ajhalthor / transformer-neural-network

Code Transformer neural network components piece by piece

Transformer-Neural-Network

Code Transformer neural network components piece by piece.

Full Playlist

transformer-neural-network's Issues

Word embeddings input size

Hi and thanks for the great series about transformers!

I noticed that you initialize the nn.embeddings layer for the word embeddings with an input size that is equal to the vocabulary size.
As we would like to add the positional encodings with dimensions max_seq_length X 512 on top of the embeddings, the dimensions of the words embeddings should be the same as the ones for the positional embeddings (max_seq_length X 512).

So the corrected code would look something like:

class SentenceEmbedding(nn.Module):
    "For a given sentence, create an embedding"
    def __init__(self, max_sequence_length, d_model, language_to_index, START_TOKEN, END_TOKEN, PADDING_TOKEN):
        super().__init__()
        self.vocab_size = len(language_to_index)
        self.max_sequence_length = max_sequence_length
        self.embedding = nn.Embedding(self.max_sequence_length, d_model) #CORRECTED LINE....
        ...
```

Regards,

M

Please help me

hello sir!
I am from myanmar
i wanna translate english to mynmar and so i just change myanmar vocabulary inset of kannada vocabulary and dataset.
i dont change any code.
and i found this error. please help me

IndexError: index out of range in self

Hello I am using a this repo to convert English to Italian and I keep running into this error
I kept all the code the same except for the vocab dictionaries which are shown below

english_vocabulary = [START_TOKEN, ' ', '!', '"', '#', '$', '%', '&', "'", '(', ')', '*', '+', ',', '-', '.', '/', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', ':', '<', '=', '>', '?', '@', 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', '[', '\\\\', ']', '^', '_', '',
'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l',
'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x',
'y', 'z',
'{', '|', '}', '~', '_', PADDING_TOKEN, END_TOKEN]
`

italian_vocabulary = [START_TOKEN, ' ', '!', '"', '#', '$', '%', '&', "'", '(', ')', '*', '+', ',', '-', '.', '/', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', ':', '<', '=', '>', '?', '@', 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'Z', 'À', 'È', 'É', 'Ì', 'Í', 'Ò', 'Ó', 'Ù', '[', '\\\\', ']', '^', '_', '',
'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l',
'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'z',
'à', 'è', 'é', 'ì', 'í', 'ò', 'ó', 'ù',
'{', '|', '}', '~', '_', PADDING_TOKEN, END_TOKEN]
`

Besides that, all the code is the same just updated from kn->it and the naming for Kannada -> italian

Epoch 0 Traceback (most recent call last): File "/Users/alessandrobongiorno/Desktop/singularity/research/transformers/gpt-3/train/train.py", line 213, in <module> it_predictions = transformer( File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/Users/alessandrobongiorno/Desktop/singularity/research/transformers/gpt-3/train/../model/model.py", line 301, in forward x = self.encoder(x, encoder_self_attention_mask, start_token=enc_start_token, end_token=enc_end_token) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/Users/alessandrobongiorno/Desktop/singularity/research/transformers/gpt-3/train/../model/model.py", line 178, in forward x = self.sentence_embedding(x, start_token, end_token) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/Users/alessandrobongiorno/Desktop/singularity/research/transformers/gpt-3/train/../model/model.py", line 71, in forward x = self.embedding(x) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/modules/sparse.py", line 163, in forward return F.embedding( File "/Users/alessandrobongiorno/.pyenv/versions/3.9.19/lib/python3.9/site-packages/torch/nn/functional.py", line 2264, in embedding return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse) IndexError: index out of range in self (env) alessandrobongiorno@Alessandros-Air train %
I believe the error is coming from this line

self.embedding = nn.Embedding(self.vocab_size, d_model)

If anyone has any ideas I would really appreciate it

I think we need the train.en and train.kn files

train.en and train.kn files are needed.

question

Please how can I implement this for word to word translation?

Permute for values after scale_dot_product operation

When got new values by scale_dot_product operation, it should permute values like before, to exchange max_sequence_length and num_heads dimensions. And then reshape its size. Such that we can make sure for one row of values, it comes from many heads.

values = values.permute(0, 2, 1, 3).reshape(batch_size, max_sequence_length, self.num_heads * self.head_dim)

RuntimeError: The size of tensor a (8) must match the size of tensor b (32) at non-singleton dimension 0

hey i am getting these error where tensor dimension are not in place

IndexError: index out of range in self

when i am converting from english to roman urdu, it is giving error:

`

IndexError Traceback (most recent call last)
Cell In[21], line 14
12 encoder_self_attention_mask, decoder_self_attention_mask, decoder_cross_attention_mask = create_masks(eng_batch, ur_batch)
13 optim.zero_grad()
---> 14 ur_predictions = transformer(eng_batch,
15 ur_batch,
16 encoder_self_attention_mask.to(device),
17 decoder_self_attention_mask.to(device),
18 decoder_cross_attention_mask.to(device),
19 enc_start_token=False,
20 enc_end_token=False,
21 dec_start_token=True,
22 dec_end_token=True)
