vllm: [Bug]:There is no module or parameter named 'base_model' in Qwen2ForCausalLM

Your current environment

The output of `python collect_env.py`
INFO 02-08 23:01:58 __init__.py:190] Automatically detected platform cuda.
Collecting environment information...
PyTorch version: 2.5.1+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 12.3.0-1ubuntu1~22.04) 12.3.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35

Python version: 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 18:56:27) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-6.8.0-52-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.4.99
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA GeForce RTX 3090
GPU 1: NVIDIA GeForce RTX 3090
GPU 2: NVIDIA GeForce RTX 3090
GPU 3: NVIDIA GeForce RTX 3090

Nvidia driver version: 570.86.16
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
架构:                                x86_64
CPU 运行模式:                        32-bit, 64-bit
Address sizes:                        46 bits physical, 48 bits virtual
字节序:                              Little Endian
CPU:                                  20
在线 CPU 列表:                       0-19
厂商 ID:                             GenuineIntel
型号名称:                            Intel(R) Core(TM) i9-7900X CPU @ 3.30GHz
CPU 系列:                            6
型号:                                85
每个核的线程数:                      2
每个座的核数:                        10
座:                                  1
步进:                                4
CPU 最大 MHz:                        4500.0000
CPU 最小 MHz:                        1200.0000
BogoMIPS:                            6599.98
标记:                                fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 pti ssbd mba ibrs ibpb stibp tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi md_clear flush_l1d arch_capabilities
虚拟化:                              VT-x
L1d 缓存:                            320 KiB (10 instances)
L1i 缓存:                            320 KiB (10 instances)
L2 缓存:                             10 MiB (10 instances)
L3 缓存:                             13.8 MiB (1 instance)
NUMA 节点:                           1
NUMA 节点0 CPU:                      0-19
Vulnerability Gather data sampling:   Mitigation; Microcode
Vulnerability Itlb multihit:          KVM: Mitigation: VMX disabled
Vulnerability L1tf:                   Mitigation; PTE Inversion; VMX conditional cache flushes, SMT vulnerable
Vulnerability Mds:                    Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Meltdown:               Mitigation; PTI
Vulnerability Mmio stale data:        Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed:               Mitigation; IBRS
Vulnerability Spec rstack overflow:   Not affected
Vulnerability Spec store bypass:      Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:             Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:             Mitigation; IBRS; IBPB conditional; STIBP conditional; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Mitigation; Clear CPU buffers; SMT vulnerable

Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-ml-py==12.570.86
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==26.2.1
[pip3] torch==2.5.1
[pip3] torchaudio==2.5.1
[pip3] torchvision==0.20.1
[pip3] transformers==4.48.3
[pip3] triton==3.1.0
[conda] cuda-cccl                 12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-command-line-tools   12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-compiler             12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cudart               12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cudart-dev           12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cudart-static        12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cuobjdump            12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cupti                12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cupti-static         12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-cuxxfilt             12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-documentation        12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-driver-dev           12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-gdb                  12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-libraries            12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-libraries-dev        12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-libraries-static     12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nsight               12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nsight-compute       12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvcc                 12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvdisasm             12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvml-dev             12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvprof               12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvprune              12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvrtc                12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvrtc-dev            12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvrtc-static         12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvtx                 12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-nvvp                 12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-opencl               12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-opencl-dev           12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-profiler-api         12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-sanitizer-api        12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-toolkit              12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-tools                12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] cuda-visual-tools         12.4.0                        0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] gds-tools                 1.9.0.20                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcublas                 12.4.2.65                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcublas-dev             12.4.2.65                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcublas-static          12.4.2.65                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufft                  11.2.0.44                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufft-dev              11.2.0.44                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufft-static           11.2.0.44                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufile                 1.9.0.20                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufile-dev             1.9.0.20                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcufile-static          1.9.0.20                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcurand                 10.3.5.119                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcurand-dev             10.3.5.119                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcurand-static          10.3.5.119                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusolver               11.6.0.99                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusolver-dev           11.6.0.99                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusolver-static        11.6.0.99                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusparse               12.3.0.142                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusparse-dev           12.3.0.142                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libcusparse-static        12.3.0.142                    0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnpp                    12.2.5.2                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnpp-dev                12.2.5.2                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnpp-static             12.2.5.2                      0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvfatbin               12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvfatbin-dev           12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvjitlink              12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvjitlink-dev          12.4.99                       0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvjpeg                 12.3.1.89                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvjpeg-dev             12.3.1.89                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] libnvjpeg-static          12.3.1.89                     0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] nsight-compute            2024.1.0.13                   0    https://mirrors.sustech.edu.cn/anaconda-extra/cloud/nvidia/label/cuda-12.4.0
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] nvidia-cublas-cu12        12.4.5.8                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.4.127                 pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.1.0.70                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.2.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.5.147               pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.6.1.9                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.3.1.170               pypi_0    pypi
[conda] nvidia-ml-py              12.570.86                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.21.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.4.127                 pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.4.127                 pypi_0    pypi
[conda] pyzmq                     26.2.1                   pypi_0    pypi
[conda] torch                     2.5.1                    pypi_0    pypi
[conda] torchaudio                2.5.1                    pypi_0    pypi
[conda] torchvision               0.20.1                   pypi_0    pypi
[conda] transformers              4.48.3                   pypi_0    pypi
[conda] triton                    3.1.0                    pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.7.2
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0	GPU1	GPU2	GPU3	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	PIX	NODE	NODE	0-19	0		N/A
GPU1	PIX	 X 	NODE	NODE	0-19	0		N/A
GPU2	NODE	NODE	 X 	PIX	0-19	0		N/A
GPU3	NODE	NODE	PIX	 X 	0-19	0		N/A

