17 lines
580 B
Bash
17 lines
580 B
Bash
#!/bin/sh
|
|
set -eu
|
|
export CUDA_VISIBLE_DEVICES=0
|
|
export PROJ_BASE="../"
|
|
python _tuning.fix.modified.py \
|
|
--model_path "../../models/Qwen3-8B/" \
|
|
--data_path "../3-1_model_training_preprocess/model_training/crafted_instruction_data_tri_injection_qa_tool_simple_l.jsonl" \
|
|
--head_path "../2-2_head_identification/head_scoring/qwen3-8b_sep/heads_sorted/all_roc_inst_0.1.json" \
|
|
--output_dir "../3-2_model_training/lora/qwen3-8b_sep_tool_simple" \
|
|
--topk "21.875p" \
|
|
--epochs "30" \
|
|
--batch_size "8" \
|
|
--batch-save-interval "100" \
|
|
--lr "1e-3"
|
|
|
|
|
|
# --topk "21.875p" \ |