wjbmattingly/lfm2-vl-450m-catmus

final

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: ✅ Complete
• Generated: 2025-08-18 19:08:04
• Training Steps: 10,000

Training Details

Dataset

• Dataset: CATMuS/medieval
• Training Examples: 148,000
• Validation Examples: 1,999

Training Configuration

• Max Steps: 10,000
• Batch Size: 2
• Learning Rate: 5e-06
• Gradient Accumulation: 2 steps
• Evaluation Frequency: Every 1,000 steps

Current Performance

• Training Loss: 1.841697
• Evaluation Loss: 3.094146

Pre-Training Evaluation

Initial Model Performance (before training):

• Loss: 6.023905
• Perplexity: 413.19
• Character Accuracy: 33.1%
• Word Accuracy: 18.0%

Evaluation History

All Checkpoint Evaluations

Step	Checkpoint Type	Loss	Perplexity	Char Acc	Word Acc	Improvement vs Pre
Pre	pre_training	6.0239	413.19	33.1%	18.0%	+0.0%
1,000	checkpoint	4.0098	55.14	26.1%	11.8%	+33.4%
2,000	checkpoint	3.6743	39.42	30.7%	16.3%	+39.0%
3,000	checkpoint	3.4875	32.70	32.1%	16.0%	+42.1%
4,000	checkpoint	3.3974	29.88	35.8%	18.6%	+43.6%
5,000	checkpoint	3.3062	27.28	33.5%	16.7%	+45.1%
6,000	checkpoint	3.2316	25.32	34.9%	17.9%	+46.4%
7,000	checkpoint	3.1867	24.21	34.1%	17.9%	+47.1%
8,000	checkpoint	3.1549	23.45	32.3%	15.8%	+47.6%
9,000	checkpoint	3.1265	22.80	31.6%	16.6%	+48.1%
10,000	final	3.0941	22.07	34.3%	17.6%	+48.6%

Training Progress

Recent Training Steps (Loss Only)

Step	Training Loss	Timestamp
9,991	1.941270	2025-08-18T19:07
9,992	2.647601	2025-08-18T19:07
9,993	3.605345	2025-08-18T19:07
9,994	3.034668	2025-08-18T19:07
9,995	2.445682	2025-08-18T19:07
9,996	3.361138	2025-08-18T19:07
9,997	1.670197	2025-08-18T19:07
9,998	2.518688	2025-08-18T19:07
9,999	2.755938	2025-08-18T19:07
10,000	1.841697	2025-08-18T19:07

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("./final")
tokenizer = AutoTokenizer.from_pretrained("./final")

# Your inference code here

Training Configuration

{
  "dataset_name": "CATMuS/medieval",
  "model_name": "LiquidAI/LFM2-VL-450M",
  "max_steps": 10000,
  "eval_steps": 1000,
  "num_accumulation_steps": 2,
  "learning_rate": 5e-06,
  "train_batch_size": 2,
  "val_batch_size": 2,
  "train_select_start": 0,
  "train_select_end": 148000,
  "val_select_start": 148001,
  "val_select_end": 150000,
  "train_field": "train",
  "val_field": "train",
  "image_column": "im",
  "text_column": "text",
  "user_text": "Transcribe this medieval manuscript line.",
  "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 19:08:04