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

LD_LIBRARY_PATH=/usr/local/cuda-12.1/lib64:/usr/local/cuda/lib64:
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

When fine - tuning Qwen2.5 using the LoRA and GRPO algorithms, an error message was displayed: “There is no module or parameter named ‘base_model’ in Qwen2ForCausalLM”.

The content of the script is as follows

import re
import torch
from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM
from trl import GRPOConfig, GRPOTrainer
from peft import LoraConfig

# 系统提示词模板
SYSTEM_PROMPT = """
Respond in the following format:
<reasoning>
...
</reasoning>
<answer>
...
</answer>
"""

# XML格式的思维链模板
XML_COT_FORMAT = """\
<reasoning>
{reasoning}
</reasoning>
<answer>
{answer}
</answer>
"""
def extract_xml_answer(text: str) -> str:
    """从XML格式文本中提取答案"""
    # 示例输入:"...<answer>3</answer>..."
    # 输出:"3"
    answer = text.split("<answer>")[-1]
    answer = answer.split("</answer>")[0]
    return answer.strip()

def extract_hash_answer(text: str) -> str | None:
    """从原始答案中提取数值答案"""
    # 示例输入:"...#### 3"
    # 输出:"3"
    if "####" not in text:
        return None
    return text.split("####")[1].strip()

def get_gsm8k_questions(split = "train") -> Dataset:
    """加载并预处理GSM8K数据集"""
    data = load_dataset('openai/gsm8k', 'main')[split] # type: ignore
    data = data.map(lambda x: { # type: ignore
        'prompt': [
            {'role': 'system', 'content': SYSTEM_PROMPT},
            {'role': 'user', 'content': x['question']}
        ],
        'answer': extract_hash_answer(x['answer'])
    }) # type: ignore
    return data # type: ignore

dataset = get_gsm8k_questions()
def correctness_reward_func(prompts, completions, answer, **kwargs) -> list[float]:
    """答案正确性奖励(0或2分)"""
    responses = [completion[0]['content'] for completion in completions]
    q = prompts[0][-1]['content']
    extracted_responses = [extract_xml_answer(r) for r in responses]
    # 打印调试信息
    print('-'*20, f"问题:\n{q}", 
          f"\n正确答案:\n{answer[0]}", 
          f"\n模型回答:\n{responses[0]}", 
          f"\n提取答案:\n{extracted_responses[0]}")
    return [2.0 if r == a else 0.0 for r, a in zip(extracted_responses, answer)]

def int_reward_func(completions, **kwargs) -> list[float]:
    """答案是否为整数的奖励(0或0.5分)"""
    responses = [completion[0]['content'] for completion in completions]
    extracted_responses = [extract_xml_answer(r) for r in responses]
    return [0.5 if r.isdigit() else 0.0 for r in extracted_responses]

def strict_format_reward_func(completions, **kwargs) -> list[float]:
    """严格格式检查奖励(正则匹配)"""
    pattern = r"^<reasoning>\n.*?\n</reasoning>\n<answer>\n.*?\n</answer>\n$"
    responses = [completion[0]["content"] for completion in completions]
    matches = [re.match(pattern, r) for r in responses]
    return [0.5 if match else 0.0 for match in matches]

