RichardErkhov/pszemraj_-_pythia-31m-KI_v1-2048-scratch-4bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

pythia-31m-KI_v1-2048-scratch - bnb 4bits

• Model creator: https://huggingface.co/pszemraj/
• Original model: https://huggingface.co/pszemraj/pythia-31m-KI_v1-2048-scratch/

Original model description:

tags: - generated_from_trainer metrics: - accuracy inference: parameters: max_new_tokens: 64 do_sample: true repetition_penalty: 1.1 no_repeat_ngram_size: 5 guidance_scale: 1.01 eta_cutoff: 0.001 widget: - text: My name is El Microondas the Wise and example_title: El Microondas - text: A meme is example_title: meme - text: >- Barack Obama nominated Hilary Clinton as his secretary of state on Monday. He chose her because she had example_title: Coreference resolution - text: >- On a shelf, there are five books: a gray book, a red book, a purple book, a blue book, and a black book example_title: Logic puzzles - text: >- The two men running to become New York City's next mayor will face off in their first debate Wednesday night example_title: Reading comprehension pipeline_tag: text-generation license: apache-2.0 language: - en

pythia-31m-KI_v1-2048-scratch

Initialized from random weights based on config of EleutherAI/pythia-31m, 3 epochs bf16 It achieves the following results on the evaluation set:

• Loss: 4.6160
• Accuracy: 0.2448

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
• train_batch_size: 2
• eval_batch_size: 2
• seed: 80085
• gradient_accumulation_steps: 64
• total_train_batch_size: 128
• 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
6.3874	0.16	100	6.4212	0.1487
5.7088	0.32	200	5.7926	0.1725
5.4575	0.48	300	5.5160	0.1903
5.2451	0.64	400	5.3429	0.1995
5.0954	0.8	500	5.2109	0.2059
5.0358	0.96	600	5.1068	0.2123
4.94	1.12	700	5.0321	0.2157
4.8532	1.28	800	4.9605	0.2202
4.7602	1.44	900	4.9047	0.224
4.6965	1.6	1000	4.8526	0.2276
4.6855	1.76	1100	4.8139	0.2300
4.6573	1.91	1200	4.7739	0.2327
4.5968	2.07	1300	4.7451	0.2346
4.5688	2.23	1400	4.7152	0.2370
4.5205	2.39	1500	4.6842	0.2396
4.5369	2.55	1600	4.6598	0.2410
4.5106	2.71	1700	4.6352	0.2433
4.4375	2.87	1800	4.6160	0.2448

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	25.21
ARC (25-shot)	23.12
HellaSwag (10-shot)	25.23
MMLU (5-shot)	23.12
TruthfulQA (0-shot)	51.67
Winogrande (5-shot)	51.78
GSM8K (5-shot)	0.0
DROP (3-shot)	1.52