pytorch 模型 Syntax error in input! 是什么原因导致的？

模型及代码：
链接：https://pan.baidu.com/s/1j9MPyahGshjuNfw5ijIqBQ
提取码：tbbr

pytorch 1.2 rknn v1.3

1.     rknn = RKNN(verbose=True)
   
2.     rknn.config(batch_size=1,
   
3.                 epochs=1)  # asymmetric_quantized-u8,  quantized_dtype='dynamic_fixed_point-16' , channel_mean_value='0 0 0 1', reorder_channel='0 1 2',
   
4. 
   
5.     system = platform.system()
   
6. 
   
7.     seq_len = 64
   
8.     isize = 672
   
9.     # hsize = 6
   
10.     # hsize1 = 4
    
11. 
    
12.     pt_file = 'toml_fp32_acoustic.pt'
    
13.     #pt_file = 'toml_fp32_feat.pt'
    
14. 
    
15.     m = torch.jit.load(pt_file)
    
16.     m.eval()
    
17. 
    
18.     #x = torch.rand((1, seq_len, isize))
    
19.     x = torch.linspace(-1, 1, steps=(isize * seq_len))
    
20.     x = x.view(1, seq_len, -1)
    
21.     input_size_list = [[1, seq_len, isize]]
    
22.     #x1 = torch.full((1, hsize), fill_value=0.2, dtype=torch.float)
    
23.     ix = x.numpy()
    
24.     out = m.forward(x)
    
25.     print("torch out:")
    
26.     print(out)
    
27. 
    
28.     rknn.load_pytorch(model=pt_file, input_size_list=input_size_list)
    
29. 
    
30.     rknn_file = pt_file + '.rknn'
    
31.     print('--> Building model')
    
32.     ret = rknn.build(do_quantization=False, dataset='./dataset.txt')
    
33.     if ret != 0:
    
34.         print('Build pytorch failed!')
    
35.         exit(ret)
    
36.     print('build done')
    
37.     rm(rknn_file)
    
38.     rknn.export_rknn(rknn_file)
    

复制代码

.load_pytorch时错
Syntax error in input! LexToken(NAMED_IDENTIFIER,'str',15,1100)

.build时报错
--> Building model
W Genreate input meta fail, please check model.
W External input meta file "C:\Users\dm\AppData\Local\Temp\tmpm1i9kbvb\toml_fp32_acoustic_inputmeta.yml" is not exists.
Traceback (most recent call last):
  File "E:/Work/_Code/x4c/jasper/check-pt.py", line 77, in <module>
    ret = rknn.build(do_quantization=False, dataset='./dataset.txt')
  File "D:\Program Files\Python\Python36\lib\site-packages\rknn\api\rknn.py", line 238, in build
    ret = self.rknn_base.build(do_quantization=do_quantization, dataset=dataset, pack_vdata=pre_compile, batch_size=rknn_batch_size)
  File "rknn\api\rknn_base.py", line 719, in rknn.api.rknn_base.RKNNBase.build
  File "rknn\api\rknn_base.py", line 2249, in rknn.api.rknn_base.RKNNBase._generate_inputmeta
IndexError: list index out of range



这错误怎么定位啊， torch.jit.load加载.pt模型是可以正常.frward()得到结果的。

pytorch版本是多少？
rknn那边只能支持1.2的

jefferyzhang 发表于 2020-3-12 17:05
pytorch版本是多少？
rknn那边只能支持1.2的

pytorch是1.2的
rknn 1.3

问题已经上报给NPU部门了，等待回复

jefferyzhang 发表于 2020-3-12 17:19
问题已经上报给NPU部门了，等待回复

好的，谢谢啊

jefferyzhang 发表于 2020-3-12 17:19
问题已经上报给NPU部门了，等待回复

您好 我也遇到了相同的问题
.load_pytorch时出错   提示Syntax error in input! LexToken(NAMED_IDENTIFIER,'str',
后边build报  IndexError: list index out of range

你好，torch1.2也遇到相同的错误
--> config model
done
--> Loading model
/home/user/.local/lib/python3.6/site-packages/onnx_tf/common/__init__.py:87: UserWarning: FrontendHandler.get_outputs_names is deprecated. It will be removed in future release.. Use node.outputs instead.
  warnings.warn(message)
./fire.pt ********************
WARNING: Token 'COMMENT' defined, but not used
WARNING: There is 1 unused token
Syntax error in input! LexToken(<,'<',3,101)
done
--> Building model
W Genreate input meta fail, please check model.
W External input meta file "/tmp/tmpzzzps_ss/fire_inputmeta.yml" is not exists.
Traceback (most recent call last):
  File "/home/user/wuzhe/fire/model_trans.py", line 51, in <module>
    trans_model()
  File "/home/user/wuzhe/fire/model_trans.py", line 36, in trans_model
    ret = rknn.build(do_quantization=True, dataset='./dataset.txt')
  File "/home/user/.local/lib/python3.6/site-packages/rknn/api/rknn.py", line 238, in build
    ret = self.rknn_base.build(do_quantization=do_quantization, dataset=dataset, pack_vdata=pre_compile, batch_size=rknn_batch_size)
  File "rknn/api/rknn_base.py", line 719, in rknn.api.rknn_base.RKNNBase.build
  File "rknn/api/rknn_base.py", line 2249, in rknn.api.rknn_base.RKNNBase._generate_inputmeta
IndexError: list index out of range

jefferyzhang 发表于 2020-3-12 17:19
问题已经上报给NPU部门了，等待回复

您好 请问这个问题还没有得到回复吗？

yjcn 发表于 2020-3-27 21:52
您好 请问这个问题还没有得到回复吗？

没有，有回复我就会转过来

NPU Team回复：
该问题已经解决，下一个版本发布就可以用了。