Prompt discovery with attention refinement
An embodiment senses a sequence, responsive to the sensed sequence, segments the sensed sequence into a prompt and text. The embodiment generates an attention embedding representative of the text and computes an attention weight based on the attention embedding. The embodiment generates a prompt embedding representative of the prompt. The embodiment computes a relationship between the prompt embedding and the attention weight comprising correlating the attention embedding and the attention weight based on a beam search; bin packing the prompt embedding where a bin is defined by the attention embedding; matching the prompt embedding to the bin based on a Minkowski distance metric; where the bin packing causes the computing of the relationship between the prompt and the text represented by the attention weight.
1 . A computer-implemented method comprising:
sensing a sequence; responsive to the sensed sequence, segmenting the sensed sequence into a prompt and text;
generating an attention embedding representative of the text and computing an attention weight based on the attention embedding;
generating a prompt embedding representative of the prompt; and
computing a relationship between the prompt embedding and the attention weight comprising:
correlating the attention embedding and the attention weight based on a beam search;
bin packing the prompt embedding wherein a bin is defined by the attention embedding; and
matching the prompt embedding to the bin based on a Minkowski distance metric; wherein the bin packing causes the computing of the relationship between the prompt and the text represented by the attention weight.
2 . The computer-implemented method of claim 1 , wherein the bin is defined by a dimension of the attention embedding.
3 . The computer-implemented method of claim 1 , wherein computing the attention weight is based on a 5W framework, the 5W framework comprising of what, where, when, why and who questions.
4 . The computer-implemented method of claim 1 , wherein the segmenting comprises zero-shot, one-shot, and few-shot learning.
5 . The computer-implemented method of claim 1 , wherein a Minkowski distance metric p value comprises of 1, 2, or 3.
6 . The computer-implemented method of claim 1 , wherein the correlating comprises selecting a highest R-squared value.
7 . The computer-implemented method of claim 1 , wherein the segmenting the sensed sequence further comprises segmenting the sensed sequence into a context wherein the context determines the attention embedding representative of the text.
8 . A computer program product comprising one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by a processor to cause the processor to perform operations comprising:
sensing a sequence; responsive to the sensed sequence, segmenting the sensed sequence into a prompt and text;
generating an attention embedding representative of the text and computing an attention weight based on the attention embedding;
generating a prompt embedding representative of the prompt; and
computing a relationship between the prompt embedding and the attention weight comprising:
correlating the attention embedding and the attention weight based on a beam search;
bin packing the prompt embedding wherein a bin is defined by the attention embedding; and
matching the prompt embedding to the bin based on a Minkowski distance metric; wherein the bin packing causes the computing of the relationship between the prompt and the text represented by the attention weight.
9 . The computer program product of claim 8 , wherein the bin is defined by a dimension of the attention embedding.
10 . The computer program product of claim 8 , wherein computing the attention weight is based on a 5W framework, the 5W framework comprising of what, where, when, why and who questions.
11 . The computer program product of claim 8 , wherein the segmenting comprises zero-shot, one-shot, and few-shot learning.
12 . The computer program product of claim 8 , wherein a Minkowski distance metric p value comprises of 1, 2, or 3.
13 . The computer program product of claim 8 , wherein the correlating comprises selecting a highest R-squared value.
14 . The computer program product of claim 8 , wherein the segmenting the sensed sequence further comprises segmenting the sensed sequence into a context wherein the context determines the attention embedding representative of the text.
15 . A computer system comprising a processor and one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by the processor to cause the processor to perform operations comprising:
sensing a sequence; responsive to the sensed sequence, segmenting the sensed sequence into a prompt and text;
generating an attention embedding representative of the text and computing an attention weight based on the attention embedding;
generating a prompt embedding representative of the prompt; and
computing a relationship between the prompt embedding and the attention weight comprising:
correlating the attention embedding and the attention weight based on a beam search;
bin packing the prompt embedding wherein a bin is defined by the attention embedding; and
matching the prompt embedding to the bin based on a Minkowski distance metric; wherein the bin packing causes the computing of the relationship between the prompt and the text represented by the attention weight.
16 . The computer system of claim 15 , wherein the bin is defined by a dimension of the attention embedding.
17 . The computer system of claim 15 , wherein computing the attention weight is based on a 5W framework, the 5W framework comprising of what, where, when, why and who questions.
18 . The computer system of claim 15 , wherein the segmenting comprises zero-shot, one-shot, and few-shot learning.
19 . The computer system of claim 15 , wherein a Minkowski distance metric p value comprises of 1, 2, or 3.
20 . The computer system of claim 15 , wherein the segmenting the sensed sequence further comprises segmenting the sensed sequence into a context wherein the context determines the attention embedding representative of the text.