This model is a fine-tuned version of ondevicellm/tinyllama_moe on the HuggingFaceH4/ultrachat_200k and the ondevicellm/SlimOrca datasets. It achieves the following results on the evaluation set:

Loss: 1.1526

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: 2e-05
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: 1

Training results

Training Loss	Epoch	Step	Validation Loss
1.4601	0.05	100	1.3361
1.3324	0.1	200	1.2566
1.2946	0.14	300	1.2279
1.2767	0.19	400	1.2111
1.2298	0.24	500	1.1995
1.2247	0.29	600	1.1902
1.2208	0.34	700	1.1833
1.2375	0.39	800	1.1775
1.2038	0.43	900	1.1726
1.1926	0.48	1000	1.1683
1.1933	0.53	1100	1.1649
1.1893	0.58	1200	1.1618
1.2029	0.63	1300	1.1593
1.2201	0.68	1400	1.1572
1.1741	0.72	1500	1.1557
1.1813	0.77	1600	1.1545
1.1668	0.82	1700	1.1536
1.1495	0.87	1800	1.1530
1.1595	0.92	1900	1.1527
1.1607	0.97	2000	1.1526

Framework versions

Transformers 4.36.2
Pytorch 2.0.1+cu117
Datasets 2.16.1
Tokenizers 0.15.0

ondevicellm/tinyllama_moe_sft_ultrachat-slimorca

tinyllama_moe_sft_ultrachat-slimorca

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

ondevicellm/
tinyllama_moe_sft_ultrachat-slimorca