From 7d861c7741effaeb03bc28429c3deeb1471bf15d Mon Sep 17 00:00:00 2001 From: Hanbyul Kim Date: Tue, 22 Jul 2025 04:12:00 +0000 Subject: [PATCH 1/3] Fix: Correct tensor-parallel-size argument --- verifiers/examples/math_python.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/verifiers/examples/math_python.py b/verifiers/examples/math_python.py index 41a75e4302..9f8c0e0008 100644 --- a/verifiers/examples/math_python.py +++ b/verifiers/examples/math_python.py @@ -5,7 +5,7 @@ """ Multi-GPU training (single node, 4 training + 4 inference) -CUDA_VISIBLE_DEVICES=0,1,2,3 vf-vllm --model 'willcb/Qwen2.5-7B-Math-Python-SFT' --tensor_parallel_size 4 +CUDA_VISIBLE_DEVICES=0,1,2,3 vf-vllm --model 'willcb/Qwen2.5-7B-Math-Python-SFT' --tensor-parallel-size 4 CUDA_VISIBLE_DEVICES=4,5,6,7 accelerate launch --config-file configs/zero3.yaml verifiers/examples/math_train.py """ From 3cbdd537545f2ad91ed37446f40cb916a2d5b7a4 Mon Sep 17 00:00:00 2001 From: Hanbyul Kim Date: Tue, 22 Jul 2025 04:12:07 +0000 Subject: [PATCH 2/3] Fix: Update import path for data_utils --- verifiers/examples/math_python.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/verifiers/examples/math_python.py b/verifiers/examples/math_python.py index 9f8c0e0008..7997b1e7dd 100644 --- a/verifiers/examples/math_python.py +++ b/verifiers/examples/math_python.py @@ -1,6 +1,6 @@ import verifiers as vf from verifiers.tools import python -from verifiers.utils import load_example_dataset +from verifiers.utils.data_utils import load_example_dataset """ Multi-GPU training (single node, 4 training + 4 inference) From ef7e5a097957257c9aace5b8b11f9963a188f41f Mon Sep 17 00:00:00 2001 From: Hanbyul Kim Date: Tue, 22 Jul 2025 04:12:29 +0000 Subject: [PATCH 3/3] Add logprob param to ToolEnv initialization --- verifiers/examples/math_python.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/verifiers/examples/math_python.py b/verifiers/examples/math_python.py index 7997b1e7dd..54f9c47071 100644 --- a/verifiers/examples/math_python.py +++ b/verifiers/examples/math_python.py @@ -50,7 +50,8 @@ system_prompt=TOOL_PROMPT, few_shot=[], tools=[python], - max_steps=3 + max_steps=3, + sampling_args={"extra_body": {"logprobs": True}} ) print(vf_env.system_prompt)