Daniel Voigt Godoy - Deep Learning with PyTorch Step-by-Step A Beginner’s Guide-leanpub
Next, we shift our focus to the self-attention mechanism on the right:• It is the second data point's turn to be the "query" (Q), being paired with both"keys" (K), generating attention scores and a context vector, resulting in thesecond "hidden state":Equation 9.12 - Context vector for second input (x 1 )As you probably already noticed, the context vector (and thusthe "hidden state") associated with a data point is basically afunction of the corresponding "query" (Q), and everything else("keys" (K), "values" (V), and the parameters of the self-attentionmechanism) is held constant for all queries.Therefore, we can simplify a bit our previous diagram and depict only one selfattentionmechanism, assuming it will be fed a different "query" (Q) every time.Figure 9.25 - Encoder with self-attentionSelf-Attention | 743
The alphas are the attention scores, and they are organized as follows in thealphas attribute (as we’ve already seen in the "Visualizing Attention" section):Equation 9.13 - Attention scoresFor the encoder, the shape of the alphas attribute is given by (N, L source , L source ) sinceit is looking at itself.Even though I’ve described the process as if it were sequential,these operations can be parallelized to generate all "hiddenstates" at once, which is much more efficient than using arecurrent layer that is sequential in nature.We can also use an even more simplified diagram of the encoder that abstracts thenitty-gritty details of the self-attention mechanism.Figure 9.26 - Encoder with self-attention (diagram)The code for our encoder with self-attention is actually quite simple since most ofthe moving parts are inside the attention heads:744 | Chapter 9 — Part II: Sequence-to-Sequence
- Page 718 and 719: Let’s pretend for a moment that t
- Page 720 and 721: to initialize the hidden state and
- Page 722 and 723: predictions in previous steps have
- Page 724 and 725: the second set of predicted coordin
- Page 726 and 727: Let’s create an instance of the m
- Page 728 and 729: Model Configuration & TrainingThe m
- Page 730 and 731: Sure, we can!AttentionHere is a (no
- Page 732 and 733: based on "the" and "zone," I’ve j
- Page 734 and 735: Figure 9.12 - Matching a query to t
- Page 736 and 737: Outputtensor([[[ 0.0832, -0.0356],[
- Page 738 and 739: utmost importance for the correct i
- Page 740 and 741: Its formula is:Equation 9.3 - Cosin
- Page 742 and 743: second hidden state contributes to
- Page 744 and 745: Outputtensor([[[ 0.5475, 0.0875, -1
- Page 746 and 747: alphas = F.softmax(scaled_products,
- Page 748 and 749: Outputtensor([[[ 0.2138, -0.3175]]]
- Page 750 and 751: Attention Mechanism1 class Attentio
- Page 752 and 753: "Why would I want to force it to do
- Page 754 and 755: 1 Sets attention module and adjusts
- Page 756 and 757: encdec = EncoderDecoder(encoder, de
- Page 758 and 759: fig = sbs_seq_attn.plot_losses()Fig
- Page 760 and 761: Figure 9.20 - Attention scoresSee?
- Page 762 and 763: Wide vs Narrow AttentionThis mechan
- Page 764 and 765: "What’s so special about it?"Even
- Page 766 and 767: Once again, the affine transformati
- Page 770 and 771: Encoder + Self-Attention1 class Enc
- Page 772 and 773: Figure 9.27 - Encoder with self- an
- Page 774 and 775: The figure below depicts the self-a
- Page 776 and 777: shifted_seq = torch.cat([source_seq
- Page 778 and 779: Equation 9.17 - Decoder’s (masked
- Page 780 and 781: At evaluation / prediction time we
- Page 782 and 783: Outputtensor([[[0.4132, 0.3728],[0.
- Page 784 and 785: Figure 9.33 - Encoder + decoder + a
- Page 786 and 787: 64 return outputsThe encoder-decode
- Page 788 and 789: Figure 9.34 - Losses—encoder + de
- Page 790 and 791: curse. On the one hand, it makes co
- Page 792 and 793: "Are we done now? Is this good enou
- Page 794 and 795: Figure 9.46 - Consistent distancesA
- Page 796 and 797: Let’s recap what we’ve already
- Page 798 and 799: Let’s see it in code:max_len = 10
- Page 800 and 801: Let’s put it all together into a
- Page 802 and 803: Outputtensor([[[-1.0000, 0.0000],[-
- Page 804 and 805: Decoder with Positional Encoding1 c
- Page 806 and 807: Visualizing PredictionsLet’s plot
- Page 808 and 809: Next, we’re moving on to the thre
- Page 810 and 811: Data Generation & Preparation1 # Tr
- Page 812 and 813: 59 self.trg_masks)60 else:61 # Deco
- Page 814 and 815: Model Configuration1 class EncoderS
- Page 816 and 817: 1617 @property18 def alphas(self):1
The alphas are the attention scores, and they are organized as follows in the
alphas attribute (as we’ve already seen in the "Visualizing Attention" section):
Equation 9.13 - Attention scores
For the encoder, the shape of the alphas attribute is given by (N, L source , L source ) since
it is looking at itself.
Even though I’ve described the process as if it were sequential,
these operations can be parallelized to generate all "hidden
states" at once, which is much more efficient than using a
recurrent layer that is sequential in nature.
We can also use an even more simplified diagram of the encoder that abstracts the
nitty-gritty details of the self-attention mechanism.
Figure 9.26 - Encoder with self-attention (diagram)
The code for our encoder with self-attention is actually quite simple since most of
the moving parts are inside the attention heads:
744 | Chapter 9 — Part II: Sequence-to-Sequence