README.md

metadata

tags:
  - generated_from_trainer
model-index:
  - name: falcon-7b-ft-self_instruct
    results: []

falcon-7b-ft-self_instruct

This model is a fine-tuned version of ybelkada/falcon-7b-sharded-bf16 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7991

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: 0.0002
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.9935	0.09	100	1.0028
1.0094	0.17	200	0.9614
0.8897	0.26	300	0.9396
0.9435	0.34	400	0.9260
0.9618	0.43	500	0.9146
0.8103	0.52	600	0.9035
0.8889	0.6	700	0.8966
0.8915	0.69	800	0.8884
0.8814	0.77	900	0.8833
0.9207	0.86	1000	0.8742
0.8284	0.95	1100	0.8693
0.8283	1.03	1200	0.8682
0.7445	1.12	1300	0.8659
0.8238	1.2	1400	0.8640
0.7448	1.29	1500	0.8537
0.7866	1.38	1600	0.8534
0.7224	1.46	1700	0.8480
0.7235	1.55	1800	0.8399
0.7936	1.64	1900	0.8385
0.7487	1.72	2000	0.8337
0.7842	1.81	2100	0.8284
0.7198	1.89	2200	0.8251
0.7507	1.98	2300	0.8188
0.622	2.07	2400	0.8353
0.6592	2.15	2500	0.8358
0.6043	2.24	2600	0.8330
0.7259	2.32	2700	0.8324
0.6388	2.41	2800	0.8290
0.7284	2.5	2900	0.8224
0.6166	2.58	3000	0.8202
0.6132	2.67	3100	0.8122
0.6323	2.75	3200	0.8094
0.6686	2.84	3300	0.8034
0.6457	2.93	3400	0.7991

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3