ALFWorld-MPO

This model is a fine-tuned version of Llama-3.1-8B-Instruct on the alfworld-metaplan-preference-pairs dataset as described in MPO: Boosting LLM Agents with Meta Plan Optimization. It achieves the following results on the evaluation set:

• Loss: 0.8390
• Rewards/chosen: -0.5836
• Rewards/rejected: -1.2646
• Rewards/accuracies: 0.6318
• Rewards/margins: 0.6810
• Logps/chosen: -12.9009
• Logps/rejected: -19.8890
• Logits/chosen: -0.3349
• Logits/rejected: -0.3405

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 4
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 3.0

Training results

Framework versions

• Transformers 4.46.1
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3

Code

https://github.com/WeiminXiong/MPO