Hugging Face's logo

htxu91
/
zero_tir_0727

Safetensors
Model card Files
xet
Community
zero_tir_0727
File size: 4,615 Bytes 
c8140ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
 
# apt-get update && \
#     apt-get install -y gosu && \
#     rm -rf /var/lib/apt/lists/*

# apt-get update && apt-get -y install sudo

echo "Number of GPUS: $N_GPUS"
echo "Number of process: $NUM_PROCESSES"
echo "WORLD_SIZE: $WORLD_SIZE"
echo "RANK: $RANK"
echo "MASTER_ADDR: $MASTER_ADDR"
echo "MASTER_PORT: $MASTER_PORT"

# export VLLM_PATH=/cpfs/user/chenhao/vllm
# export PYTHONPATH=$VLLM_PATH:$PYTHONPATH

export RANK=${RANK}
export MY_RANK=2
export NUM_PROCESSES=$(expr $RANK \* $MY_RANK)
echo "MY_RANK: $MY_RANK"
echo "RANK: $RANK"
echo "NUM_PROCESSES: $NUM_PROCESSES"
# export VLLM_USE_V1=0

# pip3 install deepspeed==0.16.0

# cd /cpfs/user/chenhao/debug/
# cp nccl.conf /etc/nccl.conf
# echo "COPY nccl.conf to etc"
# cp parameter_offload.py /usr/local/lib/python3.10/dist-packages/deepspeed/runtime/zero/parameter_offload.py
# echo "COPY parameter_offload to deepspeed"
# cp partitioned_param_coordinator.py /usr/local/lib/python3.10/dist-packages/deepspeed/runtime/zero/partitioned_param_coordinator.py
# echo "COPY partitioned_param_coordinator to deepspeed"

pip3 install math-verify tabulate markdown pysbd jsonlines coloredlogs func_timeout timeout-decorator word2number Pebble -i  https://mirrors.cloud.aliyuncs.com/pypi/simple --trusted-host mirrors.cloud.aliyuncs.com

pip3 install loguru fastapi uvicorn httpx python-multipart aiohttp aiolimiter pysbd jsonlines coloredlogs pebble aiolimiter -i  https://mirrors.cloud.aliyuncs.com/pypi/simple --trusted-host mirrors.cloud.aliyuncs.com
pip3 install func_timeout sentencex requests_futures timeout_decorator flashtext pygments -i  https://mirrors.cloud.aliyuncs.com/pypi/simple --trusted-host mirrors.cloud.aliyuncs.com

pip3 install math-verify loguru fastapi uvicorn httpx python-multipart aiohttp aiolimiter pysbd jsonlines coloredlogs pebble aiolimiter -i  https://mirrors.cloud.aliyuncs.com/pypi/simple --trusted-host mirrors.cloud.aliyuncs.com
pip3 install func_timeout sentencex requests_futures timeout_decorator flashtext pygments -i  https://mirrors.cloud.aliyuncs.com/pypi/simple --trusted-host mirrors.cloud.aliyuncs.com

# export ROOT_PATH=/cpfs/user/chenhao/outputs/qwen25_7B_reinforce_baseline_zero_tir_fix_boxed_lr1e-6_warmup0.0_kl0.0_zero_tir_0426_nginx_prefetch_fix_env_mask_vllm083_xverify_dapo_async_iternum2/

# export ROOT_PATH=/cpfs/user/chenhao/outputs/qwen25_7B_reinforce_baseline_zero_tir_fix_boxed_lr1e-6_warmup0.0_kl0.0_zero_tir_0502_nginx_prefetch_fix_env_mask_vllm083_xverify_orz_async_pipline_iternum2/

# export ROOT_PATH=/cpfs/user/chenhao/outputs/qwen25_7B_reinforce_baseline_zero_tir_fix_boxed_lr1e-6_warmup0.0_kl0.0_zero_tir_0504_nginx_prefetch_fix_env_mask_vllm083_xverify_deepmath_async_pipline_iternum2/

export ROOT_PATH=/newcpfs/user/chenhao/outputs/qwen25_32B_reinforce_baseline_zero_tir_lr1e-6_warmup0.0_kl0.0_zero_0812_agent_tir_iternum8_queue_size1_rolloutn16_orz_dapo_seqbalance_raw_adamw_before_select_dualclip_lossmask_dynamicbs_globaltoken_correction_latest/

# for step in 250 200 150 100 50
# do
#     cd ${ROOT_PATH}_actor/
#     mkdir ./global_step${step}/ckpt/
#     rm -r ./global_step${step}/ckpt/
#     python /cpfs/user/chenhao/debug/zero_to_fp32.py . ./global_step${step}/ckpt/pytorch_model.bin -t global_step${step}
#     cp -r /cpfs/user/chenhao/pretrained_models/Qwen/Qwen2.5-7B-local/*.json ./global_step${step}/ckpt/pytorch_model.bin/
# done

cd /cpfs/user/chenhao/Qwen2-Math/evaluation
# export VLLM_ENABLE_V1_MULTIPROCESSING='0'

export NGINX_IP_FILE=/cpfs/user/chenhao/hf_datasets/qwen25_qwq/nginx_conf/nginx_ip.txt
export COMPILE_SERVER_PORT='10003'
export MATH_VERIFY_SERVER_PORT='10008'
export XVERIFY_MATH_MODEL_SERVER_PORT='10005'
export REMOTE_RM_URL='http://10.39.2.54:10007'
export OPENRLHF_PATH=/cpfs/user/chenhao/debug/OpenRLHF_082/
export PRETRAIN=/newcpfs/user/chenhao/pretrained_models/Qwen/Qwen2.5-7B-local/

export DEBUG_FLAG='yes'
export CUDA_VISIBLE_DEVICES="0,1"
export INPUT_KEY='problem,question'
export ANSWER_KEY='answer,final_answer'
export DATA_NAME="aime25,aime24,hmmt_feb_2025,hmmt_feb_2024,cmimc"
export N_SAMPLING=32
export TEMPERATURE=1.0
# export VLLM_USE_V1='0'
export USE_TIR='yes'
export TASK_MAX_CONCURRENT=32

export VLLM_VERSION='vllm085'
export USE_SEPERATE='no'
export USE_ID='USE_ID'

for step in 100 150
do
    for iter in 1 2 4 8 16 18 20
    do
        export ENV_ITER_NUM=${iter}
        export MODEL_NAME_OR_PATH=${ROOT_PATH}global_step${step}_hf_actor/
        export OUTPUT_DIR=${MODEL_NAME_OR_PATH}/math_eval_useid
        export PROMPT_TYPE='orz_tir'
        export USE_SEPERATE='yes'
        bash run_evaluation.sh
    done
done