RichardErkhov/kanishka_-_smolm-autoreg-bpe-seed_8128-8bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

smolm-autoreg-bpe-seed_8128 - bnb 8bits

• Model creator: https://huggingface.co/kanishka/
• Original model: https://huggingface.co/kanishka/smolm-autoreg-bpe-seed_8128/

Original model description:

tags: - generated_from_trainer metrics: - accuracy model-index: - name: smolm-autoreg-bpe-seed_8128 results: []

smolm-autoreg-bpe-seed_8128

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.4757
• Accuracy: 0.4994

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: 0.003
• train_batch_size: 16
• eval_batch_size: 128
• seed: 8128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 24000
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.0573	1.0	2928	3.0221	0.4374
2.7148	2.0	5856	2.7910	0.4589
2.5912	3.0	8784	2.6989	0.4683
2.5153	4.0	11712	2.6402	0.4762
2.4585	5.0	14640	2.6094	0.4799
2.4202	6.0	17568	2.5849	0.4829
2.395	7.0	20496	2.5703	0.4845
2.363	8.0	23424	2.5577	0.4859
2.2878	9.0	26352	2.5095	0.4940
2.1407	10.0	29280	2.4757	0.4994

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1