feat: BFGS cpu multiple-thread support

mace-bench @ 2207c831

-Subproject commit b43f7f90237c7e2bda5dd82390e152726692c6c2
+Subproject commit 2207c831ce6d8a9e15f51d6afadbe64b6d4e15cf

scripts/relaxation.sh

@@ -59,6 +59,7 @@ FILTER2             = ${FILTER2:-<unset>}
 [Model]
 MODEL               = ${MODEL}
 CUEQ                = ${CUEQ}
+BFGS_CPU_THREADS    = ${BFGS_CPU_THREADS:-<unset>}
 USE_ORDERED_FILES   = ${USE_ORDERED_FILES}
 COMPILE_MODE        = ${COMPILE_MODE:-<unset>}
 
@@ -91,6 +92,9 @@ ORDERED_FILES_ARG=""
 CUEQ_ARG=""
 [ "${CUEQ}" = true ] && CUEQ_ARG="--cueq true"
 
+BFGS_CPU_THREADS_ARG=""
+[ -n "${BFGS_CPU_THREADS}" ] && BFGS_CPU_THREADS_ARG="--bfgs_cpu_thread ${BFGS_CPU_THREADS}"
+
 python "${PROJECT_DIR}/mace-bench/scripts/mace_opt_batch.py" \
     --target_folder "${TARGET_FOLDER}" \
     --molecule_single ${MOLECULE_SINGLE} \
@@ -105,7 +109,7 @@ python "${PROJECT_DIR}/mace-bench/scripts/mace_opt_batch.py" \
     --num_threads ${NUM_THREADS} \
     --model ${MODEL} \
     --output_path "${OUTPUT_PATH}" \
-    ${FILTER1_ARG} ${FILTER2_ARG} ${COMPILE_ARG} ${BIND_CORES_ARG} ${PROFILE_ARG} ${ORDERED_FILES_ARG} ${CUEQ_ARG} \
+    ${FILTER1_ARG} ${FILTER2_ARG} ${COMPILE_ARG} ${BIND_CORES_ARG} ${PROFILE_ARG} ${ORDERED_FILES_ARG} ${CUEQ_ARG} ${BFGS_CPU_THREADS_ARG} \
     > "${OUTPUT_PATH}/opt.log" 2>&1
 
 echo "Relaxation completed. Results in ${OUTPUT_PATH}"

scripts/relaxation_config.sh

@@ -33,6 +33,7 @@
 #   MODEL               ML potential model to use. Choices:
 #                       mace, chgnet, sevennet, uma-s-1.
 #   CUEQ                Whether to use cuEquivariance Library (true/false).
+#   BFGS_CPU_THREADS    Number of threads for BFGS CPU eigh parallel execution
 #   USE_ORDERED_FILES   Whether to sort files by atomic number descending (true/false).
 #   COMPILE_MODE        torch.compile mode: none / default / reduce-overhead /
 #                       max-autotune / max-autotune-no-cudagraphs.
@@ -47,15 +48,17 @@
 #                       auto-generated timestamped directory.
 # =============================================================================
 
-TARGET_FOLDER="${PROJECT_DIR}/csp_results/structures"
-MOLECULE_SINGLE=13
+#TARGET_FOLDER="${PROJECT_DIR}/csp_results/structures_1"
+#TARGET_FOLDER="/data/wangleping/csp_data/perf_46"
+TARGET_FOLDER="/data/wangleping/csp_data/perf_v2_sorted/perf_v2_92"
+MOLECULE_SINGLE=46
 N_GPUS=8
 GPU_OFFSET=0
-NUM_WORKERS=8
-BATCH_SIZE=25
-NUM_THREADS=4
+NUM_WORKERS=32
+BATCH_SIZE=32
+NUM_THREADS=1
 BIND_CORES=""
-MAX_STEPS=50
+MAX_STEPS=3000
 FMAX=0.01
 SCALAR_PRESSURE=0.0006
 SKIP_SECOND_STAGE=false
@@ -65,7 +68,8 @@ FILTER1="UnitCellFilter"
 FILTER2="UnitCellFilter"
 MODEL="mace"
 CUEQ=true
-USE_ORDERED_FILES=false
+BFGS_CPU_THREADS=16
+USE_ORDERED_FILE=true
 COMPILE_MODE=""
 PROFILE="false"
 OUTPUT_PATH=""