lstm decoder : dimension must be equal

[DorianKodelja]

If i'm using tensorflow-gpu 1.8.0, the other issue I raised (concerning AUTO_REUSE) doesn't appear.
Yet I have another error in lstm_decoder():
ValueError: Dimensions must be equal, but are 740 and 640 for 'LSTMD/basic_lstm_cell/MatMul' (op: 'MatMul') with input shapes: [?,740], [640,400].

It occurs at
(cell_output, (c_state, h_state)) = lstm_cell(lstm_outputs[:, time_step, :], (c_state, h_state))
because it's expecting h_state to be of dimension (?,100) but it is in fact of dimension (?,200) because of the concatenation with the previous prediction at the preceding line
h_state = tf.concat([h_state, tag_pre], -1) # (?, 200)

[ponnyhdworker]

Thanks for your awesome work. I read the paper carefully, but I don't know how to solve this problem based on the TLSTM1 formulas, can you give me a solution?

[Lc]

use tensorflow==1.4.0 can solve this problom

[AshleyCry]

Just delete the CKPT folder

[Yaoyu]

this problem occor.How should I do??
W tensorflow/core/framework/op_kernel.cc:1502] OP_REQUIRES failed at save_restore_v2_ops.cc:184 : Not found: Key LSTMD2/basic_lstm_cell/bias not found in checkpoint