TakalaWang/Discussion-Phi-4-multimodal-instruct-audio-dimp-tag-beta2

Discussion-Phi-4-multimodal-instruct-audio-dimp-tag

This model is a fine-tuned version of microsoft/Phi-4-multimodal-instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 33.4176

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: 4e-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.95) and epsilon=1e-07 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1160211.625	0.1117	10	7426.7354
0.6156	0.2235	20	31.8351
12.5708	0.3352	30	38.5527
120.2882	0.4469	40	34.2812
0.5927	0.5587	50	30.9922
2.379	0.6704	60	30.3796
0.6282	0.7821	70	29.5946
0.4296	0.8939	80	28.8131
0.1972	1.0	90	26.5031
0.2999	1.1117	100	29.2309
1.3594	1.2235	110	30.3200
0.5613	1.3352	120	27.7579
0.1148	1.4469	130	30.7001
2.4258	1.5587	140	31.5338
0.2405	1.6704	150	32.9445
0.1763	1.7821	160	33.0411
0.3052	1.8939	170	29.8340
0.0634	2.0	180	30.9146
0.1446	2.1117	190	34.0906
0.1127	2.2235	200	34.1306
2.9534	2.3352	210	31.8096
0.1518	2.4469	220	35.4243
0.148	2.5587	230	33.2750
0.0649	2.6704	240	32.6477
0.0999	2.7821	250	33.6155
0.3312	2.8939	260	33.4176

Framework versions

• Transformers 4.48.2
• Pytorch 2.4.1+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1