hoanganhpham/4b-res-reranker

Model Checkpoint

This checkpoint was created during training with the following details:

• Step: 1050
• Accuracy: 0.9497
• Original Model: Qwen/Qwen3-Reranker-4B
• Training Arguments: {'experiment': 'train', 'model_path': 'Qwen/Qwen3-Reranker-4B', 'max_length': 256, 'learning_rate': 5e-06, 'adam_beta2': 0.95, 'weight_decay': 0.01, 'per_device_train_batch_size': 2, 'per_device_eval_batch_size': 2, 'gradient_checkpointing': True, 'gradient_accumulation_steps': 8, 'datasets_dir': '/vast/share/andrew/code/tablecheck-machine-learning-takehome/data', 'dataset_name': 'text_train', 'valid_name': 'text_validation', 'output_dir': 'ckpts/lm_text_train_Qwen3-Reranker-4B_lr5e-06_bs128_text_validation', 'eval_steps': 50, 'scheduler': 'linear', 'warmup_steps': 50, 'testing': False, 'num_epochs': 20, 'run_name': 'lm_text_train_Qwen3-Reranker-4B_lr5e-06_bs128_text_validation'}

Token IDs

• False token ID (no): 2152
• True token ID (yes): 9693

Usage

This checkpoint can be loaded using transformers:

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ckpts/lm_text_train_Qwen3-Reranker-4B_lr5e-06_bs128_text_validation/checkpoint-step-1050-acc-0.9497", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("ckpts/lm_text_train_Qwen3-Reranker-4B_lr5e-06_bs128_text_validation/checkpoint-step-1050-acc-0.9497", trust_remote_code=True)