File size: 733 Bytes
1d13cae |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 |
set -eux
LLM_RECIPES_DIR=/code/llm-recipes
source $LLM_RECIPES_DIR/scripts/wmt2024/tokens.sh
MAX_INPUT_TOKENS=158
BEAM_SIZE=50
python /code/llm-recipes/tools/hf_inference_distrubuted.py \
--model /work/models/additiona_trained_hf/llama2-en-ja-continuous-pretrained-v0-dev-finetune-chunked-docs-all-averaged-841-845 \
-i /work/wmt2024_test/LLM/wmttest2024.src.sentence_splited.with_template.en-ja.en.jsonl \
-o /work/translation/wmt2024_test/en-ja/llama2-beam \
-g 0 1 2 3 4 5 6 7 \
--attn_implementation sdpa \
--dynamic_max_new_token_ratio 3.0 \
--num_return_sequences ${BEAM_SIZE} \
--num_beams ${BEAM_SIZE} \
--max_input_tokens ${MAX_INPUT_TOKENS} \
-b 158
|