Merge branch 'swx' into 'master'

Swx

See merge request !1

analysis/README.md

+### beam size关键路径
+prm/openr/reason/evaluation/evaluate.py line 201
+prm/openr/reason/evaluation/methods.py line 122
+prm/openr/reason/guided_search/tree.py line 434
+prm/openr/reason/guided_search/tree.py line 449
+prm/openr/reason/guided_search/tree.py line 461
\ No newline at end of file

openr/envs/base_env.py

@@ -218,7 +218,7 @@ class CoTEnv(BaseEnv):
             processed_act = self.post_process_act(texts[i])
             if (
                 len(processed_act) > 0
-                and processed_act not in text_list
+                # and processed_act not in text_list
                 # only stop is valid, otherwise the output action is truncated actually
                 and result.finish_reason[i] == "stop" 
             ):

openr/infer.sh

 #!/bin/bash
+if [ ! -d "ret_one" ]; then
+    mkdir -p "ret_one"
+fi 
 
-#- Job parameters
-
-# (TODO)
-# Please modify job name
-
-#SBATCH -J inference            # The job name
-#SBATCH -o inference.out        # Write the standard output to file named 'ret-<job_number>.out'
-#SBATCH -e inference.err        # Write the standard error to file named 'ret-<job_number>.err'
-
-
-#- Resources
-
-# (TODO)
-# Please modify your requirements
-
-#SBATCH -p r8nv-gpu-hw                   # Submit to 'nv-gpu' Partitiion
-#SBATCH -t 1-06:00:00                # Run for a maximum time of 0 days, 12 hours, 00 mins, 00 secs
-#SBATCH --nodes=1                    # Request N nodes
-#SBATCH --gres=gpu:8                 # Request M GPU per node
-#SBATCH --gres-flags=enforce-binding # CPU-GPU Affinity
-#SBATCH --qos=gpu-normal           # Request QOS Type
-#SBATCH --constraint="L40S"
-
-###
-### The system will alloc 8 or 16 cores per gpu by default.
-### If you need more or less, use following:
-### #SBATCH --cpus-per-task=K            # Request K cores
-###
-### 
-### Without specifying the constraint, any available nodes that meet the requirement will be allocated
-### You can specify the characteristics of the compute nodes, and even the names of the compute nodes
-###
-### #SBATCH --nodelist=gpu-v00           # Request a specific list of hosts 
-### #SBATCH --constraint="Volta|RTX8000" # Request GPU Type: Volta(V100 or V100S) or RTX8000
-###
-
-#- Log information
-
-echo "Job start at $(date "+%Y-%m-%d %H:%M:%S")"
-echo "Job run at:"
-echo "$(hostnamectl)"
-# echo "$(df -h | grep -v tmpfs)"
-
-#- Load environments
-#- Load environments
-module unload cuda-cudnn
-source ~/.bashrc
-module list                       # list modules loaded
-conda activate open_reasoner
-
-echo $(module list)              # list modules loaded
-echo $(which gcc)
-echo $(which python)
-echo $(which python3)
-
-cluster-quota                    # nas quota
-
-# nvidia-smi --format=csv --query-gpu=name,driver_version,power.limit # gpu info
-nvidia-smi
-
-#- Warning! Please not change your CUDA_VISIBLE_DEVICES
-#- in `.bashrc`, `env.sh`, or your job script
-echo "Use GPU ${CUDA_VISIBLE_DEVICES}"                              # which gpus
-#- The CUDA_VISIBLE_DEVICES variable is assigned and specified by SLURM
-
-# export CUDA_DEVICE_ORDER="PCI_BUS_ID"
-
-#- Important setting!!!
-##  otherwise it will cause an error of insufficient RDMA resources:
-ulimit -l unlimited
-##  otherwise it will result in an insufficient virtual memory size error, especially when loading LLM:
-ulimit -v unlimited
-ulimit -n 65535
-ulimit -u 4125556
-
-#- Job step
-# sleep 30h
-cd /nfs_global/S/zhangxiaoyun/prm/openr
-export PYTHONPATH=$(pwd)
-
-bash reason/llm_service/create_service_qwen2.5_math_vllm_gold_prm_speed.sh --acc 1.0 --policy_model_name Qwen2.5-Math-7B-Instruct
-sleep 100s
-bash scripts/eval/beam_search.sh --acc 1.0 --policy_model_name Qwen2.5-Math-7B-Instruct
-
-#- End
-echo "Job end at $(date "+%Y-%m-%d %H:%M:%S")"
+sbatch --job-name=test -o "ret_one/%j.out" -e "ret_one/%j.err" infer.slurm
\ No newline at end of file

