This model is a fine-tuned version of W-61/llama-3-8b-base-sft-ultrachat-8xh200 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.5470
Fcm Dpo/beta: 0.0074
Margin Dpo/margin Mean: 68.8444
Margin Dpo/margin Std: 101.7556
Logps/chosen: -427.3680
Logps/rejected: -475.3157
Logps/ref Chosen: -287.8268
Logps/ref Rejected: -266.9300
Logits/chosen: -0.8574
Logits/rejected: -0.8453

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: 2
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 8
total_train_batch_size: 128
total_eval_batch_size: 8
optimizer: Use OptimizerNames.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: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Fcm Dpo/beta	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
4.8039	0.4188	200	0.5598	0.0109	45.8351	73.3853	-368.0491	-392.9875	-287.8268	-266.9300	-0.8747	-0.8592
4.4415	0.8377	400	0.5470	0.0074	68.8444	101.7556	-427.3680	-475.3157	-287.8268	-266.9300	-0.8574	-0.8453

Framework versions

Transformers 4.51.0
Pytorch 2.3.1+cu121
Datasets 2.21.0
Tokenizers 0.21.4

W-61/llama3-8b-base-new-method-q_t-0.4-s_star0.65-20260427-043943

llama3-8b-base-new-method-q_t-0.4-s_star0.65-20260427-043943

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

W-61/
llama3-8b-base-new-method-q_t-0.4-s_star0.65-20260427-043943