wjbmattingly/lfm2-vl-medieval-page

model_step_15000

Model Description

This model is a fine-tuned version of LiquidAI/LFM2-VL-450M using the brute-force-training package.

• Base Model: LiquidAI/LFM2-VL-450M
• Training Status: 🔄 In Progress
• Generated: 2025-08-18 23:13:09
• Training Steps: 15,000

Training Details

Dataset

• Dataset: wjbmattingly/medieval-synthetic-dataset
• Training Examples: 11,000
• Validation Examples: 99

Training Configuration

• Max Steps: 50,000
• Batch Size: 2
• Learning Rate: 1e-05
• Gradient Accumulation: 1 steps
• Evaluation Frequency: Every 5,000 steps

Current Performance

• Training Loss: 0.910276
• Evaluation Loss: 0.854880

Pre-Training Evaluation

Initial Model Performance (before training):

• Loss: 1.175152
• Perplexity: 3.24
• Character Accuracy: 13.2%
• Word Accuracy: 5.0%

Evaluation History

All Checkpoint Evaluations

Step	Checkpoint Type	Loss	Perplexity	Char Acc	Word Acc	Improvement vs Pre
Pre	pre_training	1.1752	3.24	13.2%	5.0%	+0.0%
5,000	checkpoint	0.8849	2.42	9.4%	4.4%	+24.7%
10,000	checkpoint	0.8629	2.37	9.4%	4.8%	+26.6%
15,000	checkpoint	0.8549	2.35	9.9%	4.9%	+27.3%

Training Progress

Recent Training Steps (Loss Only)

Step	Training Loss	Timestamp
14,991	0.975032	2025-08-18T23:12
14,992	0.670720	2025-08-18T23:12
14,993	0.850654	2025-08-18T23:12
14,994	0.935257	2025-08-18T23:12
14,995	0.870635	2025-08-18T23:12
14,996	0.942344	2025-08-18T23:12
14,997	0.785241	2025-08-18T23:12
14,998	0.754749	2025-08-18T23:12
14,999	0.950578	2025-08-18T23:12
15,000	0.910276	2025-08-18T23:12

Training Visualizations

Training Progress and Evaluation Metrics

This chart shows the training loss progression, character accuracy, word accuracy, and perplexity over time. Red dots indicate evaluation checkpoints.

Evaluation Comparison Across All Checkpoints

Comprehensive comparison of all evaluation metrics across training checkpoints. Red=Pre-training, Blue=Checkpoints, Green=Final.

Available Visualization Files:

• training_curves.png - 4-panel view: Training loss with eval points, Character accuracy, Word accuracy, Perplexity
• evaluation_comparison.png - 4-panel comparison: Loss, Character accuracy, Word accuracy, Perplexity across all checkpoints

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
# For vision-language models, use appropriate imports

model = AutoModelForCausalLM.from_pretrained("./model_step_15000")
tokenizer = AutoTokenizer.from_pretrained("./model_step_15000")

# Your inference code here

Training Configuration

{
  "dataset_name": "wjbmattingly/medieval-synthetic-dataset",
  "model_name": "LiquidAI/LFM2-VL-450M",
  "max_steps": 50000,
  "eval_steps": 5000,
  "num_accumulation_steps": 1,
  "learning_rate": 1e-05,
  "train_batch_size": 2,
  "val_batch_size": 2,
  "train_select_start": 0,
  "train_select_end": 11000,
  "val_select_start": 11001,
  "val_select_end": 11100,
  "train_field": "train",
  "val_field": "train",
  "image_column": "image",
  "text_column": "text",
  "user_text": "Transcribe this medieval manuscript page.",
  "max_image_size": 200
}

Model Card Metadata

• Base Model: LiquidAI/LFM2-VL-450M
• Training Framework: brute-force-training
• Training Type: Fine-tuning
• License: Inherited from base model
• Language: Inherited from base model

---

This model card was automatically generated by brute-force-training on 2025-08-18 23:13:09