htlou/backup_0202_llamafactory_Compcap_cooccur_0_80-llava-mistral

Compcap_cooccur_0_80

This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the Compcap_cooccur_0_80 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7911

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: 1e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.9418	0.1692	50	0.9316
0.8739	0.3384	100	0.8767
0.8408	0.5076	150	0.8505
0.8415	0.6768	200	0.8333
0.8121	0.8460	250	0.8218
0.7579	1.0152	300	0.8134
0.7469	1.1844	350	0.8087
0.7411	1.3536	400	0.8031
0.7378	1.5228	450	0.7980
0.7423	1.6920	500	0.7938
0.7363	1.8613	550	0.7900
0.6778	2.0305	600	0.7938
0.6716	2.1997	650	0.7936
0.6809	2.3689	700	0.7926
0.6973	2.5381	750	0.7919
0.6804	2.7073	800	0.7912
0.6684	2.8765	850	0.7911

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.20.3