KeyError: "The name 'cudnn_lstm/opaque_kernel_saveable' refers to an Operation not in the graph."

[LangR7]

hi，I am using “deepspeech-0.9.3-checkpoint-zh-CN.tar.gz” to restore output_graph.pb。
when I use tf.compat.v1.train.import_meta_graph to load best_dev-408475.meta，I met this error：“”KeyError: "The name 'cudnn_lstm/opaque_kernel_saveable' refers to an Operation not in the graph."
my tensorflow version is 1.15.4
thanks for your help

[Adorablepet]

@LangR7 I also meet too. Have you resolved?Thanks.

[LangR7]

i use [pbmm_to_pb.py] (https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/public/mozilla-deepspeech-0.8.2/pbmm_to_pb.py)
convert pbmm to pb

[Adorablepet]

i use [pbmm_to_pb.py] (https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/public/mozilla-deepspeech-0.8.2/pbmm_to_pb.py) convert pbmm to pb

我有个疑问：pbmm_to_pb.py只是转换英文数据集下训练的模型，虽然运行中文数据集模型时候代码没有报错。但是我看了下维度是29，如果是中文的话肯定不止，不知道你有没有注意到这点。

[LangR7]

i use [pbmm_to_pb.py] (https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/public/mozilla-deepspeech-0.8.2/pbmm_to_pb.py) convert pbmm to pb

我有个疑问：pbmm_to_pb.py只是转换英文数据集下训练的模型，虽然运行中文数据集模型时候代码没有报错。但是我看了下维度是29，如果是中文的话肯定不止，不知道你有没有注意到这点。

是的啊，可是这个也可以把deepspeech最新版本的CN模型从pbmm转成pb哇

[tailangjun]

下训练的模型，虽然运行中文数据集模型时候代码没有报错。但是我看了下维度是29，如果是中文的话肯定不止，不知道你有没有注意到这点。

deepspeech-0.9.3-zh会报错，貌似维度做了调整
Tensor("deepspeech/logits:0", shape=(16, 1, 256), dtype=float32)
ValueError: Cannot feed value of shape (1, 240, 494) for Tensor 'deepspeech/input_node:0', which has shape '(1, 16, 19, 26)'