Upload folder using huggingface_hub

README.md

@@ -3,68 +3,75 @@ base_model:
 - LiquidAI/LFM2-VL-450M
 ---
 
-# model_step_300
+# 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**: 🔄 In Progress
-- **Generated**: 2025-08-14 21:03:04
-- **Training Steps**: 300
+- **Training Status**: ✅ Complete
+- **Generated**: 2025-08-18 19:51:23
+- **Training Steps**: 10,000
 
 ## Training Details
 
 ### Dataset
-- **Dataset**: wjbmattingly/medieval-letters-htr-synthetic
+- **Dataset**: CATMuS/medieval
 - **Training Examples**: 148,000
 - **Validation Examples**: 1,999
 
 ### Training Configuration
-- **Max Steps**: 1,000
-- **Batch Size**: 10
-- **Learning Rate**: 1e-05
-- **Gradient Accumulation**: 4 steps
-- **Evaluation Frequency**: Every 100 steps
+- **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**: 4.575191
-- **Evaluation Loss**: 4.492388
+- **Training Loss**: 0.918973
+- **Evaluation Loss**: 3.461486
 
 ## Pre-Training Evaluation
 
 **Initial Model Performance (before training):**
-- **Loss**: 5.419081
-- **Perplexity**: 225.67
-- **Character Accuracy**: 11.4%
-- **Word Accuracy**: 6.9%
+- **Loss**: 6.255835
+- **Perplexity**: 521.04
+- **Character Accuracy**: 27.7%
+- **Word Accuracy**: 11.6%
 
 ## Evaluation History
 
 ### All Checkpoint Evaluations
 | Step | Checkpoint Type | Loss | Perplexity | Char Acc | Word Acc | Improvement vs Pre |
 |------|----------------|------|------------|----------|----------|--------------------|
-| Pre | pre_training | 5.4191 | 225.67 | 11.4% | 6.9% | +0.0% |
-| 100 | checkpoint | 4.8218 | 124.19 | 10.6% | 6.4% | +11.0% |
-| 200 | checkpoint | 4.6127 | 100.76 | 10.8% | 6.4% | +14.9% |
-| 300 | checkpoint | 4.4924 | 89.33 | 10.6% | 6.4% | +17.1% |
+| Pre | pre_training | 6.2558 | 521.04 | 27.7% | 11.6% | +0.0% |
+| 1,000 | checkpoint | 4.4533 | 85.91 | 33.8% | 16.1% | +28.8% |
+| 2,000 | checkpoint | 4.1024 | 60.49 | 32.5% | 15.1% | +34.4% |
+| 3,000 | checkpoint | 3.9043 | 49.62 | 33.2% | 16.9% | +37.6% |
+| 4,000 | checkpoint | 3.7561 | 42.78 | 30.9% | 14.2% | +40.0% |
+| 5,000 | checkpoint | 3.6675 | 39.15 | 33.5% | 17.0% | +41.4% |
+| 6,000 | checkpoint | 3.6180 | 37.26 | 31.8% | 15.1% | +42.2% |
+| 7,000 | checkpoint | 3.5651 | 35.34 | 32.2% | 15.6% | +43.0% |
+| 8,000 | checkpoint | 3.5113 | 33.49 | 30.6% | 14.2% | +43.9% |
+| 9,000 | checkpoint | 3.4908 | 32.81 | 33.4% | 16.9% | +44.2% |
+| 10,000 | final | 3.4615 | 31.86 | 32.5% | 16.5% | +44.7% |
 
 ## Training Progress
 
 ### Recent Training Steps (Loss Only)
 | Step | Training Loss | Timestamp |
 |------|---------------|-----------|
-| 291 | 4.513980 | 2025-08-14T20:59 |
-| 292 | 4.423133 | 2025-08-14T20:59 |
-| 293 | 4.640058 | 2025-08-14T20:59 |
-| 294 | 4.488780 | 2025-08-14T20:59 |
-| 295 | 4.027132 | 2025-08-14T20:59 |
-| 296 | 4.805581 | 2025-08-14T20:59 |
-| 297 | 4.652530 | 2025-08-14T21:00 |
-| 298 | 4.494508 | 2025-08-14T21:00 |
-| 299 | 4.449580 | 2025-08-14T21:00 |
-| 300 | 4.575191 | 2025-08-14T21:00 |
+| 9,991 | 0.764521 | 2025-08-18T19:50 |
+| 9,992 | 0.948460 | 2025-08-18T19:50 |
+| 9,993 | 0.758166 | 2025-08-18T19:50 |
+| 9,994 | 0.898506 | 2025-08-18T19:50 |
+| 9,995 | 0.784889 | 2025-08-18T19:50 |
+| 9,996 | 0.786168 | 2025-08-18T19:50 |
+| 9,997 | 0.674831 | 2025-08-18T19:50 |
+| 9,998 | 0.950868 | 2025-08-18T19:50 |
+| 9,999 | 0.960045 | 2025-08-18T19:50 |
+| 10,000 | 0.918973 | 2025-08-18T19:50 |
 
