kmseong/qwen2_5_7b-instruct-gsm8k-rsn-tuned-lr5e-5
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: kmseong/qwen2_5_7b-instruct-gsm8k-rsn-tuned-lr5e-5
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-08-11 13:59:37 +08:00
commit e1a0b5a5cd
17 changed files with 152257 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
finetune_config.json Normal file
View File

@@ -0,0 +1,21 @@
{
"base_model": "/home/edgeai_lab/minseong_results/rsn_tuned_qwen2_5_7b-instruct_lr5e-5_20260715_124936",
"fine_tuning_type": "GSM8K Fine-tuning with Safety Neuron Freezing",
"safety_neurons_file": "/home/edgeai_lab/minseong_results/critical_safety_qwen2_5_7b-instruct/critical_safety_neurons.txt",
"dataset": "GSM8K",
"num_train_samples": 7473,
"batch_size": 4,
"grad_accum": 4,
"learning_rate": 5e-05,
"weight_decay": 0.01,
"warmup_ratio": 0.1,
"epochs": 3,
"max_length": 1024,
"max_grad_norm": 1.0,
"lr_scheduler_type": "cosine",
"optimizer": "adamw_torch",
"gradient_checkpointing": false,
"dtype": "bf16",
"trainer_type": "Trainer",
"strategy": "Freeze safety neurons, train others"
}