CTFR 18/623,195 CTFR 96546 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Introduction This office action is in response to communications filed 05/12/2026. Claims 1-25 are pending and likewise have been examined. Information Disclosure Statement The information disclosure statement (IDS) submitted on 03/03/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner. Response to Amendment Amendment filed 05/12/2026 ahs been fully considered by examiner. Response to Arguments Applicant’s arguments, see Remarks, filed 05/12/2026, with respect to the rejection(s) of claim(s) 1-25 under 35 U.S.C. 101 have been fully considered but they are not persuasive. Examiner believes the claimed limitations are directed towards an abstract idea without significantly more. Applicant argues that Independent Claims 1, 8, 15, 22 and 24 do not recite an abstract idea, and the deployment of the AI on the edge devices could not be done mentally. Examiner argues that the limitations of the independent claims do recite and abstract idea. The generation of the model and data descriptions, AI task capacity profile and selection, are not limited to being more than specifications, which a human could work on using pen and paper. While the claims may indent for a computer to do these functions, the broadest reasonable interpretation of the claim includes a human doing these functions. The deployment of the model is not actually claimed. The final limitation recites the intended use of deploying the model, but the limitation is a selection limitation, would a human could do. Applicant cites the specification to show that the claims recites computer-implemented AI deployment operations and not merely mental observation and judgment. Examiner argues that although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns , 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). For these reasons Examiner believes the independent claims contain an abstract idea, specifically a mental process. Examiner believes the abstract ideas in the independent claims are not integrated into a practical application. Applicant argues that the portions cited in the specification provide a practical application of deploying AI models to edge devices efficiently. Examiner argues that although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns , 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). Examiner acknowledges that the cited portions of the specification could provide a practical application for the claimed limitations. However, these limitations are not in the claims. The last limitations recites an intended use for deploying the model, but the limitation is just a selection, the deployment of the AI is not explicitly recited in a form other than intended use. For these reasons Examiner believes the claims are not integrated into a practical application. Examiner believes the Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Applicant argues the specifications, capacity profile, operating system constraints, and supported formats are additional elements that amount to more than the judicial exception. Applicant cites the specification to show in further detail the claimed limitations. Examiner argues that the claimed limitations of specifications, capacity profile, operating system constraints, and supported formats are not limited to the extent that they cannot be interpreted as generic specifications and requirements that could be written down on paper and compared mentally. Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns , 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993). For these reasons Examiner believes the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. 07-38-02 AIA Applicant’s arguments, see Remarks, Pg 18-19 , filed 05/12/2026 , with respect to the rejection(s) of claim(s) 1-3 under 35 U.S.C. 102 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Sindwani et al. (US 12541687 B1) . 07-38-02 AIA Applicant’s arguments, see Remarks, Pg 19-21 , filed 05/12/2026 , with respect to the rejection(s) of claim(s) 4-25 under 35 U.S.C. 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Sindwani et al. (US 12541687 B1) . Claim Rejections - 35 USC § 101 07-04-01 AIA 07-04 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Independent claims 1, 8 and 15 recite the limitations “comprising: receiving a text-based service request for an artificial intelligence (AI) model for an edge device;”, “generating model and data descriptions using the text-based service request, wherein generating the model and data descriptions comprises translating requirements, prompts, and other requested features in the text-based service request into the model and data descriptions;”, “generating an AI task capacity profile, wherein the Al task capacity profile is generated based on a capacity profile of the edge device, and wherein the capacity profile of the edge device comprises operating system constraints and at least one of supported Al model formats of the edge device and available inference engines of the edge device;”, “and selecting a resource-optimal AI model for deployment on the edge device based on the AI task capacity profile”. Claim 1 also recites “A computer-implemented method comprising:”. Claim 8 also recites “A system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:”. Claim 15 also recites “a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform operations comprising:”. The limitations “comprising: receiving a text-based service request for an artificial intelligence (AI) model for an edge device;”, “generating model and data descriptions using the text-based service request, wherein generating the model and data descriptions comprises translating requirements, prompts, and other requested features in the text-based service request into the model and data descriptions;”, “generating an AI task capacity profile, wherein the Al task capacity