 ## Training Visualizations
 
@@ -88,8 +95,8 @@ This model is a fine-tuned version of **LiquidAI/LFM2-VL-450M** using the brute-
 from transformers import AutoModelForCausalLM, AutoTokenizer
 # For vision-language models, use appropriate imports
 
-model = AutoModelForCausalLM.from_pretrained("./model_step_300")
-tokenizer = AutoTokenizer.from_pretrained("./model_step_300")
+model = AutoModelForCausalLM.from_pretrained("./final")
+tokenizer = AutoTokenizer.from_pretrained("./final")
 
 # Your inference code here
 ```
@@ -98,23 +105,23 @@ tokenizer = AutoTokenizer.from_pretrained("./model_step_300")
 
 ```json
 {
-  "dataset_name": "wjbmattingly/medieval-letters-htr-synthetic",
+  "dataset_name": "CATMuS/medieval",
   "model_name": "LiquidAI/LFM2-VL-450M",
-  "max_steps": 1000,
-  "eval_steps": 100,
-  "num_accumulation_steps": 4,
-  "learning_rate": 1e-05,
-  "train_batch_size": 10,
-  "val_batch_size": 10,
+  "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": "image",
-  "text_column": "transcription",
-  "user_text": "Transcribe this medieval manuscript line",
+  "image_column": "im",
+  "text_column": "text",
+  "user_text": "Transcribe this medieval manuscript line.",
   "max_image_size": 200
 }
 ```
@@ -129,4 +136,4 @@ tokenizer = AutoTokenizer.from_pretrained("./model_step_300")
 
 ---
 
-*This model card was automatically generated by brute-force-training on 2025-08-14 21:03:04*
+*This model card was automatically generated by brute-force-training on 2025-08-18 19:51:23*

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94846411f2c1171e1ec1a3aff354d161896539bd14085e6bd518449a8c04eaf4
+oid sha256:7d1c7bf9a67a65d0305165d7d04b3a5275ece0b8040925556407941f0b971da8
 size 901692416

model_card_metadata.json

@@ -1,31 +1,31 @@
 {
   "base_model": "LiquidAI/LFM2-VL-450M",
   "training_framework": "brute-force-training",
-  "training_date": "2025-08-14T21:03:04.407697",
-  "training_steps": 300,
-  "dataset": "wjbmattingly/medieval-letters-htr-synthetic",
+  "training_date": "2025-08-18T19:51:23.156386",
+  "training_steps": 10000,
+  "dataset": "CATMuS/medieval",
   "training_config": {
-    "dataset_name": "wjbmattingly/medieval-letters-htr-synthetic",
+    "dataset_name": "CATMuS/medieval",
     "model_name": "LiquidAI/LFM2-VL-450M",
-    "max_steps": 1000,
-    "eval_steps": 100,
-    "num_accumulation_steps": 4,
-    "learning_rate": 1e-05,
-    "train_batch_size": 10,
-    "val_batch_size": 10,
+    "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": "image",
-    "text_column": "transcription",
-    "user_text": "Transcribe this medieval manuscript line",
+    "image_column": "im",
+    "text_column": "text",
+    "user_text": "Transcribe this medieval manuscript line.",
     "max_image_size": 200
   },
-  "final_training_loss": 4.575191020965576,
-  "final_evaluation_loss": 4.492388168970744,
-  "final_char_accuracy": 0.1061965003999563,
-  "final_word_accuracy": 0.06438001131166525
+  "final_training_loss": 0.9189727306365967,
+  "final_evaluation_loss": 3.461486041545868,
+  "final_char_accuracy": 0.3254617025073297,
+  "final_word_accuracy": 0.16456672762721147
 }