Llama-3.1-8B-Base

This model is a fine-tuned version of meta-llama/Llama-3.1-8B on the self_ask_train_data dataset. It achieves the following results on the evaluation set:

Loss: 0.8267

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-06
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 2
total_train_batch_size: 16
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: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.9149	0.0889	100	0.9035
0.84	0.1778	200	0.8476
0.8347	0.2667	300	0.8385
0.827	0.3556	400	0.8327
0.8376	0.4444	500	0.8221
0.7863	0.5333	600	0.8132
0.7957	0.6222	700	0.8041
0.8213	0.7111	800	0.7987
0.8157	0.8	900	0.7951
0.7784	0.8889	1000	0.7914
0.8097	0.9778	1100	0.7857
0.5958	1.0667	1200	0.8021
0.5881	1.1556	1300	0.7966
0.5749	1.2444	1400	0.7988
0.6056	1.3333	1500	0.7950
0.579	1.4222	1600	0.7947
0.6103	1.5111	1700	0.7883
0.6097	1.6	1800	0.7877
0.5635	1.6889	1900	0.7841
0.5749	1.7778	2000	0.7817
0.5661	1.8667	2100	0.7780
0.5856	1.9556	2200	0.7757
0.4189	2.0444	2300	0.8264
0.4021	2.1333	2400	0.8262
0.3829	2.2222	2500	0.8276
0.3857	2.3111	2600	0.8322
0.4063	2.4	2700	0.8273
0.4143	2.4889	2800	0.8291
0.3953	2.5778	2900	0.8304
0.4051	2.6667	3000	0.8251
0.3984	2.7556	3100	0.8273
0.3972	2.8444	3200	0.8260
0.378	2.9333	3300	0.8266