full Fine-Tuning (FFT) support alongside LoRA adapters, AWS announced on November 18, 2026, letting builders run full-rank updates through the same serverless job type. In AWS's experiment fine-tuning Qwen3-4B on the 110,000-pair Magicoder-Evol-Instruct-110K coding dataset (99,000 train, 5,500 validation, 5,500 test), FFT reached roughly 90% LLM-as-Judge semantic correctness at 100K+ scale and beat LoRA on aggregate PRIMECODE and large-scale judge scores, though LoRA won one LLM-as-Judge comparison. AWS warns FFT carries higher memory cost and catastrophic-forgetting risk at aggressive learning rates, and the sample code sets a 2e-5 learning rate, global batch size 8, and 2 epochs.", "body_md": "**Amazon SageMaker AI Serverless Model Customization now supports Full Fine-Tuning (FFT)**, not only parameter-efficient LoRA adapters. AWS published the workflow on **[November 18, 2026](https://aws.amazon.com/blogs/machine-learning/full-fine-tuning-fft-on-sagemaker-ai-serverless-model-customization/?ref=explainx)**, arguing FFT can close quality gaps on **large, complex corpora** when LoRA plateaus.\n\nFor builders on **[Bedrock AgentCore](https://explainx.ai/blog/aws-bedrock-agentcore-identity-private-ca-november-2026)** or **[custom model routes](https://explainx.ai/blog/baseten-loops-serverless-sft-rl-training-glm-5-3-qwen3-6-per-token-billing-ga-what-docs-confirm-2026)**, the headline is operational: **one serverless job type** now spans **cheap adapters** and **full-rank updates** without a separate training stack.\n\n## FFT versus LoRA in plain language\n\n**[LoRA](https://arxiv.org/abs/2106.09685?ref=explainx)** freezes base weights and trains small low-rank matrices. It saves GPU memory and trains fast. AWS warns that low-rank updates can **cap representation power** as tasks grow—extra rank does not always catch up to full fine-tuning.\n\n**FFT** updates **all parameters** together. AWS says that preserves layer alignment better on **complex domain shifts**, especially when datasets exceed **tens of thousands** of high-quality examples.\n\n| Method | Memory | Best when | Risk AWS highlights | \n|---|---|---|---|\n| LoRA | Lower | Smaller shifts, budget fine-tunes | Rank plateaus on hard tasks | \n| FFT | Higher | Large curated corpora, code style absorption | Catastrophic forgetting at aggressive LR | \n\nThis is not an either-or religion. AWS's own charts show **LoRA winning** one **LLM-as-Judge** comparison while **FFT wins** aggregate PRIMECODE and large-scale judge scores.\n\n## The November 18 experiment design\n\nAWS ran **supervised fine-tuning** on **Qwen3-4B** using **Magicoder-Evol-Instruct-110K**—** 110,000** coding instruction pairs spanning algorithms, debugging, and systems topics. Split:\n\n- **99,000** train\n- **5,500** validation\n- **5,500** held-out test\n\nJobs ran on **SageMaker AI Serverless Model Customization** with identical prompts and evaluation code for LoRA and FFT branches.\n\n### Code entry point\n\nAWS sample code constructs `SFTTrainer` with `TrainingType.FULL`:\n\n``` python\nfrom sagemaker.train.sft_trainer import SFTTrainer\nfrom sagemaker.train.common import TrainingType\n\ntrainer = SFTTrainer(\n    model=base_model_id,\n    training_type=TrainingType.FULL,\n    model_package_group=model_package_group_name,\n    training_dataset=training_dataset,\n    validation_dataset=val_dataset,\n    s3_output_path=output_path,\n    sagemaker_session=sess,\n    role=role,\n)\n\ntrainer.hyperparameters.learning_rate = 2e-5\ntrainer.hyperparameters.global_batch_size = 8\ntrainer.hyperparameters.max_epochs = 2\ntrainer.hyperparameters.lr_warmup_steps_ratio = 0.1\n\ntrainer.train(wait=False)\n```\n\nSwitching to LoRA is **`TrainingType.LORA`** only—everything else unchanged per AWS.\n\nFine-tuned artifacts register as **versioned Model Packages** in the SageMaker Model Registry when jobs complete.\n\n## Evaluation trio AWS ships serverlessly\n\nLLM-as-Judge semantic correctness FFT versus LoRA at 100K examples\n*Image: [Amazon Web Services](https://aws.amazon.com/blogs/machine-learning/full-fine-tuning-fft-on-sagemaker-ai-serverless-model-customization/?ref=explainx)*\n\nAWS evaluates inside the same serverless product:\n\n1. **LLM-as-Judge** — Amazon**Nova Pro** judges whether generated code is logically correct versus the prompt.\n2. **PRIMECODE** — Deterministic BLEU, ROUGE, and F1 against reference solutions for reproducible code-style metrics.\n3. **MMLU** — Forgetting detector across**57** academic subjects; tracks whether code fine-tuning erases general knowledge.\n\nAt **100K+** scale AWS charts **~90%** LLM-as-Judge semantic correctness for FFT on the coding task setup they describe. Radar charts show FFT extending farther than LoRA across multiple quality axes in the same post.\n\n**Caveat:** Those figures are **AWS-run benchmarks** on a **single public dataset**. Replicate on **your** validation set before production routing.