pmahdavi/Llama-3.1-8B-math-reasoning
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: pmahdavi/Llama-3.1-8B-math-reasoning
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-06-12 02:07:16 +08:00
commit a26a31c63d
27 changed files with 5441 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
training_config.yaml Normal file
View File

@@ -0,0 +1,27 @@
bf16: true
cutoff_len: 4096
dataset: tulu3_mixture_math_reasoning
ddp_timeout: 180000000
deepspeed: examples/deepspeed/ds_z3_config.json
do_train: true
eval_strategy: 'no'
finetuning_type: full
gradient_accumulation_steps: 32
learning_rate: 5.0e-06
logging_steps: 10
lr_scheduler_type: cosine
model_name_or_path: meta-llama/Llama-3.1-8B
num_train_epochs: 1.0
output_dir: /scratch/pxm5426/runs/lora-exploration/llama-factory/Llama-3.1-8B_tulu3_mixture_math_reasoning_full_ebs128_lr5e-06
overwrite_cache: true
overwrite_output_dir: true
per_device_train_batch_size: 2
plot_loss: true
preprocessing_num_workers: 12
report_to: wandb
save_steps: 600
save_strategy: steps
stage: sft
template: tulu_v3
trust_remote_code: true
warmup_ratio: 0.03