scripts/decode/en-ja/mistral/top_p_inference_1.sh

set -eux
LLM_RECIPES_DIR=/code/llm-recipes
source $LLM_RECIPES_DIR/scripts/wmt2024/tokens.sh

for i in `seq 0 7`; do
	python /code/llm-recipes/tools/hf_inference.py \
	       --model /work/models/translation_finetuned_hf/mistral-llm-recipes-en-ja-continuous-pretrained-v1-dev-finetune-chunked-docs-all-averaged-71-7 \
	       -i /work/wmt2024_test/LLM/split/en-ja/wmttest2024.src.sentence_splited.with_template.en-ja.en.jsonl.0${i}  \
	       -o /work/translation/wmt24_test/en-ja/mistral-top-p-0.95/split_0${i} \
	       -g ${i} \
	       -b 400 \
	       --attn_implementation sdpa \
	       --dynamic_max_new_token_ratio 3.0 \
	       --num_return_sequences 100 \
	       --do_sample \
	       --top_p 0.95 &
done
wait