1. 系统环境
硬件环境(Ascend/GPU/CPU): Ascend/GPU/CPU
MindSpore版本: 2.1
执行模式（PyNative/ Graph）: 不限
2. 报错信息
2.1 问题描述
模型推理时验证model.generate()功能报错。
Traceback (most recent call last):
File "predict dx.py". line 26, in <module>
result = model.generate(input_ids=inputs["input_ids"], max_length=model.config.seq_length)
File "/home/wizardcoder/1_wiz ardcoder-mindformers7mindformers/generation/text_generator.py", line, in generate**model kwargs,
File "/home/wizardcoder/1 wizardcoder-mindformers/mindformers/generation/text_generator.py", line 329, in _forward
input ids. **model kwargs
File. "/home/wizardcoder/1 wizardcoder-mindformers /mindformers/generation/text generator.py", line 58, in prepare inputs for generation
"A model class needs to-define a `prepare inputs for aeneration*
RuntimeError A model class needs to define a `prepare inputs fordgeneration` method in order to use .generate()`
3. 根因分析
分析代码可知（见text_generator.py第48行），如果需要使用model.generate()功能，必须要重写prepare_inputs_for_generation()方法。
class GeneratorMixin:
"""Generator For the nlp models"""
def __init__(self):
pass
# pylint: disable=W0613
def prepare_inputs_for_generation(self, input_ids, **kwargs):
"""
prepare inputs for generation.
A model class needs to define a `prepare_inputs_for_generation' method
in orderto use `.generate()
Raises:
RuntimeError: Not implemented in model but call `.generate()'
"""
raise RuntimeError(
"A model class needs to define a 'prepare_inputs_for_generation'"
"method in order to use '.generate()'."
)
4. 解决方案
需要在待迁移模型中实现这个功能方法，参照GPT-2的写法，wizardcoder在WizardCoderLMHeadModel类中实现了这个功能。
def prepare_inputs_for_generation(self, input_ids, **kwargs):
input_ids = {"input_ids":Tensor(input_ids, mstype.int32)}
# input_ids = {"input_ids": input_ids}
Return input_ids