meng-lab/codellama_34b_instruct_paradec_humaneval

CodeLlama-34b-Instruct-sft-5e-3-epoch-100-human-eval-final

This model is a fine-tuned version of meta-llama/CodeLlama-34b-Instruct-hf on the meng-lab/CodeLlama-34B-Instruct-humaneval dataset. It achieves the following results on the evaluation set:

• Loss: 3.7616
• Loss Layer 6 Head: 1.0709
• Loss Layer 12 Head: 0.8047
• Loss Layer 18 Head: 0.7212
• Loss Layer 24 Head: 0.4396
• Loss Layer 30 Head: 0.3042
• Loss Layer 36 Head: 0.2040
• Loss Layer 42 Head: 0.1346

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: 0.005
• train_batch_size: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Loss Layer 6 Head	Loss Layer 12 Head	Loss Layer 18 Head	Loss Layer 24 Head	Loss Layer 30 Head	Loss Layer 36 Head	Loss Layer 42 Head
3.7028	9.1168	200	4.6352	1.4035	0.8311	1.0429	0.4931	0.4457	0.2349	0.1599
2.736	18.2336	400	4.7219	1.2158	0.8316	0.7490	1.0869	0.3238	0.2666	0.1723
2.0128	27.3504	600	3.8953	1.1598	0.8030	0.7230	0.4451	0.3500	0.2027	0.1459
3.3605	36.4672	800	4.9203	1.1038	0.8175	1.6655	0.4410	0.3091	0.2055	0.1365
2.5177	45.5840	1000	4.2388	1.0907	0.8042	1.1115	0.4403	0.3038	0.2217	0.1412
2.0743	54.7009	1200	3.9221	1.0727	0.8050	0.8689	0.4418	0.3012	0.2044	0.1362
1.8844	63.8177	1400	3.8140	1.0723	0.8028	0.7729	0.4389	0.3045	0.2036	0.1350
1.8019	72.9345	1600	3.7777	1.0726	0.8038	0.7376	0.4401	0.3042	0.2032	0.1345
1.7339	82.0513	1800	3.7662	1.0703	0.8056	0.7246	0.4394	0.3041	0.2042	0.1347
1.6981	91.1681	2000	3.7616	1.0709	0.8047	0.7212	0.4396	0.3042	0.2040	0.1346

Framework versions

• Transformers 4.43.2
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1