stg-cli13b-t6-cdp-ca.mt.him.cln.inter-b4s1e1-20231220-1052

This model is a fine-tuned version of codellama/CodeLlama-13b-Instruct-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0472

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.3435	0.03	100	0.0703
0.0654	0.07	200	0.0586
0.0579	0.1	300	0.0563
0.0567	0.14	400	0.0562
0.0551	0.17	500	0.0547
0.0547	0.21	600	0.0526
0.0532	0.24	700	0.0516
0.0534	0.28	800	0.0515
0.0521	0.31	900	0.0520
0.0522	0.35	1000	0.0517
0.0518	0.38	1100	0.0511
0.051	0.42	1200	0.0502
0.0517	0.45	1300	0.0494
0.0506	0.49	1400	0.0499
0.0511	0.52	1500	0.0496
0.05	0.56	1600	0.0493
0.05	0.59	1700	0.0497
0.049	0.63	1800	0.0485
0.0487	0.66	1900	0.0484
0.0492	0.7	2000	0.0483
0.0493	0.73	2100	0.0481
0.0483	0.77	2200	0.0478
0.048	0.8	2300	0.0478
0.048	0.83	2400	0.0476
0.0476	0.87	2500	0.0474
0.0471	0.9	2600	0.0473
0.0472	0.94	2700	0.0472
0.0469	0.97	2800	0.0472

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: QuantizationMethod.BITS_AND_BYTES
• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: bfloat16

Framework versions

• PEFT 0.6.2