openr/infer.slurm

+#!/bin/bash
+
+#- Job parameters
+
+# (TODO)
+# Please modify job name
+
+#SBATCH -J inference            # The job name
+#SBATCH -o inference.out        # Write the standard output to file named 'ret-<job_number>.out'
+#SBATCH -e inference.err        # Write the standard error to file named 'ret-<job_number>.err'
+
+
+#- Resources
+
+# (TODO)
+# Please modify your requirements
+
+#SBATCH -p r8nv-gpu-hw                   # Submit to 'nv-gpu' Partitiion
+#SBATCH -t 1-06:00:00                # Run for a maximum time of 0 days, 12 hours, 00 mins, 00 secs
+#SBATCH --nodes=1                    # Request N nodes
+#SBATCH --gres=gpu:8                 # Request M GPU per node
+#SBATCH --gres-flags=enforce-binding # CPU-GPU Affinity
+#SBATCH --qos=gpu-normal           # Request QOS Type
+#SBATCH --constraint="L40"
+
+###
+### The system will alloc 8 or 16 cores per gpu by default.
+### If you need more or less, use following:
+### #SBATCH --cpus-per-task=K            # Request K cores
+###
+### 
+### Without specifying the constraint, any available nodes that meet the requirement will be allocated
+### You can specify the characteristics of the compute nodes, and even the names of the compute nodes
+###
+### #SBATCH --nodelist=gpu-v00           # Request a specific list of hosts 
+### #SBATCH --constraint="Volta|RTX8000" # Request GPU Type: Volta(V100 or V100S) or RTX8000
+###
+
+#- Log information
+
+echo "Job start at $(date "+%Y-%m-%d %H:%M:%S")"
+echo "Job run at:"
+echo "$(hostnamectl)"
+# echo "$(df -h | grep -v tmpfs)"
+
+#- Load environments
+#- Load environments
+module unload cuda-cudnn
+source ~/.bashrc
+module list                       # list modules loaded
+conda activate open_reasoner
+
+echo $(module list)              # list modules loaded
+echo $(which gcc)
+echo $(which python)
+echo $(which python3)
+
+cluster-quota                    # nas quota
+
+# nvidia-smi --format=csv --query-gpu=name,driver_version,power.limit # gpu info
+nvidia-smi
+
+#- Warning! Please not change your CUDA_VISIBLE_DEVICES
+#- in `.bashrc`, `env.sh`, or your job script
+echo "Use GPU ${CUDA_VISIBLE_DEVICES}"                              # which gpus
+#- The CUDA_VISIBLE_DEVICES variable is assigned and specified by SLURM
+
+# export CUDA_DEVICE_ORDER="PCI_BUS_ID"
+
+#- Important setting!!!
+##  otherwise it will cause an error of insufficient RDMA resources:
+ulimit -l unlimited
+##  otherwise it will result in an insufficient virtual memory size error, especially when loading LLM:
+ulimit -v unlimited
+ulimit -n 65535
+ulimit -u 20000
+
+#- Job step
+# sleep 30h
+cd /nfs_global/S/shiwenxuan/prm/openr
+export PYTHONPATH=$(pwd)
+
+bash reason/llm_service/create_service_qwen2.5_math_vllm_gold_prm_speed.sh --acc 0.3 --policy_model_name Qwen2.5-Math-1.5B-Instruct
+sleep 100s
+bash scripts/eval/beam_search.sh --acc 0.3 --policy_model_name Qwen2.5-Math-1.5B-Instruct
+
+# sleep 6h
+
+#- End
+echo "Job end at $(date "+%Y-%m-%d %H:%M:%S")"