# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
# 
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# 
#     http://www.apache.org/licenses/LICENSE-2.0
# 
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

export PYTHONPATH=../../../:$PYTHONPATH # PaddleNLP dir
export PYTHONPATH=../../:$PYTHONPATH # PaddleNLP/llm dir
export USE_FAST_TOKENIZER=true
export NCCL_ALGO=Tree

# MODEL_PATH="Qwen/Qwen2.5-Math-7B"
# MODEL_PATH="deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"
MODEL_PATH=checkpoints/Qwen2.5-Math-7B/gsm8k_distilled

DTYPE="bfloat16"
DECODE_STRATEGY="greedy_search"
TEMPERATURE=0.95
TOP_P=0.6
SRC_LENGTH=1024
MAX_LENGTH=3072
TOTAL_MAX_LENGTH=4096

# EVAL_PROMPT_EN="\nPlease reason step by step, and put your final answer within \\boxed{}."
# EVAL_PROMPT_ZH="\n请一步一步地推理，并将你的最终答案放在\boxed{}中。"
EVAL_PROMPT_EN=""
EVAL_PROMPT_ZH=""

INPUT_FILE="./data/gsm8k_distilled_zh/GSM8K_distilled_zh-test.json"
OUTPUT_DIR="results-gsm8k/${MODEL_PATH}"

EVAL_RESULTS="${OUTPUT_DIR}/output_zh.json"
mkdir -p ${OUTPUT_DIR} && touch ${EVAL_RESULTS}
python -u -m paddle.distributed.launch \
    --devices "0,1,2,3" \
    distill_eval.py \
    --eval_file ${INPUT_FILE} \
    --eval_question_key "question_zh" \
    --eval_answer_key "answer_only" \
    --eval_prompt "${EVAL_PROMPT_ZH}" \
    --model_name_or_path ${MODEL_PATH} \
    --inference_model true \
    --append_attn true \
    --dtype ${DTYPE} \
    --batch_size 32 \
    --use_flash_attention true \
    --src_length ${SRC_LENGTH} \
    --max_length ${MAX_LENGTH} \
    --total_max_length ${TOTAL_MAX_LENGTH} \
    --decode_strategy ${DECODE_STRATEGY} \
    --temperature ${TEMPERATURE} \
    --top_p ${TOP_P} \
    --data_file ${INPUT_FILE} \
    --eval_results ${EVAL_RESULTS}


EVAL_RESULTS="${OUTPUT_DIR}/output_en.json"
mkdir -p ${OUTPUT_DIR} && touch ${EVAL_RESULTS}
python -u -m paddle.distributed.launch \
    --devices "0,1,2,3" \
    distill_eval.py \
    --eval_file ${INPUT_FILE} \
    --eval_question_key "question" \
    --eval_answer_key "answer_only" \
    --eval_prompt "${EVAL_PROMPT_EN}" \
    --model_name_or_path ${MODEL_PATH} \
    --inference_model true \
    --append_attn true \
    --dtype ${DTYPE} \
    --batch_size 32 \
    --use_flash_attention true \
    --src_length ${SRC_LENGTH} \
    --max_length ${MAX_LENGTH} \
    --total_max_length ${TOTAL_MAX_LENGTH} \
    --decode_strategy ${DECODE_STRATEGY} \
    --temperature ${TEMPERATURE} \
    --top_p ${TOP_P} \
    --data_file ${INPUT_FILE} \
    --eval_results ${EVAL_RESULTS}