LLamaRecipes/getting-started/finetuning/vision/README.md

Fine-tuning Llama 3.2 11B Vision for Structured Data Extraction

This recipe demonstrates how to fine-tune Llama 3.2 11B Vision model on a synthetic W-2 tax form dataset for structured data extraction. The tutorial compares LoRA (Low-Rank Adaptation) and full parameter fine-tuning approaches, evaluating their trade-offs in terms of accuracy, memory consumption, and computational requirements.

Objectives

• Showcase how to fine-tune and evaluate on a specific extraction use case
• Demonstrate custom benchmarking for structured output tasks
• Compare trade-offs between LoRA and Full Parameter Fine-tuning on both task-specific and general benchmarks
• Provide guidance on data preparation, training configuration, and evaluation methodologies

Results

Task-Specific Performance (W2 Extraction)

Benchmark	11B bf16 (Baseline)	LoRA	FPFT	FPFT int4
W2 extraction acc	58	72	97	96

General Benchmark Performance (llama-verifications)

Benchmark	11B bf16 (Baseline)	LoRA	FPFT	FPFT int4
bfclv3	39.87	39.87	39.85	34.67
docvqa	86.88	85.08	86.3	78.95
gpqa-cot-diamond	27.78	27.78	26	28
ifeval	74.79	74.78	74.54	74.42
mmlu-pro-cot	48.43	48.13	48.33	46.14

LM Evaluation Harness Results

Benchmark	11B bf16 (Baseline)	FPFT
gsm8k_cot_llama_strict	85.29	85.29
gsm8k_cot_llama_flexible	85.44	85.44
chartqa llama exact	0	0
chartqa llama relaxed	34.16	35.58
chartqa llama anywhere	43.53	46.52

Prerequisites

• CUDA-compatible GPU with at least 40GB VRAM
• HuggingFace account with access to Llama models
• Python 3.10+

Setup

Environment Creation

git clone git@github.com:meta-llama/llama-cookbook.git
cd llama-cookbook/getting-started/finetuning/vision
conda create -n image-ft python=3.12 -y
conda activate image-ft

Dependencies Installation

pip install torchtune torch torchvision torchaudio torchao bitsandbytes transformers==4.51.1 accelerate vllm==0.9.2 lm_eval wandb

Install torchtune nightly for the latest vision model support:

pip install --pre --upgrade torchtune --extra-index-url https://download.pytorch.org/whl/nightly/cpu

Important: Log in to your HuggingFace account to download the model and datasets:

huggingface-cli login

Dataset Preparation

The dataset contains 2,000 examples of synthetic W-2 forms with three splits: train (1,800), test (100), and validation (100). For this use case, we found that fewer training examples (30% train, 70% test) provided sufficient improvement while allowing for more comprehensive evaluation.

The preparation script:

1. Reshuffles the train/test splits according to the specified ratio
2. Removes unnecessary JSON structure wrappers from ground truth
3. Adds standardized prompts for training

python prepare_w2_dataset.py --train-ratio 0.3

This creates a new dataset directory: fake_w2_us_tax_form_dataset_train30_test70

Configuration Note: If you change the train ratio, update the dataset.data_files.train path in the corresponding YAML configuration files.

Model Download

Download the base Llama 3.2 11B Vision model:

tune download meta-llama/Llama-3.2-11B-Vision-Instruct --output-dir Llama-3.2-11B-Vision-Instruct

This downloads to the expected directory structure used in the provided YAML files. If you change the directory, update these keys in the configuration files:

• checkpointer.checkpoint_dir
• tokenizer.path

Baseline Evaluation

Before fine-tuning, establish a baseline by evaluating the pre-trained model on the test set.

Start vLLM Server

For single GPU (H100):

CUDA_VISIBLE_DEVICES="0" python -m vllm.entrypoints.openai.api_server --model Llama-3.2-11B-Vision-Instruct/ --port 8001 --max-model-len 65000 --max-num-seqs 10

For multi-GPU setup:

CUDA_VISIBLE_DEVICES="0,1" python -m vllm.entrypoints.openai.api_server --model Llama-3.2-11B-Vision-Instruct/ --port 8001 --max-model-len 65000 --tensor-parallel-size 2 --max-num-seqs 10

Run Baseline Evaluation

python evaluate.py --server_url http://localhost:8001/v1 --model Llama-3.2-11B-Vision-Instruct/ --structured --dataset fake_w2_us_tax_form_dataset_train30_test70/test --limit 100 --max_workers 25

Fine-tuning

Configuration Overview

The repository includes two pre-configured YAML files:

• 11B_full_w2.yaml: Full parameter fine-tuning configuration
• 11B_lora_w2.yaml: LoRA fine-tuning configuration

Key differences:

Full Parameter Fine-tuning:

• Trains encoder and fusion layers, leaving decoder frozen
• Higher memory requirements but potentially better performance
• Learning rate: 2e-5
• Optimizer: PagedAdamW8bit for memory efficiency

LoRA Fine-tuning:

• Only trains low-rank adapters across enconder and fusion layers only as well. Decoder is frozen.
• Significantly lower memory requirements
• Learning rate: 1e-4
• LoRA rank: 8, alpha: 16
• Frozen decoder with LoRA on encoder and fusion layers

WandB Configuration

Before training, update the WandB entity in your YAML files:

metric_logger:
  _component_: torchtune.training.metric_logging.WandBLogger
  project: llama3_2_w2_extraction
  entity: <your_wandb_entity>  # Update this

Training Commands

Full Parameter Fine-tuning:

tune run full_finetune_single_device --config 11B_full_w2.yaml

LoRA Fine-tuning:

tune run lora_finetune_single_device --config 11B_lora_w2.yaml

Note: The VQA dataset component in torchtune is pre-configured to handle the multimodal format, eliminating the need for custom preprocessors.

