ESFT/scripts/eval_expert.sh

19 lines
636 B
Bash

export PYTHONPATH=$PYTHONPATH:$(pwd)
python scripts/expert/get_expert_scores.py \
--eval_dataset=intent \
--base_model_path=deepseek-ai/ESFT-vanilla-lite \
--output_dir=results/expert_scores/intent \
--n_sample_tokens=131072 \
--world_size=4 \
--gpus_per_rank=2
python scripts/expert/generate_expert_config.py \
--eval_dataset=intent \
--expert_scores_dir=results/expert_scores/intent \
--output_path=results/expert_configs/intent.json \
--score_function=token \
--top_p=0.2 # the scoring function and top_p are hyperparameters
# --train_shared_experts
# --train_non_expert_modules