\n\n## Hyperparameters AWS sweeps\n\nLearning rate dataset size risk map for SageMaker full fine-tuning\n*Image: [Amazon Web Services](https://aws.amazon.com/blogs/machine-learning/full-fine-tuning-fft-on-sagemaker-ai-serverless-model-customization/?ref=explainx)*\n\n### Learning rate\n\nFFT touches every weight. AWS maps **learning rate × dataset size** risk: **2e-5** at **100K+** examples landed in their **safe zone**. **≥5e-5** wrecked **MATH exact match** and drove **MMLU** toward **~3.6%** in aggressive configs—classic **catastrophic forgetting**.\n\n### Epochs\n\nAWS treats max epochs as a **forgetting knob**, not a loss-curve game. **~3 epochs** raised judge scores toward **90%**, but **beyond 3** degraded variants they tried—even when validation loss still looked fine on in-distribution code.\n\n### Batch size\n\nSmaller batches add noise akin to higher LR; larger batches smooth updates but reduce steps. AWS includes batch-size trade-off charts in the post.\n\n### Warmup\n\nA configuration with **5% warmup**, **batch size 8**, and **LR 2e-5** best preserved **MMLU** in their forgetting study.\n\n## When AWS says to stay on LoRA\n\nThe post is explicit FFT is **not universal**:\n\n- Both FFT and LoRA **trailed the base model** on**MATH exact match** in some runs.\n- **LoRA at 81.5%** beat**FFT at 79.5%** on one LLM-as-Judge setup AWS lists.\n- Aggressive FFT schedules **destroyed** general benchmarks.\n\nPick FFT when you have **large, clean in-domain data** and monitoring for **MMLU-style drift**. Stay on LoRA when budgets, iteration speed, or subjective judge alignment matter more.\n\n## How this compares to other serverless trainers\n\n| Platform | Full-weight SFT serverless | Notes | \n|---|---|---|\n| **SageMaker (this launch)** | Yes, `TrainingType.FULL` | Bundled judge + PRIMECODE + MMLU | \n| **[Baseten Loops](https://explainx.ai/blog/baseten-loops-serverless-sft-rl-training-glm-5-3-qwen3-6-per-token-billing-ga-what-docs-confirm-2026)** | SFT and RL per token | Different billing model | \n| **[Snowflake Cortex Training](https://explainx.ai/blog/snowflake-cortex-training-preview-serverless-sft-rl-open-weight-models-aws-what-we-know-2026)** | Preview SFT and RL | Warehouse-adjacent | \n\nSageMaker's angle is **enterprise registry + evaluation batteries** in one managed lane—valuable if you already standardize on **[AgentCore identity](https://explainx.ai/blog/aws-bedrock-agentcore-identity-private-ca-november-2026)**.\n\n## Builder checklist for this week\n\n1. **Inventory data size.** Below ~10K curated examples, start LoRA; plan FFT trials when growth toward**100K** is realistic.\n2. **Pin evaluation.** Export baseline MMLU and domain-held-out sets**before** first FFT job.\n3. **Cap LR** at AWS's**2e-5** starting point unless your regression suite proves otherwise.\n4. **Stop at epoch 3** until you see forgetting curves flatten safely.\n5. **Register packages** in Model Registry for rollback when**[coding model scores rise but merge rates fall](https://explainx.ai/blog/higher-scoring-coding-models-swe-bench-merge-rate-natural-experiment-2026)** .\n\n## Spectral view: why AWS says rank limits LoRA\n\nThe AWS post dives into **weight matrix spectra**: LoRA applies a **low-rank delta** that cannot express every direction full fine-tuning can move. As datasets scale, the **missing directions** show up as plateaus in judge scores even when you increase LoRA rank.\n\nFFT modifies matrices **uniformly**, which AWS ties to better **activation alignment across layers**. You do not need to accept the theory wholesale to use the practical signal: when validation loss stalls while train loss falls on LoRA, FFT is the next serverless knob—if your **MMLU guardrail** passes.\n\n## Rapid experimentation story\n\nAWS emphasizes **tens of jobs in a short window** because serverless removes cluster babysitting. They swept:\n\n- Training type (FFT vs LoRA)\n- Learning rate, batch, warmup, epochs\n- Base models: **Qwen3-4B** ,**DeepSeek-R1-1.5B** ,**Nemotron-30B** ,**Gemma-4-26B**\n\nEach job registered outputs in **Model Registry**, so promotion to endpoints stays auditable—important when **[audit agents](https://explainx.ai/blog/kpmg-project-exponential-ai-agents-ftse-250-audit-2026)** ask for lineage.\n\n## PRIMECODE versus LLM-as-Judge\n\n**PRIMECODE** rewards n-gram overlap with reference solutions—great for **detecting formatting conventions** and API usage patterns in code corpora. **LLM-as-Judge** rewards logical correctness from Nova Pro's reading of the prompt plus code.\n\nAWS uses both because neither alone is sufficient: a solution can look like the reference yet be wrong, or be correct yet novel. If you only track loss, you miss the split.\n\n## Operational details builders ask in search\n\n**Do I provision GPUs?** AWS says no—serverless customization allocates compute for the job.\n\n**Where do checkpoints live?** S3 output paths you provide; registered packages surface in Model Registry.