Model Evaluation

Local Evaluation Setup

Start a vLLM server with your fine-tuned model:

For LoRA model:

CUDA_VISIBLE_DEVICES="0,1" python -m vllm.entrypoints.openai.api_server --model ./outputs/Llama-3.2-11B-Instruct-w2-lora/epoch_4/ --port 8001 --max-model-len 65000 --tensor-parallel-size 2 --max-num-seqs 10

For full fine-tuned model:

CUDA_VISIBLE_DEVICES="0,1" python -m vllm.entrypoints.openai.api_server --model ./outputs/Llama-3.2-11B-Instruct-w2-full/epoch_4/ --port 8001 --max-model-len 65000 --tensor-parallel-size 2

Task-Specific Evaluation

python evaluate.py --server_url http://localhost:8001/v1 --model ./outputs/Llama-3.2-11B-Instruct-w2-full/epoch_4/ --structured --dataset fake_w2_us_tax_form_dataset_train30_test70/test --limit 100 --max_workers 25

Cloud Evaluation Setup

To evaluate bigger models, like Llama 4 Maverick, we leverage cloud providers that server these models, like together.ai. Any OpenAI compatible provider should work out of the box with our evaluation script, as it uses the OpenAI SDK.

TOGETHER_API_KEY=<your_api_key> python evaluate.py --server_url https://api.together.xyz/v1 --model meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8  --structured --dataset fake_w2_us_tax_form_dataset_train30_test70/test --limit 100 --max_workers 25

General Benchmark Evaluation

Install llama-verifications for standard benchmarks:

pip install llama-verifications

Run benchmark evaluation:

uvx llama-verifications run-benchmarks \
    --benchmarks mmlu-pro-cot,gpqa-cot-diamond,bfclv3,docvqa \
    --provider http://localhost:8001/v1 \
    --model <model_served_name> \
    --continue-on-failure \
    --max-parallel-generations 100

LM Evaluation Harness

For additional benchmarks using lm-eval:

With vLLM backend:

CUDA_VISIBLE_DEVICES=0,1 lm_eval --model vllm \
    --model_args pretrained=<model_path>,tensor_parallel_size=2,dtype=auto,gpu_memory_utilization=0.9 \
    --tasks gsm8k_cot_llama \
    --batch_size auto \
    --seed 4242

With transformers backend:

CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch -m lm_eval --model hf-multimodal \
    --model_args pretrained=<model_path> \
    --tasks chartqa_llama_90 \
    --batch_size 16 \
    --seed 4242 \
    --log_samples \
    --output_path results

Key Findings

Task-Specific Performance

• Full Parameter Fine-tuning achieved the best task-specific performance (97% accuracy on W2 extraction)
• LoRA fine-tuning provided substantial improvement (72% vs 58% baseline) with significantly lower resource requirements
• Both approaches showed dramatic improvement over the baseline for the specific task

General Capability Preservation

• LoRA fine-tuning preserved general capabilities better, showing minimal degradation on standard benchmarks
• Full Parameter fine-tuning showed minimal degradation on industry benchmarks, making it the preferred choice for this small dataset and FT results. With a larger dataset, as the original split of 80/10/10 and more steps, we do see additional degradation on the benchmarks.
• Both methods maintained strong performance on mathematical reasoning tasks (gsm8k)

Resource Efficiency

• LoRA requires significantly less GPU memory and training time
• Full Parameter fine-tuning requires more resources but achieves better task-specific performance

Performance Graphs

Performance

Peak memory

Loss curve graph

Comparison with Llama API

You can benchmark against the Llama API for comparison:

LLAMA_API_KEY="<your_api_key>" python evaluate.py \
  --server_url https://api.llama.com/compa/v1 \
  --limit 100 \
  --model Llama-4-Maverick-17B-128E-Instruct-FP8 \
  --structured \
  --max_workers 50 \
  --dataset fake_w2_us_tax_form_dataset_train30_test70/test

Best Practices

1. Data Preparation: Ensure your dataset format matches the expected structure. The preparation script handles common formatting issues.

2. Configuration Management: Always update paths in YAML files when changing directory structures.

3. Memory Management: Use PagedAdamW8bit optimizer for full parameter fine-tuning to reduce memory usage.

4. Evaluation Strategy: Evaluate both task-specific and general capabilities to understand trade-offs.

5. Monitoring: Use WandB for comprehensive training monitoring and comparison.

Troubleshooting

Common Issues

1. CUDA Out of Memory: Reduce batch size, enable gradient checkpointing, or use LoRA instead of full fine-tuning.

2. Dataset Path Errors: Verify that dataset paths in YAML files match your actual directory structure.

3. Model Download Issues: Ensure you're logged into HuggingFace and have access to Llama models.

4. vLLM Server Connection: Check that the server is running and accessible on the specified port.

References

• Torchtune Documentation
• vLLM Documentation
• LM Evaluation Harness
• Synthetic Data Kit