llama-3-8b-base-kto-ultrafeedback-8xh200

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: 0.3658
• Rewards/chosen: 0.1622
• Logps/chosen: -286.2337
• Rewards/rejected: -2.5444
• Logps/rejected: -292.3963
• Rewards/margins: 2.7066
• Kl: 0.0
• Logits/chosen: -140467840.0
• Logits/rejected: -139209600.0

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• 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	Rewards/chosen	Logps/chosen	Rewards/rejected	Logps/rejected	Rewards/margins	Kl	Logits/chosen	Logits/rejected
1.5841	0.2094	200	0.3971	-0.1699	-289.5548	-1.7146	-284.0978	1.5447	0.0	-151004736.0	-149476768.0
1.404	0.4188	400	0.3773	-0.0342	-288.1983	-2.3874	-290.8255	2.3531	0.0	-143785152.0	-142386976.0
1.4253	0.6283	600	0.3684	-0.3211	-291.0670	-3.1407	-298.3589	2.8196	0.0	-145117536.0	-143700400.0
1.4432	0.8377	800	0.3658	0.1622	-286.2337	-2.5444	-292.3963	2.7066	0.0	-140467840.0	-139209600.0

Framework versions

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