pythia-31m-simplewiki-2048

This was initialized from random weights based on the config of EleutherAI/pythia-31m and trained on pszemraj/simple_wikipedia_LM for 3 epochs.

It achieves the following results on the evaluation set:

• Loss: 3.6874
• Accuracy: 0.4105

Model description

More information needed

Intended uses & limitations

This is a baseline for comparison to other models.

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 1
• eval_batch_size: 1
• seed: 80085
• gradient_accumulation_steps: 64
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-07
• lr_scheduler_type: inverse_sqrt
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
6.0657	0.22	100	5.6210	0.2414
5.2447	0.45	200	4.9316	0.3054
4.8397	0.67	300	4.6011	0.3343
4.7933	0.9	400	4.3878	0.3530
4.274	1.12	500	4.2352	0.3646
4.4867	1.35	600	4.1224	0.3723
4.3434	1.57	700	4.0282	0.3791
4.1857	1.8	800	3.9552	0.3841
4.229	2.02	900	3.8890	0.3909
3.9189	2.25	1000	3.8301	0.3967
4.084	2.47	1100	3.7782	0.4023
3.8965	2.7	1200	3.7302	0.4069
3.915	2.92	1300	3.6874	0.4105

Framework versions

• Transformers 4.33.1
• Pytorch 2.2.0.dev20230907+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	24.35
ARC (25-shot)	22.18
HellaSwag (10-shot)	25.55
MMLU (5-shot)	23.12
TruthfulQA (0-shot)	49.37
Winogrande (5-shot)	49.41
GSM8K (5-shot)	0.0
DROP (3-shot)	0.81