meng-lab/llama_3.1_8b_instruct_paradec_gsm8k

Llama-3.1-8B-Instruct-sft-5e-3-epoch-100-gsm8k

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the meng-lab/Llama-3.1-8B-Instruct-gsm8k dataset. It achieves the following results on the evaluation set:

• Loss: 6.0386
• Loss Layer 4 Head: 1.7920
• Loss Layer 8 Head: 1.4613
• Loss Layer 12 Head: 1.4113
• Loss Layer 16 Head: 0.7502
• Loss Layer 20 Head: 0.3285
• Loss Layer 24 Head: 0.1760
• Loss Layer 28 Head: 0.0876

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: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• total_eval_batch_size: 8
• 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 4 Head	Loss Layer 8 Head	Loss Layer 12 Head	Loss Layer 16 Head	Loss Layer 20 Head	Loss Layer 24 Head	Loss Layer 28 Head
5.3945	23.6162	200	7.0517	2.0468	1.7131	1.5892	0.8683	0.4093	0.2547	0.1286
3.8702	47.2325	400	6.2998	1.8651	1.5127	1.4674	0.7790	0.3461	0.1892	0.1047
3.146	70.8487	600	6.0631	1.8044	1.4661	1.4136	0.7520	0.3300	0.1769	0.0885
3.0395	94.4649	800	6.0386	1.7920	1.4613	1.4113	0.7502	0.3285	0.1760	0.0876

Framework versions

• Transformers 4.43.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1