Some cleanup stuff

olmocr/train/configs/example_config.yaml

@@ -10,7 +10,7 @@ model:
   trust_remote_code: true
   torch_dtype: auto
   use_flash_attention: true
-  attn_implementation: sdpa
+  attn_implementation: flash_attention_2
   
   # LoRA settings (disabled by default)
   use_lora: false
@@ -65,9 +65,11 @@ training:
   per_device_train_batch_size: 1
   per_device_eval_batch_size: 1
   gradient_accumulation_steps: 8
+
+  gradient_checkpointing: False
   
   # Learning rate
-  learning_rate: 2e-5
+  learning_rate: 1e-6
   lr_scheduler_type: cosine
   warmup_ratio: 0.1
   

olmocr/train/train.py

@@ -162,9 +162,13 @@ def main():
     total_eval_samples = sum(len(dataset) for dataset in eval_datasets.values())
     logger.info(f"Total evaluation samples across {len(eval_datasets)} datasets: {total_eval_samples}")
 
+    # Construct full output directory by appending run_name to base output_dir
+    full_output_dir = os.path.join(config.training.output_dir, config.run_name)
+    logger.info(f"Setting output directory to: {full_output_dir}")
+
     # Set up training arguments
     training_args = TrainingArguments(
-        output_dir=config.training.output_dir,
+        output_dir=full_output_dir,
         num_train_epochs=config.training.num_train_epochs,
         per_device_train_batch_size=config.training.per_device_train_batch_size,
         per_device_eval_batch_size=config.training.per_device_eval_batch_size,

pyproject.toml

@@ -37,7 +37,7 @@ dependencies = [
   "boto3",
   "httpx",
   "torch>=2.7.0",
-  "transformers>=4.51.1",
+  "transformers==4.52.4",
   "img2pdf",
   "beaker-py",
 ]

scripts/train/newtrainer-gantry.sh

@@ -52,7 +52,7 @@ gantry run \
     --priority normal \
     --gpus 1 \
     --preemptible \
-    --cluster "ai2/jupiter-cirrascale-2" \
+    --cluster "ai2/titan-cirrascale" \
     --budget ai2/oe-data \
     --env LOG_FILTER_TYPE=local_rank0_only \
     --env OMP_NUM_THREADS=8 \
@@ -64,4 +64,4 @@ gantry run \
     --weka oe-training-default:/weka/oe-training-default \
     --shared-memory 10GiB \
     --yes \
-    -- /bin/bash -c "source scripts/beaker/jupiter-ib.sh && python -m olmocr.train.train --config olmocr/train/configs/example_config.yaml"
+    -- /bin/bash -c "pip install flash-attn==2.8.0.post2 --no-build-isolation && python -m olmocr.train.train --config olmocr/train/configs/example_config.yaml"