griffin-v0.01-c3t-8layer-simplewiki-silu

• griffin/recurrent_gemma arch
• claude3 tokenizer (as an HF gpt2 tokenizer)

Model description

pretrain experiment on the pszemraj/simple_wikipedia_LM dataset.

It achieves the following results on the evaluation set:

• Loss: 4.0476
• Accuracy: 0.4224

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 4
• eval_batch_size: 4
• seed: 80085
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-07
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
13.3276	0.2548	100	12.0402	0.0131
8.9207	0.5095	200	8.0312	0.0360
7.2681	0.7643	300	6.4775	0.0506
6.3187	1.0190	400	5.6227	0.0434
5.5695	1.2738	500	4.7796	0.3635
5.2926	1.5285	600	4.3923	0.3952
4.878	1.7833	700	4.1877	0.4085

Framework versions

• Transformers 4.40.1
• Pytorch 2.2.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1