23 labels = transformer.decoder.sentence_embedding.batch_tokenize(ur_batch, start_token=False, end_token=True)
24 loss = criterian(
25 ur_predictions.view(-1, ur_vocab_size).to(device),
26 labels.view(-1).to(device)
27 ).to(device)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1511, in Module._wrapped_call_impl(self, *args, **kwargs)
1509 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1510 else:
-> 1511 return self._call_impl(*args, **kwargs)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1520, in Module._call_impl(self, *args, **kwargs)
1515 # If we don't have any hooks, we want to skip the rest of the logic in
1516 # this function, and just call forward.
1517 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1518 or _global_backward_pre_hooks or _global_backward_hooks
1519 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1520 return forward_call(*args, **kwargs)
1522 try:
1523 result = None

File ~\Desktop\FYP\Web App\Meme_AI\transformer model\transformer_model.py:302, in Transformer.forward(self, x, y, encoder_self_attention_mask, decoder_self_attention_mask, decoder_cross_attention_mask, enc_start_token, enc_end_token, dec_start_token, dec_end_token)
292 def forward(self,
293 x,
294 y,
(...)
300 dec_start_token=False, # We should make this true
301 dec_end_token=False): # x, y are batch of sentences
--> 302 x = self.encoder(x, encoder_self_attention_mask, start_token=enc_start_token, end_token=enc_end_token)
303 out = self.decoder(x, y, decoder_self_attention_mask, decoder_cross_attention_mask, start_token=dec_start_token, end_token=dec_end_token)
304 out = self.linear(out)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1511, in Module._wrapped_call_impl(self, *args, **kwargs)
1509 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1510 else:
-> 1511 return self._call_impl(*args, **kwargs)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1520, in Module._call_impl(self, *args, **kwargs)
1515 # If we don't have any hooks, we want to skip the rest of the logic in
1516 # this function, and just call forward.
1517 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1518 or _global_backward_pre_hooks or _global_backward_hooks
1519 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1520 return forward_call(*args, **kwargs)
1522 try:
1523 result = None

File ~\Desktop\FYP\Web App\Meme_AI\transformer model\transformer_model.py:179, in Encoder.forward(self, x, self_attention_mask, start_token, end_token)
178 def forward(self, x, self_attention_mask, start_token, end_token):
--> 179 x = self.sentence_embedding(x, start_token, end_token)
180 x = self.layers(x, self_attention_mask)
181 return x

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1511, in Module._wrapped_call_impl(self, *args, **kwargs)
1509 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1510 else:
-> 1511 return self._call_impl(*args, **kwargs)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1520, in Module._call_impl(self, *args, **kwargs)
1515 # If we don't have any hooks, we want to skip the rest of the logic in
1516 # this function, and just call forward.
1517 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1518 or _global_backward_pre_hooks or _global_backward_hooks
1519 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1520 return forward_call(*args, **kwargs)
1522 try:
1523 result = None

File ~\Desktop\FYP\Web App\Meme_AI\transformer model\transformer_model.py:72, in SentenceEmbedding.forward(self, x, start_token, end_token)
70 def forward(self, x, start_token, end_token): # sentence
71 x = self.batch_tokenize(x, start_token, end_token)
---> 72 x = self.embedding(x)
73 pos = self.position_encoder().to(get_device())
74 x = self.dropout(x + pos)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1511, in Module._wrapped_call_impl(self, *args, **kwargs)
1509 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1510 else:
-> 1511 return self._call_impl(*args, **kwargs)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\module.py:1520, in Module._call_impl(self, *args, **kwargs)
1515 # If we don't have any hooks, we want to skip the rest of the logic in
1516 # this function, and just call forward.
1517 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1518 or _global_backward_pre_hooks or _global_backward_hooks
1519 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1520 return forward_call(*args, **kwargs)
1522 try:
1523 result = None

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\modules\sparse.py:163, in Embedding.forward(self, input)
162 def forward(self, input: Tensor) -> Tensor:
--> 163 return F.embedding(
164 input, self.weight, self.padding_idx, self.max_norm,
165 self.norm_type, self.scale_grad_by_freq, self.sparse)

File c:\Users\AAHIL ALWANI\Desktop\FYP\Web App\Meme_AI\meme\Lib\site-packages\torch\nn\functional.py:2237, in embedding(input, weight, padding_idx, max_norm, norm_type, scale_grad_by_freq, sparse)
2231 # Note [embedding_renorm set_grad_enabled]
2232 # XXX: equivalent to
2233 # with torch.no_grad():
2234 # torch.embedding_renorm_
2235 # remove once script supports set_grad_enabled
2236 no_grad_embedding_renorm(weight, input, max_norm, norm_type)
-> 2237 return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)

IndexError: index out of range in self
`

permute before reshape

Hi Ajay,
In "Multi Head Attention " section, after scaled_dot_product, it needed to reshape, but I think before the reshape, it need to permute the head dimension and the sequence dimension.
Please let me know if I'm wrong.