Modality selection for responding to prompts
A server obtains a prompt from a user device. The server determines a score representing a likelihood that the prompt matches an intent. The server selects, based on the score, a modality to generate a response to the prompt. The modality is selected from a group comprising a workflow for the intent and a generative pretrained transformer engine. The server generates the response using the modality. The server transmits the response to the user device.
1 . A method, comprising:
obtaining a prompt from a user device;
determining a score representing a likelihood that the prompt matches an intent;
selecting, based on the score not meeting a threshold, a generative pretrained transformer (GPT) engine as a modality to generate a response to the prompt, wherein the modality is selected from a group comprising a workflow for the intent and the GPT engine, and wherein the threshold is dynamically determined based on computing resources available to the GPT engine;
matching, using the GPT engine, the prompt to a second intent;
executing a second workflow corresponding to the second intent;
generating the response using the second workflow; and
transmitting the response to the user device.
2 . The method of claim 1 , wherein
the workflow for the intent is selected as the modality responsive to the score meeting the threshold.
3 . The method of claim 1 , wherein the intent is one of multiple intents, wherein selecting the GPT engine as the modality comprises:
identifying, from the multiple intents, an intent having a highest score; and
selecting the GPT engine responsive to the highest score not meeting the threshold.
4 . The method of claim 1 , comprising:
removing noise from the prompt using a noise removal filter to generate a noise-free prompt, wherein the score is determined based on the noise-free prompt.
5 . The method of claim 1 , wherein the score represents a probability that the prompt matches the intent.
6 . The method of claim 1 , wherein the score is determined using an intent matching engine, the method further comprising:
training, using online learning, the intent matching engine based on the prompt and the second intent.
7 . The method of claim 1 , further comprising:
removing noise from the prompt using a noise removal engine to generate a noise-free prompt, wherein the score is determined based on the noise-free prompt; and
training, using online learning, the noise removal engine based on an output of the GPT engine.
8 . A non-transitory computer-readable medium storing instructions operable to cause one or more processors to perform operations comprising:
obtaining a prompt from a user device;
determining a score representing a likelihood that the prompt matches an intent;
selecting, based on the score not meeting a threshold, a generative pretrained transformer (GPT) engine as a modality to generate a response to the prompt, wherein the modality is selected from a group comprising a workflow for the intent and the GPT engine, and wherein the threshold is dynamically determined based on computing resources available to the GPT engine;
matching, using the GPT engine, the prompt to a GPT-identified intent;
executing a second workflow corresponding to the GPT-identified intent;
generating the response using the second workflow; and
transmitting the response to the user device.
9 . The non-transitory computer-readable medium of claim 8 , wherein
the workflow for the intent is selected as the modality responsive to the score meeting the threshold.
10 . The non-transitory computer-readable medium of claim 8 , wherein
the score represents a probability that the prompt matches the intent.
11 . The non-transitory computer-readable medium of claim 8 , wherein the intent is one of a plurality of intents, wherein selecting the GPT engine as the modality comprises:
identifying, from the plurality of intents, an intent having a highest score; and
selecting the GPT engine responsive to the highest score not meeting the threshold.
12 . The non-transitory computer-readable medium of claim 8 , the operations comprising:
removing noise from the prompt using a noise removal engine to generate a noise-free prompt, wherein the score is determined based on the noise-free prompt.
13 . The non-transitory computer-readable medium of claim 8 , the operations further comprising:
training, using online learning and based on the prompt and the second intent, an intent matching engine for determining the score.
14 . The non-transitory computer-readable medium of claim 8 , the operations further comprising:
removing noise from the prompt using a noise removal filter to generate a noise-free prompt, wherein the score is determined based on the noise-free prompt; and
training the noise removal filter based on an output of the GPT engine.
15 . A system, comprising:
a memory subsystem; and
processing circuitry configured to execute instructions stored in the memory subsystem to:
obtain a prompt from a user device;
determine a score representing a likelihood that the prompt matches an intent;
select, based on the score, a modality to generate a response to the prompt, the modality being selected from a group comprising a workflow for the intent and a generative pretrained transformer (GPT) engine, wherein to select the modality comprises to:
select the workflow for the intent responsive to the score meeting a threshold; or
select the GPT engine responsive to the score not meeting the threshold, wherein the threshold is dynamically determined based on computing resources available to the GPT engine;
generate the response using the modality; and
transmit the response to the user device.
16 . The system of claim 15 , wherein the intent is one of multiple intents, wherein to select the modality further comprises to:
identify, from the multiple intents, an intent having a score associated with a highest probability of the prompt matching to the intent.
17 . The system of claim 15 , wherein the instructions further comprise instructions to:
remove noise from the prompt using a noise removal filter, wherein the score is determined using the prompt with the noise removed.
18 . The system of claim 17 , wherein the instructions further comprise instructions to:
train, using online learning, the noise removal filter based on an output of the GPT engine.
19 . The system of claim 15 , wherein the score represents a probability that the prompt matches the intent.
20 . The system of claim 16 , wherein to select the modality further comprises to:
select the workflow for the intent having the score associated with the highest probability responsive to the score associated with the highest probability meeting the threshold; or
select the GPT engine responsive to the score associated with the highest probability not meeting the threshold.