AttributeError: 'Graph' object has no attribute 'find_nodes'

[Ruoxuan-Li]

当我使用AWQ量化自己训练的Qwen3-8B模型时，出现如下报错：
Traceback (most recent call last):
File "/media/lrx/LLaMA-Factory/AWQ_quantize.py", line 135, in
oneshot(
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/entrypoints/oneshot.py", line 311, in oneshot
one_shot()
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/entrypoints/oneshot.py", line 149, in call
self.apply_recipe_modifiers(
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/entrypoints/oneshot.py", line 193, in apply_recipe_modifiers
pipeline(self.model, calibration_dataloader, self.dataset_args)
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/pipelines/independent/pipeline.py", line 49, in call
pipeline(model, dataloader, dataset_args)
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/pipelines/sequential/pipeline.py", line 66, in call
subgraphs = trace_subgraphs(model, sample_input, sequential_targets, ignore)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/pipelines/sequential/helpers.py", line 142, in trace_subgraphs
partitions = topological_partition(graph, targets)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/pipelines/sequential/helpers.py", line 299, in topological_partition
for node in graph.graph.find_nodes(op="get_attr"):
^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'Graph' object has no attribute 'find_nodes'
不知道是什么原因造成的？

[Ruoxuan-Li]

已成功解决~

进入报错位置：/home/a/miniconda3/envs/llama/lib/python3.11/site-packages/llmcompressor/pipelines/sequential/helpers.py，
在helpers.py文件中出现了两处graph.find_nodes函数，找到这两处位置，进行如下修改：
第①处：
for node in graph.graph.find_nodes(op="get_attr"):
user_partitions = []
for user in node.users:
for index in range(len(partitions)):
if user in partitions[index]:
user_partitions.append(index)
break
partition_index = min(user_partitions)
partitions[partition_index].insert(0, node)
修改为：
for node in graph.graph.nodes: # 兼容新版PyTorch
if node.op == "get_attr":
user_partitions = []
for user in node.users:
for i, partition in enumerate(partitions):
if user in partition:
user_partitions.append(i)
break
if not user_partitions: # 处理无用户节点
continue
partition_index = min(user_partitions)
partitions[partition_index].insert(0, node)

第②处：
if len(graph.find_nodes(op="output")) <= 0:
output_dict = {
node.name: node_map[node]
for node in partition_nodes
if any(user not in partition_nodes for user in node.users.keys())
}
graph.output(output_dict)
修改为：
if not any(node.op == "output" for node in graph.nodes):
output_dict = {
node.name: node_map[node]
for node in partition_nodes
if any(user not in partition_nodes for user in node.users.keys())
}
graph.output(output_dict)

成功！

[brian-dellabetta]

Hi @Ruoxuan-Li , please provide your AWQ_Quantize.py code to reproduce this error so we can troubleshoot. Thanks