metadata

base_model: ondevicellm/tinyllama_mole_v1
tags:
  - trl
  - sft
  - generated_from_trainer
datasets:
  - generator
model-index:
  - name: tinyllama_mole_sft_router05_lr1e-4_ep3
    results: []

tinyllama_mole_sft_router05_lr1e-4_ep3

This model is a fine-tuned version of ondevicellm/tinyllama_mole_v1 on the generator dataset. It achieves the following results on the evaluation set:

Loss: 2.1035

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: 0.0001
train_batch_size: 16
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 2
total_train_batch_size: 128
total_eval_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 120
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
2.2617	0.09	100	2.2410
2.2246	0.18	200	2.2165
2.1994	0.26	300	2.1994
2.1767	0.35	400	2.1869
2.1532	0.44	500	2.1792
2.171	0.53	600	2.1717
2.1588	0.61	700	2.1645
2.145	0.7	800	2.1567
2.1366	0.79	900	2.1507
2.1219	0.88	1000	2.1450
2.1415	0.96	1100	2.1387
1.9765	1.05	1200	2.1446
1.9837	1.14	1300	2.1430
1.9952	1.23	1400	2.1388
1.9868	1.31	1500	2.1351
1.9864	1.4	1600	2.1316
1.987	1.49	1700	2.1263
1.9678	1.58	1800	2.1230
1.9827	1.66	1900	2.1164
1.9846	1.75	2000	2.1134
1.9694	1.84	2100	2.1068
1.9429	1.93	2200	2.1035
1.8079	2.01	2300	2.1369
1.8132	2.1	2400	2.1375
1.8043	2.19	2500	2.1360
1.7927	2.28	2600	2.1334
1.7935	2.37	2700	2.1335
1.7982	2.45	2800	2.1321
1.8029	2.54	2900	2.1311
1.7919	2.63	3000	2.1298
1.7953	2.72	3100	2.1287
1.798	2.8	3200	2.1280
1.7947	2.89	3300	2.1282
1.8015	2.98	3400	2.1283

Framework versions

Transformers 4.37.0
Pytorch 2.1.2+cu118
Datasets 2.16.1
Tokenizers 0.15.0