This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 1.7260
Rewards/chosen: -5.0723
Rewards/rejected: -8.1169
Rewards/accuracies: 0.3075
Rewards/margins: 3.0445
Logps/rejected: -601.5961
Logps/chosen: -399.7386
Logits/rejected: 4.1594
Logits/chosen: 3.9982

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: 8.19247218325348e-07
train_batch_size: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 2
total_train_batch_size: 64
total_eval_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 2

Training results

Training Loss	Epoch	Step	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
0.66	1.0417	100	2.2227	2.3998	-231.4086	-327.2254	0.6710	0.3155	-0.1518	0.0862	-0.2380

Framework versions

Transformers 4.44.2
Pytorch 2.4.0
Datasets 2.21.0
Tokenizers 0.19.1

RichardErkhov/taicheng_-_zephyr-7b-align-scan-0.0-0.0-cosine-2-8bits

Original model description:

zephyr-7b-align-scan-0.0-0.0-cosine-2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
taicheng_-_zephyr-7b-align-scan-0.0-0.0-cosine-2-8bits