profile is generated based on a capacity profile of the edge device, and wherein the capacity profile of the edge device comprises operating system constraints and at least one of supported Al model formats of the edge device and available inference engines of the edge device;”, “and selecting a resource-optimal AI model for deployment on the edge device based on the AI task capacity profile” as drafted, covers a mental process, as this could be done mentally or by hand with pen and paper. This judicial exception is not integrated into a practical application. Claim 1 recites “A computer-implemented method comprising”. Claim 8 recites “A system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising”. Claim 15 recites “a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform operations comprising”. These limitations direct towards using a computer for the method, and do not impose any meaningful limits on practicing the abstract idea. Claim 1, 8 and 15 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The addition of the generic computer components recited above with regard to Claims 1, 8 and 15, does not amount to more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Claims 1, 8 and 15 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent claims 2, 9 and 16 recites the additional limitations “wherein the text-based service request comprises a description of an AI task, a description of an AI model architecture, a description of an input to the AI model, a description of an output of the AI model, an example of a deployment scenario of the AI model, an example of a specific use-case for the AI model, an example of a re-use of the AI model, a list of performance requirements of the AI model, or a list of generative prompts to the AI model”. These limitations cover mental processes, as they could be done mentally or by hand with pen and paper. These judicial exceptions are not integrated into a practical application. Claims 2, 9 and 16 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 2, 9 and 16 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claims 3, 10 and 17 recite the additional limitations “wherein generating(“the operations to generate” in claims 10 and 17) the model and data descriptions using the text-based service request further comprises: providing the text-based service request to a pre-trained large language model as input;”, “and generating the model and data descriptions using results received from the pre-trained large language model.” These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. Specifically, the act of providing data to a model and doing something using results from a model, is still an abstract idea. The use of the model is addressed below. These judicial exceptions are not integrated into a practical application Claims 3 10 and 17 recite “generating the model and data descriptions using results received from the pre-trained large language model”. These limitations direct towards using a computer for the method, and do not impose any meaningful limits on practicing the abstract idea. Claims 3, 10 and 17 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The addition of the generic computer components recited above with regard to Claims 3, 10 and 17, does not amount to more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Claims 3, 10 and 17 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claim 4, 11 and 18 recite the additional limitations “wherein generating(“the operations to generate” in claims 11 and 18) the AI task capacity profile further comprises: retrieving a capacity profile of the edge device;”, “identifying performance and resource parameters by comparing the capacity profile of the edge device to an AI model requirements mapping;”, “and generating the AI task capacity profile using the performance and resource parameters.”. These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. These judicial exceptions are not integrated into a practical application. Claims 4, 11 and 18 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 4, 11 and 18 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claims 5, 12 and 19 recite the additional limitations “wherein the AI task capacity profile comprises a compatibility list that comprises hardware and software mismatches between a potential AI model and edge device or potential bottlenecks in memory, CPU, GPU, or software infrastructure of the edge device.”. These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. These judicial exceptions are not integrated into a practical application. Claims 5, 12 and 19 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 5, 12 and 19 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claims 6, 13 and 20 recite the additional limitations “wherein selecting(“the operations to select” in claims 13 and 20) the resource-optimal AI model for deployment on the edge device based on the AI task capacity profile further comprises: identifying an AI model family using the model and data descriptions and the AI task capacity profile;”, “and selecting a model variant of the AI model family based on the AI task capacity profile and resources of the edge device.”