flan-t5-finetune-v3

This model is a fine-tuned version of google/flan-t5-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6234
• Rouge1: 0.3289
• Rouge2: 0.1822
• Rougel: 0.3016
• Rougelsum: 0.3018
• Gen Len: 19.8627

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-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_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_steps: 500
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
1.571	0.1739	500	1.0387	0.3684	0.2174	0.337	0.3372	19.727
1.0043	0.3478	1000	0.7315	0.3288	0.1686	0.3007	0.3009	19.5687
0.9122	0.5217	1500	0.6687	0.3276	0.1692	0.2977	0.2979	19.7673
0.8667	0.6956	2000	0.6440	0.3331	0.1795	0.3041	0.3044	19.8282
0.8554	0.8695	2500	0.6328	0.3276	0.1768	0.2997	0.3001	19.8418
0.8431	1.0431	3000	0.6282	0.3334	0.1832	0.3054	0.3057	19.8683
0.8288	1.2170	3500	0.6260	0.3304	0.1821	0.3027	0.303	19.855
0.841	1.3909	4000	0.6243	0.3301	0.1825	0.3026	0.3028	19.8639
0.8423	1.5648	4500	0.6239	0.329	0.1821	0.3021	0.3023	19.8687
0.835	1.7387	5000	0.6235	0.328	0.181	0.3007	0.3011	19.8664
0.8265	1.9126	5500	0.6236	0.3286	0.1822	0.3019	0.3021	19.8621
0.8452	2.0863	6000	0.6235	0.3295	0.1829	0.3023	0.3026	19.865
0.8369	2.2602	6500	0.6235	0.3287	0.182	0.3016	0.3019	19.8674
0.8481	2.4340	7000	0.6235	0.3295	0.1826	0.3024	0.3027	19.8648
0.835	2.6079	7500	0.6234	0.3287	0.1818	0.3014	0.3017	19.865
0.8278	2.7818	8000	0.6234	0.3285	0.1816	0.3015	0.3017	19.8664
0.8257	2.9557	8500	0.6234	0.3289	0.1822	0.3016	0.3018	19.8627

Framework versions

• Transformers 4.55.4
• Pytorch 2.6.0+cu124
• Datasets 2.18.0
• Tokenizers 0.21.2