starchat2-15b-v0.1

This model is a fine-tuned version of bigcode/starcoder2-15b on the HuggingFaceH4/airoboros-3.2, the HuggingFaceH4/Code-Feedback, the HuggingFaceH4/orca-math-word-problems-200k, the HuggingFaceH4/SystemChat and the HuggingFaceH4/capybara datasets. It achieves the following results on the evaluation set:

• Loss: 0.6601

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 32
• total_train_batch_size: 128
• total_eval_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.8404	0.1099	100	0.8309
0.762	0.2198	200	0.7794
0.736	0.3297	300	0.7523
0.6853	0.4396	400	0.7334
0.6926	0.5495	500	0.7201
0.7125	0.6593	600	0.7093
0.6658	0.7692	700	0.7011
0.6515	0.8791	800	0.6940
0.6235	0.9890	900	0.6872
0.5912	1.0989	1000	0.6865
0.585	1.2088	1100	0.6831
0.5855	1.3187	1200	0.6797
0.5732	1.4286	1300	0.6747
0.5907	1.5385	1400	0.6722
0.5689	1.6484	1500	0.6689
0.5753	1.7582	1600	0.6656
0.5447	1.8681	1700	0.6621
0.5504	1.9780	1800	0.6603
0.5146	2.0879	1900	0.6647
0.5232	2.1978	2000	0.6637
0.5237	2.3077	2100	0.6633
0.5138	2.4176	2200	0.6623
0.5085	2.5275	2300	0.6616
0.4996	2.6374	2400	0.6604
0.5026	2.7473	2500	0.6602
0.5144	2.8571	2600	0.6600
0.5292	2.9670	2700	0.6601

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.1+rocm6.2
• Datasets 3.2.0
• Tokenizers 0.20.3