. These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. These judicial exceptions are not integrated into a practical application. Claims 6, 13 and 20 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 6, 13 and 20 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claims 7, 14 and 21 recite the additional limitations “wherein the model variant is a compressed, pruned, or quantized AI model to correspond to resources of the edge device. These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. These judicial exceptions are not integrated into a practical application. Claims 7, 14 and 21 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 7, 14 and 21 do not contain any additional limitations. The claims as drafted, are not patent eligible. Independent Claims 22 and 24 recite the limitations “receiving a service request for an artificial intelligence (AI) model for an edge device;”, “generating model and data specifications by using automated generative translations of the service request, wherein using the automated generative translations comprises translating requirements, prompts, and other requested features in the service request into the model and data specifications;”, “performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model, wherein the capacity profile of the edge device comprises operating system constraints and at least one of supported AI model formats of the edge device and available inference engines of the edge device;”, “and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter”. Claim 22 also recites “A computer-implemented method comprising:”. Claim 24 also recites “A system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:”. The limitations “receiving a service request for an artificial intelligence (AI) model for an edge device;”, “generating model and data specifications by using automated generative translations of the service request, wherein using the automated generative translations comprises translating requirements, prompts, and other requested features in the service request into the model and data specifications;”, “performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model, wherein the capacity profile of the edge device comprises operating system constraints and at least one of supported AI model formats of the edge device and available inference engines of the edge device;”, “and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter”, as drafted, covers a mental process, as this could be done mentally or by hand with pen and paper. This judicial exception is not integrated into a practical application. Claim 22 recites “A computer-implemented method comprising”. Claim 24 recites “A system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising”. These limitations direct towards using a computer for the method, and do not impose any meaningful limits on practicing the abstract idea. Claims 22 and 24 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claims 22 and 24 do not contain any additional limitations. The claims as drafted, are not patent eligible. Dependent Claims 23 and 25 recite the additional limitations “wherein generating(“the operations to generate” in claim 25) the model and data specifications by using the automated generative translations of the service request further comprise: providing the service request to a pre-trained large language model as input;”, “and generating the model and data specifications using results generated by the pre-trained large language model.”. These limitations cover a mental processes, as they could be done mentally or by hand with pen and paper. Specifically, the act of providing data to a model and doing something using results from a model, is still an abstract idea. The use of the model is addressed below. These judicial exceptions are not integrated into a practical application Claims 23 and 25 recite “and generating the model and data specifications using results generated by the pre-trained large language model”. These limitations direct towards using a computer for the method, and do not impose any meaningful limits on practicing the abstract idea. Claims 23 and 25 do not contain any additional limitations. The Claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The addition of the generic computer components recited above with regard to Claims 23 and 25, does not amount to more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Claims 23 and 25 do not contain any additional limitations. The claims as drafted, are not patent eligible. Claim Rejections - 35 USC § 103 07-20-aia AIA The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. 07-21-aia AIA Claim (s) 1-3 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al “Foundation Model Based Native AI Framework in 6G with Cloud-Edge-End Collaboration”, hereinafter Chen, and further in view of Sindwani et al. (US 12541687 B1) . Regarding Claim 1: Chen teaches a computer-implemented method comprising: receiving a text-based service request for an artificial intelligence (AI) model for an edge device(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); generating model and data descriptions using the text-based service request, wherein generating the model and data descriptions comprises translating requirements, prompts, and other requested features in the text-based service request into the model and data descriptions(Pg 4, Col 1, Para 2, Ln 1-1, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Then, an expert knowledge graph can be established with a tree structure based on the existing structured data knowledge, as illustrated in Fig. 2. See Fig 2 . Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Pg 4, Col 2, Para 3, Adaptor Tuning, Ln 1-8, The adaptor tuning method entails the insertion of additional trainable parameters at each layer of the PFM. These parameters are proficient in modifying the outputs to better align with specific tasks without altering the fundamental structure or weights of the model. This strategy aims to retain the majority of pre-trained knowledge while permitting a degree of fine-tuning to enhance performance on specific tasks .); generating an AI task capacity profile, wherein the Al task capacity profile is generated based on a capacity profile of the edge device, and wherein the capacity profile of the edge device comprises……and at least one of supported Al model formats of the edge device and available inference engines of the edge device(Pg 4, Col 2, Para 5, Ln 1-6, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices . See Fig 3, Edge, traditional algs toolkit, tool AI models toolkit); and selecting a resource-optimal AI model