metadata

base_model: JunxiongWang/llama3_0_875_mamba2_sft
tags:
  - alignment-handbook
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
  - HuggingFaceH4/orca_dpo_pairs
  - JunxiongWang/llama3-ultrafeedback-armorm
model-index:
  - name: JunxiongWang/Mamba2InLlama_0_875
    results: []

Please check here for details.

JunxiongWang/Mamba2InLlama_0_875

This model is a fine-tuned version of JunxiongWang/llama3_0_875_mamba2_sft on the HuggingFaceH4/ultrafeedback_binarized, the HuggingFaceH4/orca_dpo_pairs and the JunxiongWang/llama3-ultrafeedback-armorm datasets. It achieves the following results on the evaluation set:

Loss: 0.4761
Rewards/chosen: -1.4040
Rewards/rejected: -2.6012
Rewards/accuracies: 0.7982
Rewards/margins: 1.1973
Logps/rejected: -584.9104
Logps/chosen: -459.0677
Logits/rejected: 0.3408
Logits/chosen: 0.3851

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: 5e-07
train_batch_size: 4
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
total_train_batch_size: 32
total_eval_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5009	0.4798	2000	0.4998	-1.4973	-2.6147	0.7804	1.1175	-586.2582	-468.3976	0.4682	0.5136
0.4895	0.9597	4000	0.4761	-1.4040	-2.6012	0.7982	1.1973	-584.9104	-459.0677	0.3408	0.3851

Framework versions

Transformers 4.43.1
Pytorch 2.1.1+cu118
Datasets 2.20.0
Tokenizers 0.19.1

MambaInLlama

@article{junxiongdaniele2024mambainllama,
  title   = {The Mamba in the Llama: Distilling and Accelerating Hybrid Models},
  author  = {Junxiong Wang and Daniele Paliotta and Avner May and Alexander M. Rush and Tri Dao},
  journal = {arXiv preprint arXiv:2408.15237},
  year    = {2024}
}