library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	W-61/llama-3-8b-base-sft-hh-helpful-4xh200	alignment-handbook margin-dpo generated_from_trainer	Anthropic/hh-rlhf	name results llama-3-8b-base-margin-dpo-hh-helpful-4xH200-batch-64

llama-3-8b-base-margin-dpo-hh-helpful-4xH200-batch-64

This model is a fine-tuned version of W-61/llama-3-8b-base-sft-hh-helpful-4xh200 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

• Loss: 0.4046
• Margin Dpo/beta: 0.1000
• Margin Dpo/loss Margin Mean: 21.7563
• Margin Dpo/beta Margin Mean: 2.1756
• Margin Dpo/beta Margin Grad Mean: -0.2570
• Margin Dpo/beta Margin Grad Std: 0.2538
• Margin Dpo/margin Mean: 21.7563
• Margin Dpo/margin Std: 26.3378
• Logps/chosen: -105.9372
• Logps/rejected: -135.4405
• Logps/ref Chosen: -79.0510
• Logps/ref Rejected: -86.7979
• Logits/chosen: -0.6270
• Logits/rejected: -0.6013

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: 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: 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	Margin Dpo/beta	Margin Dpo/loss Margin Mean	Margin Dpo/beta Margin Mean	Margin Dpo/beta Margin Grad Mean	Margin Dpo/beta Margin Grad Std	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
0.9037	0.1468	100	0.5593	0.1000	8.4400	0.8440	-0.3668	0.2303	8.4400	15.3426	-87.1427	-103.3296	-79.0510	-86.7979	-0.6628	-0.6366
0.6607	0.2937	200	0.4791	0.1000	14.6826	1.4683	-0.3109	0.2473	14.6826	21.1628	-92.9979	-115.4274	-79.0510	-86.7979	-0.6426	-0.6197
0.699	0.4405	300	0.4414	0.1000	18.1032	1.8103	-0.2828	0.2516	18.1032	23.7825	-99.9692	-125.8193	-79.0510	-86.7979	-0.6107	-0.5845
0.4468	0.5874	400	0.4213	0.1000	20.2783	2.0278	-0.2687	0.2540	20.2783	25.4582	-102.0468	-130.0720	-79.0510	-86.7979	-0.5647	-0.5335
0.38	0.7342	500	0.4098	0.1000	21.8238	2.1824	-0.2579	0.2561	21.8238	26.5974	-106.9358	-136.5065	-79.0510	-86.7979	-0.6236	-0.5976
0.4876	0.8811	600	0.4046	0.1000	21.7563	2.1756	-0.2570	0.2538	21.7563	26.3378	-105.9372	-135.4405	-79.0510	-86.7979	-0.6270	-0.6013

Framework versions

• Transformers 4.51.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.21.4