IP Library Granted Patent US 12,705,265
Granted Patent B2
US 12,705,265 · App. 18/244,530 · Granted Aug 11, 2026

Computer network monitoring and control using a fine-tuned language model

Inventors: Jean-Philippe Vasseur (Combloux, FR); Grégory Mermoud (Venthône, CH); Pierre-André Savalle (Rueil-Malmaison, FR)
Assignee: Cisco Technology, Inc.
G06F16/3329G06F40/253
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,705,265
App. No.
18/244,530
Filed
Sep 11, 2023
Granted
Aug 11, 2026
Kind
B2
Examiner
GAY, SONIA L
Art Unit
2657
USPC
704/9
Abstract

In one implementation, a device generates a response to user input using a first language model. The device determines that the response is an erroneous response. The device generates a resolution to the erroneous response using a teacher language model. The device updates the first language model using the resolution from the teacher language model.

Claims (47)

1 . A method comprising:

generating, by a device, a response to user input using a first language model;

determining, by the device, that the response is an erroneous response;

making, by the device, a determination whether to delay querying a teacher language model regarding the erroneous response until after handling of the user input has completed;

querying, by the device and after handling of the user input has completed when the determination is to delay, the teacher language model based on the erroneous response in order to obtain a resolution to the erroneous response;

generating, by the device and based on an output from querying the teacher language model regarding the erroneous response, a resolution to the erroneous response including a corrected version of the erroneous response; and

updating, by the device, the first language model using the resolution from the teacher language model.

2 . The method as in claim 1 , further comprising:

blocking, by the device, the first language model from providing the erroneous response to a user interface associated with the user input.

3 . The method as in claim 1 , wherein determining that the response is an erroneous response comprises:

detecting a syntax error or incorrect parameter within the response.

4 . The method as in claim 1 , wherein generating the resolution to the erroneous response using the teacher language model comprises:

receiving, at the device, feedback from an expert regarding the resolution.

5 . The method as in claim 1 , wherein the determination whether to delay querying the teacher language model is based on one or more of an identification of the user or a current load for the teacher model.

6 . The method as in claim 1 , wherein the response comprises a chart or plot of a network generated by the first language model based on networking telemetry from that network.

7 . The method as in claim 1 , wherein the response comprises a command line interface (CLI) command for a networking element in a network.

8 . The method as in claim 1 , wherein the response comprises one or more of an analysis of a packet trace from a network or an application programming interface (API) query.

9 . The method as in claim 1 , wherein updating the first language model includes using the resolution to augment subsequent contexts provided to the first language model.

10 . The method as in claim 1 , wherein the first language model is configured to perform a monitoring or control action in a computer network.

11 . An apparatus, comprising:

one or more network interfaces;

a processor coupled to the one or more network interfaces and configured to execute one or more processes; and

a memory configured to store a process that is executable by the processor, the process when executed configured to:

generate a response to user input using a first language model;

determine that the response is an erroneous response;

make a determination whether to delay querying a teacher language model regarding the erroneous response until after handling of the user input has completed;

query, after handling of the user input has completed when the determination is to delay, the teacher language model based on the erroneous response in order to obtain a resolution to the erroneous response;

generate, based on an output from querying the teacher language model regarding the erroneous response, a resolution to the erroneous response including a corrected version of the erroneous response; and

update the first language model using the resolution from the teacher language model.

12 . The apparatus as in claim 11 , wherein the process when executed is further configured to:

block the first language model from providing the erroneous response to a user interface associated with the user input.

13 . The apparatus as in claim 11 , wherein the apparatus determines that the response is an erroneous response by:

detecting a syntax error or incorrect parameter within the response.

14 . The apparatus as in claim 11 , wherein the apparatus generates the resolution to the erroneous response using the teacher language model by:

receiving feedback from an expert regarding the resolution.

15 . The apparatus as in claim 11 , wherein the teacher language model is larger in size than that of the first language model.

16 . The apparatus as in claim 11 , wherein the response comprises a chart or plot of a network generated by the first language model based on networking telemetry from that network.

17 . The apparatus as in claim 11 , wherein the response comprises a command line interface (CLI) command for a networking element in a network.

18 . The apparatus as in claim 11 , wherein the response comprises an analysis of a packet trace from a network.

19 . The apparatus as in claim 11 , wherein the response comprises an application programming interface (API) query.

20 . A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:

generating, by the device, a response to user input using a first language model;

determining, by the device, that the response is an erroneous response;

making, by the device, a determination whether to delay querying a teacher language model regarding the erroneous response until after handling of the user input has completed;

querying, by the device and after handling of the user input has completed when the determination is to delay, the teacher language model based on the erroneous response in order to obtain a resolution to the erroneous response;

generating, by the device and based on an output from querying the teacher language model regarding the erroneous response, a resolution to the erroneous response including a corrected version of the erroneous response; and

updating, by the device, the first language model using the resolution from the teacher language model.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Sep 11, 2023
From: VASSEUR, JEAN-PHILIPPE; MERMOUD, GRÉGORY; SAVALLE, PIERRE-ANDRÉ
To: CISCO TECHNOLOGY, INC.
Reel/Frame 064860/0763 →
Continuity (1)
Related Publication 20250086205A1 · Mar 13, 2025
References Cited (21)
US 10482182B1 · Jankowski, Jr. · 2019 [cited by examiner]
US 11341326B2 · Wang et al. · 2022 [cited by applicant]
US 11620515B2 · Liu et al. · 2023 [cited by applicant]
US 11900921B1 · Gupta · 2024 [cited by examiner]
US 20180364975A1 · Kwong · 2018 [cited by examiner]
US 20190130251A1 · Lao · 2019 [cited by examiner]
US 20210224660A1 · Song et al. · 2021 [cited by applicant]
US 20220012637A1 · Rezazadegan Tavakoli et al. · 2022 [cited by applicant]
US 20220335303A1 · Haidar et al. · 2022 [cited by applicant]
US 20240056361A1 · Bai · 2024 [cited by examiner]
US 20240137289A1 · Wu · 2024 [cited by examiner]
US 20240319970A1 · Watson · 2024 [cited by examiner]
US 20240354423A1 · Mardikar · 2024 [cited by examiner]
US 20240362093A1 · Zhou · 2024 [cited by examiner]
US 20240362468A1 · Lee · 2024 [cited by examiner]
US 20240428079A1 · Chen · 2024 [cited by examiner]
US 20250005058A1 · Khosla · 2025 [cited by examiner]
US 20250036375A1 · Hsieh · 2025 [cited by examiner]
US 20250077551A1 · Simaria · 2025 [cited by examiner]
MIST WAN Assurance Overview, 2023, MIST (Year: 2023). [cited by examiner]
“Knowledge distillation”, online: https://en.wikipedia.org/wiki/Knowledge_distillation, accessed Sep. 11, 2023, 4 pages, Wikimedia Foundation, Inc. [cited by applicant]