\n\n**Can I A/B in production?** Use packages as versions; pair with **[higher merge-rate experiments](https://explainx.ai/blog/higher-scoring-coding-models-swe-bench-merge-rate-natural-experiment-2026)** before routing traffic.\n\n**What about RL after SFT?** This post covers SFT only; combine with your RL stack separately—compare **[MIT agent self-improvement loops](https://explainx.ai/blog/llm-agents-self-improve-without-prescribed-loop-mit-isola-2026)** if you chain RL after FFT.\n\n## Failure modes to log in your runbook\n\n1. **MMLU cliff** after epoch 3 — roll back package, reduce LR, or shorten epochs.\n2. **Judge regression vs base** on subjective tasks — LoRA may be the better tool per AWS's own tie cases.\n3. **MATH exact match collapse** — treat as forgetting alarm, not noise.\n4. **Cost spikes** — FFT uses more compute than LoRA; watch job duration against**[TSMC credit stories](https://explainx.ai/blog/tsmc-credit-limits-upfront-cash-ai-chip-orders-claimed-vs-verified-2026)** only if you finance GPU-hours centrally.\n\n## Enterprise procurement angle\n\nTeams already buying **SageMaker** for compliance get FFT without a separate vendor. That matters when **[Snowflake](https://explainx.ai/blog/snowflake-cortex-training-preview-serverless-sft-rl-open-weight-models-aws-what-we-know-2026)** and **Baseten** compete on per-token economics but your IAM boundary is AWS-only.\n\nDocument **Nova Pro judge costs** in FFT evaluation—they are not free tokens even when training is serverless.\n\n## Step-by-step first FFT job\n\n1. Register **Magicoder** or your corpora in SageMaker dataset assets.\n2. Create a **model package group** for promotion discipline.\n3. Instantiate **SFTTrainer** with`TrainingType.FULL` and conservative LR**2e-5** .\n4. Run **MMLU baseline** on the base package before training.\n5. Train with **max_epochs=2** first; inspect PRIMECODE and judge on held-out code.\n6. Compare against a **LoRA twin job** on identical splits.\n7. Promote only if **MMLU drop** sits inside your risk tolerance.\n\n## Documentation pointers\n\nAWS links hyperparameter docs for Serverless Model Customization on the SageMaker doc site—read those alongside this post for field names that change release to release. The blog date is **November 18, 2026**; if your account lacks `TrainingType.FULL`, your region may still be rolling out—open a support case rather than assuming the blog is wrong.\n\nFor dataset hygiene, Magicoder-Evol-Instruct is a **public** corpus tied to the [Magicoder paper](https://arxiv.org/abs/2312.02120?ref=explainx) on OSS-Instruct and Evol-Instruct data; enterprise teams should mirror the **90/5/5 split** discipline on **proprietary** logs with PII stripped, similar to how **[GitHub and Microsoft m-dash reports](https://explainx.ai/blog/github-microsoft-mdash-9791-ai-findings-99-reports-november-2026)** stress provenance before automation.\n\n## Cost realism\n\nServerless does not mean **free**. FFT updates more parameters per step than LoRA, so expect **longer jobs** and **higher GPU-second bills** even without cluster admin overhead. Model the **inference savings** only **after** you measure quality uplift on your held-out suite—do not assume FFT inference is cheaper; you may deploy a **smaller specialized model** that beats a frontier generalist on your task, which is the economic story AWS tells at the top of its post. Keep a LoRA checkpoint as rollback until FFT wins on **your** judge suite twice in a row.\n\n## Related reading\n\n*Benchmarks and charts come from the AWS Machine Learning Blog post dated November 18, 2026; rerun on your data before production commits. Contact AWS support if `TrainingType.FULL` is missing in your account region.*", "url": "https://wpnews.pro/news/aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization", "canonical_source": "https://explainx.ai/blog/aws-sagemaker-fft-serverless-model-customization-november-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-19 00:00:00+00:00", "updated_at": "2026-11-19 02:13:49.726599+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-infrastructure", "mlops"], "entities": ["Amazon Web Services", "Amazon SageMaker AI Serverless Model Customization", "Qwen3-4B", "Magicoder-Evol-Instruct-110K", "Amazon Nova Pro", "PRIMECODE", "MMLU", "LoRA"], "also_reported_by": [], "takeaway": "SageMaker's SFTTrainer now accepts TrainingType.FULL to run full fine-tuning on serverless jobs, with AWS's Qwen3-4B run using learning rate 2e-5, global batch size 8, and 2 epochs over 99,000 training examples.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization", "markdown": "https://wpnews.pro/news/aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization.md", "text": "https://wpnews.pro/news/aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization.txt", "jsonld": "https://wpnews.pro/news/aws-sagemaker-fft-full-fine-tuning-on-serverless-model-customization.jsonld"}}