for deployment on the edge device based on the AI task capacity profile(Pg 4, Col 2, Para 5, Ln 1-15, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices……..resource scheduling and task orchestration for a cell have shifted from its edge server to the cloud-based intent aware PFM within this framework . See Pg 5, Fig 3, Edge AI models toolkit. Pg 5, Col 2, Para 1, Ln 1-15, AI models stored in the algorithm toolkits are orchestrated by the well trained PFM to handle tasks across multiple edge/end-user devices……Through intent recognition and unified orchestration, these models are assigned to tasks that align with their capabilities, enabling them to effectively leverage the relationships between tasks ). Chen does not teach and wherein the capacity profile of the edge device comprises operating system constraints. In the same field of edge computing, Sindwani teaches and wherein the capacity profile of the edge device comprises operating system constraints(Col 12, Ln 44-60, execute respective ones of the plurality of useable versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations of the useable versions of the trained ML model and the plurality of different compute instance types. Compute instance types may include features and/or capabilities of specific or particular compute resources for running the trained ML model. In some aspects, different compute instance types may include different virtual machine instance types specified based on an operating system type ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify Chen with the edge computing system of Sindwani, as it helps minimize cost(Col 2, Ln 25-30). Regarding Claim 2: The combination of Chen and Sindwani teaches the computer-implemented method of claim 1, and Chen teaches wherein the text-based service request comprises a description of an AI task, a description of an AI model architecture, a description of an input to the AI model, a description of an output of the AI model, an example of a deployment scenario of the AI model, an example of a specific use-case for the AI model, an example of a re-use of the AI model, a list of performance requirements of the AI model, or a list of generative prompts to the AI model(Pg 5, Fig 3: Application 1, Application 2). Regarding Claim 3: The combination of Chen and Sindwani teaches the computer-implemented method of claim 1, and Chen teaches wherein generating the model and data descriptions using the text-based service request further comprises: providing the text-based service request to a pre-trained large language model as input(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); and generating the model and data descriptions using results received from the pre-trained large language model(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Also See Pg 5 Fig 3, library of expert knowledge & task data -> PFM) . 07-22-aia AIA Claim (s) 4 and 5 is/are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Chen and Sindwani as applied to claim 1 above, and further in view of Xu et al. “Joint Foundation Model Caching and Inference of Generative AI Services for Edge Intelligence”, hereinafter Xu . Regarding Claim 4: The combination of Chen and Sindwani teaches the computer-implemented method of claim 1, and Chen teaches wherein generating the AI task capacity profile further comprises: retrieving a capacity profile of the edge device(Pg 4, Col 2, Para 5, Ln 1-15, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices ); The combination of Chen and Sindwani does not specifically teach identifying performance and resource parameters by comparing the capacity profile of the edge device to an AI model requirements mapping; and generating the AI task capacity profile using the performance and resource parameters. In the same field of AI Edge Computing, Xu teaches identifying performance and resource parameters by comparing the capacity profile of the edge device to an AI model requirements mapping(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n ); and generating the AI task capacity profile using the performance and resource parameters(Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Sindwani with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12). Regarding Claim 5: The combination of Chen and Sindwani teaches the computer-implemented method of claim 1, but does not specifically teach wherein the AI task capacity profile comprises a compatibility list that comprises hardware and software mismatches between a potential AI model and edge device or potential bottlenecks in memory, CPU, GPU, or software infrastructure of the edge device. In the same field of AI Edge Computing, Xu teaches wherein the AI task capacity profile comprises a compatibility list that comprises hardware and software mismatches between a potential AI model and edge device or potential bottlenecks in memory, CPU, GPU, or software infrastructure of the edge device(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Sindwani with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12) . 