DeepSeek/ESFT
1
0
Fork 0
mirror of https://github.com/deepseek-ai/ESFT synced 2025-06-26 18:15:50 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

streamline code; add intermediate saving support for ep

Browse Source
This commit is contained in:
ZihanWang314
2025-05-22 07:21:52 +00:00
parent 98fd21ce21
commit f38f67706c
123 changed files with 710 additions and 5601 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
scripts/eval.sh
View File

@@ -2,17 +2,18 @@
python eval_multigpu.py \
--eval_datasets=translation \
--eval_dataset=intent \
--base_model_path=deepseek-ai/ESFT-vanilla-lite \
--adapter_dir=all_models/adapters/token \
--output_dir=results/completions/token \
--adapter_dir=all_models/adapters/token/intent \
--output_path=results/completions/token/intent.jsonl \
--max_new_tokens=512 \
--openai_api_key=REPLACE_WITH_YOUR_KEY \
--eval_batch_size=2 \
--world_size=4 \
--openai_api_key=REPLACE_WITH_YOUR_KEY \
--gpus_per_rank=2
# this script is used for single-gpu training and has been deprecated. If you have no multiple gpus, you can set above world_size=1 and gpus_per_rank=1
# below script is used for single-gpu training and has been deprecated. If you have only one gpu, you can set above world_size=1 and gpus_per_rank=1
# python scripts/eval.py \
# --eval_datasets=translation \