johnwee1/peft-starcoder-lora-python

peft-starcoder-lora-a100

This model is a fine-tuned version of bigcode/starcoderbase-1b on chargoddard/commitpack-ft-instruct filtered only for Python examples It achieves the following results on the evaluation set:

• Loss: 0.8388

Model description

More information needed

Intended uses & limitations

Intended for merging

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 30
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss
0.8844	0.05	100	0.8664
0.8718	0.1	200	0.8622
0.8754	0.15	300	0.8603
0.8898	0.2	400	0.8581
0.8722	0.25	500	0.8565
0.8592	0.3	600	0.8554
0.8655	0.35	700	0.8537
0.8546	0.4	800	0.8514
0.8776	0.45	900	0.8493
0.852	0.5	1000	0.8477
0.8702	0.55	1100	0.8451
0.8745	0.6	1200	0.8438
0.8613	0.65	1300	0.8422
0.8602	0.7	1400	0.8412
0.8584	0.75	1500	0.8400
0.8455	0.8	1600	0.8398
0.8388	0.85	1700	0.8393
0.8222	0.9	1800	0.8388
0.8413	0.95	1900	0.8389
0.8337	1.0	2000	0.8388

Framework versions

• PEFT 0.11.1
• Transformers 4.41.2
• Pytorch 2.3.1
• Datasets 2.20.0
• Tokenizers 0.19.1