def soft_format_reward_func(completions, **kwargs) -> list[float]:
    """宽松格式检查奖励(允许空格)"""
    pattern = r"<reasoning>.*?</reasoning>\s*<answer>.*?</answer>"
    responses = [completion[0]["content"] for completion in completions]
    matches = [re.match(pattern, r) for r in responses]
    return [0.5 if match else 0.0 for match in matches]

def count_xml(text) -> float:
    """XML标签完整性评分"""
    count = 0.0
    # 检查开始标签
    if text.count("<reasoning>\n") == 1:
        count += 0.125
    # 检查结束标签
    if text.count("\n</reasoning>\n") == 1:
        count += 0.125
    # 检查答案标签
    if text.count("\n<answer>\n") == 1:
        count += 0.125
        # 惩罚多余内容
        count -= len(text.split("\n</answer>\n")[-1])*0.001
    if text.count("\n</answer>") == 1:
        count += 0.125
        count -= (len(text.split("\n</answer>")[-1]) - 1)*0.001
    return count

def xmlcount_reward_func(completions, **kwargs) -> list[float]:
    """XML结构完整性奖励"""
    contents = [completion[0]["content"] for completion in completions]
    return [count_xml(c) for c in contents]
model_name = "/mnt/nfs/wangyu/Qwen2.5-0.5B-Instruct"  # 使用Qwen xx亿参数指令微调版
output_dir="outputs/Qwen-0.5B-GRPO"                     # 输出目录
run_name="Qwen-0.5B-GRPO-gsm8k"                         # 实验名称

training_args = GRPOConfig(
    output_dir=output_dir,
    run_name=run_name,
    learning_rate=5e-6,        # 学习率
    adam_beta1 = 0.9,          # Adam优化器参数
    adam_beta2 = 0.99,
    weight_decay = 0.1,        # 权重衰减
    warmup_ratio = 0.1,        # 热身比例
    lr_scheduler_type='cosine',# 学习率调度器
    logging_steps=1,           # 每步记录日志
    bf16=True,                 # 使用bfloat16精度
    #fp16=True,
    per_device_train_batch_size=1,  # 每个设备的批量大小
    gradient_accumulation_steps=4,  # 梯度累积步数
    num_generations=16,        # 每次生成样本数
    max_prompt_length=256,     # 提示最大长度
    max_completion_length=200, # 生成最大长度
    num_train_epochs=1,        # 训练轮数
    save_steps=100,            # 保存间隔
    max_grad_norm=0.1,         # 梯度裁剪
    log_on_each_node=False,
    # use_vllm=True,             # 启用vLLM加速
    # vllm_gpu_memory_utilization=.3,  # GPU显存使用率
    # vllm_device="cuda:3",      # 使用GPU设备
    report_to="none"           # 禁用WandB报告
)

peft_config = LoraConfig(
    r=16,
    lora_alpha=64,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "up_proj", "down_proj", "gate_proj"],
    #task_type="CAUSAL_LM",
    lora_dropout=0.05,
)

"""### 初始化模型和训练器"""

# 加载预训练模型
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    #torch_dtype=torch.float16,
    torch_dtype=torch.bfloat16,  # 使用bfloat16精度
    device_map="auto",
    # device_map=None
    
) #.to("cuda")

# 初始化分词器
tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.pad_token = tokenizer.eos_token  # 设置填充token
# use peft at your own risk; not working for me with multi-GPU training
trainer = GRPOTrainer(
    model=model,
    processing_class=tokenizer,
    reward_funcs=[              # 组合多个奖励函数
        xmlcount_reward_func,    # XML结构完整性(0-0.5分)
        soft_format_reward_func, # 宽松格式检查(0或0.5分)
        strict_format_reward_func, # 严格格式检查(0或0.5分)
        int_reward_func,         # 答案格式检查(0或0.5分)
        correctness_reward_func  # 正确答案检查(0或2分)
    ],
    args=training_args,
    train_dataset=dataset,
    peft_config=peft_config      # 使用Lora
)
# 开始训练
trainer.train()

The specific error message is as follows

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[7], [line 2](vscode-notebook-cell:?execution_count=7&line=2)
      [1](vscode-notebook-cell:?execution_count=7&line=1) # 开始训练
----> [2](vscode-notebook-cell:?execution_count=7&line=2) trainer.train()

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2171, in Trainer.train(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
   [2169](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2169)         hf_hub_utils.enable_progress_bars()
   [2170](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2170) else:
-> [2171](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2171)     return inner_training_loop(
   [2172](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2172)         args=args,
   [2173](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2173)         resume_from_checkpoint=resume_from_checkpoint,
   [2174](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2174)         trial=trial,
   [2175](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2175)         ignore_keys_for_eval=ignore_keys_for_eval,
   [2176](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2176)     )

