lean4-compile / pass_rate_atp_pass.py
rookiemango's picture
Upload folder using huggingface_hub
1d2a897 verified
raw
history blame
7.78 kB
import pdb
import subprocess
import re
# Output file
output_file = "pass_rate_output.txt"
# Clearing the output file before appending new content
with open(output_file, "w") as file:
file.write("")
# List of input paths
input_path_lists = [
"test/zero_shot/wild_test/generation/lean4_random_15k_all/2/1/",
]
def get_output(input_string, k):
pattern = r"zero_shot/(\w+)/(.+?)/(\w+)"
match = re.search(pattern, input_string)
if match:
part1 = match.group(1)
part2 = match.group(3) + f"pass{k}.jsonl"
result = "/".join([part1, part2])
print(result)
else:
print("No match found.")
assert True
return result
# List of input paths
input_path_lists = [
# "../auto-info/generate_result/zero_shot/gsm8k_train/generation/lean4_random_15k_all/2/10/",
# "../auto-info/generate_result/zero_shot/math_train/generation/lean4_random_15k_all/2/10/",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/1/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/2/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/3/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/1/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/2/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/3/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/1/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/2/1",
# "../auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/3/1",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_verifier_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_verifier_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_verifier_rft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_15k_train/generation/lean4_random_15k_all/2/20/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_random_15k_all/2/5/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_15k_all/2/5/",
# "/opt/tiger/mariana/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_5k/2/1/",
# "test/zero_shot/lean4_random_test/generation/lean4_random_15k_all/3/1/",
# "/opt/tiger/mariana/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_random_15k_all/2/1/",
# "/opt/tiger/mariana/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_15k_all/2/1/",
# "test/zero_shot/lean4_random_test/generation/lean4_random_15k_all/3/1/",
# "test/zero_shot/lean4_basic_test/generation/lean4_random_15k_all/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_random_15k_all_mathrft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_15k_all_mathrft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_random_15k_all_mathrft/1/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_random_15k_all_mathrft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_15k_all_mathrft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_random_15k_all_mathrft/2/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_basic_test/generation/lean4_random_15k_all_mathrft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_random_test/generation/lean4_random_15k_all_mathrft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/wild_test/generation/lean4_random_15k_all_mathrft/3/1/",
# "/opt/tiger/auto-info/generate_result/zero_shot/gsm8k_train/generation/lean4_random_15k_all_mathrft/2/10/",
# "/opt/tiger/auto-info/generate_result/zero_shot/math_train/generation/lean4_random_15k_all_mathrft/2/10/",
# "/opt/tiger/auto-info/generate_result/zero_shot/lean4_15k_train/generation/lean4_random_15k_all_mathrft/2/10/",
# Add more input paths as needed
]
# Iterate through the input paths and run the command
for input_path in input_path_lists:
k = 5
if "wild_test" in input_path or "gsm8k_train" in input_path or "math_train" in input_path:
print(f"wild")
print(f"Running for input path: {input_path}", file=open(output_file, "a"))
command = f"python3 pass_rate_notlean_test.py --input_path {input_path} --output_path {get_output(input_path,k)} --k {k}"
subprocess.run(command, shell=True, stdout=open(output_file, "a"), stderr=subprocess.STDOUT)
print("\n\n",file=open(output_file, "a"))
else:
print(f"lean")
print(f"Running for input path: {input_path}", file=open(output_file, "a"))
command = f"python3 pass_rate_new_test.py --input_path {input_path} --output_path {get_output(input_path, k)} --k {k}"
subprocess.run(command, shell=True, stdout=open(output_file, "a"), stderr=subprocess.STDOUT)
print("\n\n",file=open(output_file, "a"))