This model is a fine-tuned version of meta-llama/Llama-3.1-8B on the tulu3_mixture_coding dataset. It achieves the following results on the evaluation set:

Loss: 0.7736 This checkpoint was released alongside https://arxiv.org/abs/2509.11167.

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: 5e-06
train_batch_size: 2
eval_batch_size: 2
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 32
total_train_batch_size: 128
total_eval_batch_size: 4
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: warmup_stable_decay
num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.8738	0.0909	100	0.8080
0.8557	0.1818	200	0.8015
0.8619	0.2726	300	0.7972
0.8201	0.3635	400	0.7945
0.8609	0.4544	500	0.7920
0.8175	0.5453	600	0.7903
0.8462	0.6361	700	0.7885
0.8307	0.7270	800	0.7850
0.8595	0.8179	900	0.7791
0.8116	0.9088	1000	0.7748
0.8221	0.9996	1100	0.7736

Framework versions

Transformers 4.51.1
Pytorch 2.6.0+cu124
Datasets 3.4.1
Tokenizers 0.21.0

pmahdavi/Llama-3.1-8B-coding-tulu3-ebs128-lr5e6-wsdcr0p4

Llama-3.1-8B_tulu3_mixture_coding_full_adamw_ebs128_lr5e-06_wsd-cr0.4

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

pmahdavi/
Llama-3.1-8B-coding-tulu3-ebs128-lr5e6-wsdcr0p4