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2531, in Trainer._inner_training_loop(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)
   [2524](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2524) context = (
   [2525](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2525)     functools.partial(self.accelerator.no_sync, model=model)
   [2526](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2526)     if i != len(batch_samples) - 1
   [2527](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2527)     and self.accelerator.distributed_type != DistributedType.DEEPSPEED
   [2528](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2528)     else contextlib.nullcontext
   [2529](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2529) )
   [2530](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2530) with context():
-> [2531](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2531)     tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
   [2533](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2533) if (
   [2534](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2534)     args.logging_nan_inf_filter
   [2535](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2535)     and not is_torch_xla_available()
   [2536](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2536)     and (torch.isnan(tr_loss_step) or torch.isinf(tr_loss_step))
   [2537](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2537) ):
   [2538](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2538)     # if loss is nan or inf simply add the average of previous logged losses
   [2539](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:2539)     tr_loss = tr_loss + tr_loss / (1 + self.state.global_step - self._globalstep_last_logged)

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3675, in Trainer.training_step(self, model, inputs, num_items_in_batch)
   [3672](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3672)     return loss_mb.reduce_mean().detach().to(self.args.device)
   [3674](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3674) with self.compute_loss_context_manager():
-> [3675](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3675)     loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
   [3677](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3677) del inputs
   [3678](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3678) if (
   [3679](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3679)     self.args.torch_empty_cache_steps is not None
   [3680](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3680)     and self.state.global_step % self.args.torch_empty_cache_steps == 0
   [3681](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/transformers/trainer.py:3681) ):

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:394, in GRPOTrainer.compute_loss(self, model, inputs, return_outputs, num_items_in_batch)
    [392](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:392)     if self.accelerator.is_main_process:
    [393](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:393)         llm_model = self.llm.llm_engine.model_executor.driver_worker.model_runner.model
--> [394](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:394)         llm_model.load_weights(state_dict.items())
    [395](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:395)     self._last_loaded_step = self.state.global_step
    [397](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/trl/trainer/grpo_trainer.py:397) # Generate completions using vLLM: gather all prompts and use them in a single call in the main process

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:515, in Qwen2ForCausalLM.load_weights(self, weights)
    [508](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:508) def load_weights(self, weights: Iterable[Tuple[str,
    [509](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:509)                                                torch.Tensor]]) -> Set[str]:
    [510](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:510)     loader = AutoWeightsLoader(
    [511](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:511)         self,
    [512](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:512)         skip_prefixes=(["lm_head."]
    [513](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:513)                        if self.config.tie_word_embeddings else None),
    [514](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:514)     )
--> [515](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py:515)     return loader.load_weights(weights)

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:235, in AutoWeightsLoader.load_weights(self, weights, mapper)
    [232](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:232) if mapper is not None:
    [233](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:233)     weights = mapper.apply(weights)
--> [235](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:235) autoloaded_weights = set(self._load_module("", self.module, weights))
    [236](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:236) return autoloaded_weights

File /mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:224, in AutoWeightsLoader._load_module(self, base_prefix, module, weights)
    [220](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:220)     continue
    [222](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:222) msg = (f"There is no module or parameter named '{prefix}' "
    [223](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:223)        f"in {type(self.module).__name__}")
--> [224](https://vscode-remote+ssh-002dremote-002b10-002e16-002e22-002e110.vscode-resource.vscode-cdn.net/mnt/nfs/wangyu/Miniconda/envs/whs_vllm_cuda_12_4/lib/python3.12/site-packages/vllm/model_executor/models/utils.py:224) raise ValueError(msg)

ValueError: There is no module or parameter named 'base_model' in Qwen2ForCausalLM

Before submitting a new issue…

  • Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

About this issue

  • Original URL
  • State: open
  • Created 2 years ago
  • Reactions: 4
  • Comments: 15

Most upvoted comments

In my case, i use my own pretrained adapter. update latest vllm, trl (github version)

step1: model = AutoModelForCasualLM.from_pretrained(model_name_or_path) model = PeFTModel.from_pretrained(model, adapter_path)

step2: freeze all weight except adapter (only module name “lora” to requires_grad=True)

step3:

trainer = GRPOTrainer(
    model=model,
    ...,
    # peft_config=peft_config, # Do Not input peft_config
).train()

and it’s working