SYSTEMS AND METHODS FOR UPDATING AND EXECUTING LANGUAGE MODELS USING INPUT FROM DISTRIBUTED COMPUTING ENVIRONMENTS

1 . A system, comprising:

one or more processors coupled to non-transitory memory, the one or more processors configured to:

receive, from a plurality of client devices, a plurality of prompts for a language model, each of the plurality of prompts comprising at least one nickname corresponding to an athlete, the at least one nickname corresponding to an intent relating to wagers, wherein the language model is a generative pretrained transformer (GPT) model and includes an encoder that processes input text and a decoder that generates output text and a plurality of trainable parameters;

determine that the language model has not been updated using training examples that include the at least one nickname corresponding to the intent;

generate, using the plurality of prompts and additional information corresponding to the at least one nickname, a set of training examples, each training example comprising a respective prompt having the at least one nickname;

update the language model using the set of training examples by applying one or more parameter-efficient fine-tuning adapters configured to modify less than all of the plurality of trainable parameters;

receive from a client device, a prompt including the at least one nickname; and

provide, using the updated language model, an output corresponding to the at least one nickname.

2 . The system of claim 1 , wherein the one or more processors are further configured to:

determine that the language model has not been updated using training examples that include the at least one nickname based on a plurality of second prompts received from the plurality of client devices that indicate the language model provided incorrect output.

3 . The system of claim 1 , wherein the one or more processors are further configured to:

determine, using the language model, that the plurality of prompts cannot be classified as corresponding to the intent relating to wagers; and

determine that the language model has not been updated using training examples that include the at least one nickname responsive to determining that the plurality of prompts cannot be classified as corresponding to the intent relating to wagers using the language model.

4 . The system of claim 1 , wherein the one or more processors are further configured to:

determine that a number of the plurality of prompts satisfies a threshold; and

generate the set of training examples for the language model responsive to determining that the number of the plurality of prompts satisfies the threshold.

5 . The system of claim 1 , wherein the training examples further comprise an input prompt comprising the at least one nickname and a corresponding output prompt, the corresponding output prompt indicating an intent associated with the at least one nickname.

6 . The system of claim 1 , wherein the one or more processors are further configured to:

associate the at least one nickname with semantic data determined based on the plurality of prompts received from the plurality of client devices; and

store the association in a database.

7 . The system of claim 1 , wherein the one or more processors are further configured to:

generate variations of the set of training examples based on the plurality of prompts received from the plurality of client devices, the variations comprising the at least one nickname; and

update the language model using the variations.

8 . The system of claim 7 , wherein the variations are generated according to changes in at least one of a syntactic pattern, a semantic pattern, or a contextual pattern.

9 . The system of claim 1 , wherein, in response to receiving a prompt from a client device, the one or more processors are further configured to:

retrieve information from a database based on a second prompt received from the client device of the plurality of client devices;

generate an input context for the language model using the retrieved information; and

generate, using the language model, a corresponding output prompt based on the input context.

10 . The system of claim 1 , wherein the one or more processors are further configured to:

store the set of training examples and corresponding outputs in a database; and

update the database with indications that the language model has been trained to recognize the at least one nickname.

11 . A method, comprising:

receiving, by one or more processors coupled to non-transitory memory, from a plurality of client devices, a plurality of prompts for a language model, each of the plurality of prompts comprising at least one nickname corresponding to an athlete and corresponding to an intent relating to wagers, wherein the language model is a generative pretrained transformer (GPT) model and includes an encoder that processes input text and a decoder that generates output text and a plurality of trainable parameters;

determining, by the one or more processors, that the language model has not been updated using training examples that include the at least one nickname or slang term corresponding to the intent;

generating, by the one or more processors, using the plurality of prompts and additional information corresponding to the at least one nickname, a set of training examples, each training example comprising a respective prompt having the at least one nickname;

updating, by the one or more processors, the language model using the set of training examples by applying one or more parameter-efficient fine-tuning adapters configured to modify less than all of the plurality of trainable parameters;

receiving from a client device, a prompt including the at least one nickname; and

providing, using the updated language model, an output corresponding to the at least one nickname.

12 . The method of claim 11 , further comprising:

determining, by the one or more processors, that the language model has not been updated using training examples that include the at least one nickname based on a plurality of second prompts received from the plurality of client devices that indicate the language model provided incorrect output.

13 . The method of claim 11 , further comprising:

determining, by the one or more processors, using the language model, that the plurality of prompts cannot be classified as corresponding to the intent relating to wagers; and

determining, by the one or more processors, that the language model has not been updated using training examples that include the at least one nickname responsive to determining that the plurality of prompts cannot be classified as corresponding to the intent relating to wagers using the language model.

14 . The method of claim 11 , further comprising:

determining, by the one or more processors, that a number of the plurality of prompts satisfies a threshold; and

generating, by the one or more processors, the set of training examples for the language model responsive to determining that the number of the plurality of prompts satisfies the threshold.

15 . The method of claim 11 , wherein the training examples further comprise an input prompt comprising the at least one nickname and a corresponding output prompt, the corresponding output prompt indicating an intent associated with the at least one nickname.

16 . The method of claim 11 , further comprising:

associating, by the one or more processors, the at least one nickname with semantic data determined based on the plurality of prompts received from the plurality of client devices; and

storing, by the one or more processors, the association in a database.

17 . The method of claim 11 , further comprising:

generating, by the one or more processors, variations of the set of training examples based on the plurality of prompts received from the plurality of client devices, the variations comprising the at least one nickname; and

updating, by the one or more processors, the language model using the variations.

18 . The method of claim 17 , wherein the variations are generated according to changes in at least one of a syntactic pattern, a semantic pattern, or a contextual pattern.

19 . The method of claim 11 , further comprising, in response to receiving a prompt from a client device:

retrieving, by the one or more processors, information from a database based on a second prompt received from the client device of the plurality of client devices;

generating, by the one or more processors, an input context for the language model using the retrieved information; and

generating, by the one or more processors, using the language model, a corresponding output prompt based on the input context.

20 . The method of claim 11 , further comprising:

storing, by the one or more processors, the set of training examples and corresponding outputs in a database; and

updating, by the one or more processors, the database with indications that the language model has been trained to recognize the at least one nickname.