Modelpublic

pszemraj/pythia-31m-simplepile-lite-2048-scratch-2e

sourceHugging Faceapache-2.0updated 9mo agoView on Hugging Face

0likes80downloads

Model Card

BL-pythia-31m-simplepile-lite-2048-scratch

Train from scratch based on config of EleutherAI/pythia-31m on the None dataset. It achieves the following results on the evaluation set:

—Loss: 3.9891
—Accuracy: 0.3498

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: 0.0005
—trainbatchsize: 2
—evalbatchsize: 1
—seed: 80085
—gradientaccumulationsteps: 64
—totaltrainbatch_size: 128
—optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-07
—lrschedulertype: inverse_sqrt
—lrschedulerwarmup_ratio: 0.05
—num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
7.4089	0.07	100	7.3885	0.1133
6.2774	0.13	200	6.2091	0.1621
5.7019	0.2	300	5.7450	0.1890
5.4922	0.27	400	5.4697	0.2080
5.233	0.33	500	5.2846	0.2195
5.0523	0.4	600	5.1479	0.2296
4.9396	0.47	700	5.0391	0.2376
4.7633	0.53	800	4.9366	0.2458
4.7516	0.6	900	4.8339	0.2559
4.5937	0.67	1000	4.7286	0.2676
4.5079	0.73	1100	4.6293	0.2798
4.4608	0.8	1200	4.5433	0.2903
4.3426	0.87	1300	4.4719	0.2988
4.1722	0.93	1400	4.4089	0.3057
4.1655	1.0	1500	4.3585	0.3107
4.0927	1.07	1600	4.3101	0.3161
4.1439	1.13	1700	4.2714	0.3206
4.0064	1.2	1800	4.2330	0.3249
4.0633	1.27	1900	4.2015	0.3281
3.9948	1.33	2000	4.1702	0.3311
3.9389	1.4	2100	4.1439	0.3338
3.8833	1.47	2200	4.1200	0.3367
3.8411	1.53	2300	4.0949	0.3395
3.8481	1.6	2400	4.0764	0.3408
3.8397	1.67	2500	4.0578	0.3420
3.8897	1.73	2600	4.0383	0.3440
3.8785	1.8	2700	4.0206	0.3459
3.8126	1.87	2800	4.0044	0.3478
3.783	1.93	2900	3.9891	0.3498

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.	24.7
ARC (25-shot)	21.59
HellaSwag (10-shot)	25.79
MMLU (5-shot)	24.99
TruthfulQA (0-shot)	50.62
Winogrande (5-shot)	48.62
GSM8K (5-shot)	0.0
DROP (3-shot)	1.32