07-22-aia AIA Claim (s) 6-7 is/are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Chen and Sindwani as applied to claim 1 above, and further in view of Zawish et al. “Complexity-Driven Model Compression for Resource-Constrained Deep Learning on Edge” . Regarding Claim 6: The combination of Chen and Sindwani teaches the computer-implemented method of claim 1, and Chen teaches wherein selecting the resource-optimal AI model for deployment on the edge device based on the AI task capacity profile further comprises: identifying an AI model family using the model and data descriptions and the AI task capacity profile(Pg 5, Col 2, Para 1, Ln 1-15, AI models stored in the algorithm toolkits are orchestrated by the well trained PFM to handle tasks across multiple edge/end-user devices……Through intent recognition and unified orchestration, these models are assigned to tasks that align with their capabilities, enabling them to effectively leverage the relationships between tasks. Pg 4, Col 2, Para 5, Ln 1-6, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices. Pg 4, Col 1, Para 2, Ln 1-8, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods ); The combination of Chen and Sindwani does not teach and selecting a model variant of the AI model family based on the AI task capacity profile and resources of the edge device. In the same field of AI Edge Computing, Zawish teaches and selecting a model variant of the AI model family based on the AI task capacity profile and resources of the edge device(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13, reduction in memory of a CNN is critical when the aim is to achieve both computation and energy efficiency ….. Moreover, in order for the deep models to run at the edge, they must fit within the target device’s RAM without disrupting the IoT application at the runtime. To achieve this, the memory-based complexity of each convolutional layer k can be calculated using. Pg 3894, Col 1, Para 1, Ln 1-2, model M must be either less than or equal to the desired complexityCr). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Sindwani with the Pruning methods of Zawish, as it can improve computation and energy efficiency of the model(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13). Regarding Claim 7: The combination of Chen, Sindwani and Zawish teaches the computer-implemented method of claim 6, but does not teach wherein the model variant is a compressed, pruned, or quantized AI model to correspond to resources of the edge device. In the same field of AI Edge Computing, Zawish teaches wherein the model variant is a compressed, pruned, or quantized AI model to correspond to resources of the edge device(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13, reduction in memory of a CNN is critical when the aim is to achieve both computation and energy efficiency ….. Moreover, in order for the deep models to run at the edge, they must fit within the target device’s RAM without disrupting the IoT application at the runtime. To achieve this, the memory-based complexity of each convolutional layer k can be calculated using. Pg 3894, Col 1, Para 1, Ln 1-2, model M must be either less than or equal to the desired complexityCr). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen, Sindwani and Zawish with the Pruning methods of Zawish, as it can improve computation and energy efficiency of the model(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13) . 07-21-aia AIA Claim (s) 8-10 and 15-17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen, and further in view of Persia et al. (US 20220327442 A1), and further in view of Sindwani . Regarding Claim 8: Chen teaches receiving a text-based service request for an artificial intelligence (AI) model for an edge device(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); generating model and data descriptions using the text-based service request, wherein generating the model and data descriptions comprises translating requirements, prompts, and other requested features in the text-based service request into the model and data descriptions(Pg 4, Col 1, Para 2, Ln 1-1, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Then, an expert knowledge graph can be established with a tree structure based on the existing structured data knowledge, as illustrated in Fig. 2. See Fig 2 . Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Pg 4, Col 2, Para 3, Adaptor Tuning, Ln 1-8, The adaptor tuning method entails the insertion of additional trainable parameters at each layer of the PFM. These parameters are proficient in modifying the outputs to better align with specific tasks without altering the fundamental structure or weights of the model. This strategy aims to retain the majority of pre-trained knowledge while permitting a degree of fine-tuning to enhance performance on specific tasks .); generating an AI task capacity profile, wherein the AI task capacity profile is generated based on a capacity profile of the edge device, and wherein the capacity profile of the edge device comprises……and at least one of supported AI model formats of the edge device and available inference engines of the edge device; (Pg 4, Col 2, Para 5, Ln 1-6, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices. See Fig 3, Edge, traditional algs toolkit, tool AI models toolkit); and selecting a resource-optimal AI model for deployment on the edge device based on the AI task capacity profile(Pg 4, Col 2, Para 5, Ln 1-15, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices……..resource scheduling and task orchestration for a cell have shifted from its edge server to the cloud-based intent aware PFM within this framework . See Pg 5, Fig 3, Edge AI models toolkit. Pg 5, Col 2, Para 1, Ln 1-15, AI models stored in the algorithm toolkits are orchestrated by the well trained PFM to handle tasks across multiple edge/end-user devices……Through intent recognition and unified orchestration, these models are assigned to tasks that align with their capabilities, enabling them to effectively leverage the relationships between tasks ). Chen does not explicitly teach a system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising. Persia teaches a system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising(Para [0083], Ln 1-14, Processor 620 is implemented in hardware, firmware, or a combination of hardware and software. In some implementations, processor 620 includes one or more processors capable of being programmed to perform a function. Memory 630 includes a random access memory ): It would have been obvious for one skilled in the art, at the effective time of filing, to modify Chen with the computer components of Persia, as it provides an environment for the system to be realized(Para [0082], Ln 1-13, Para [0083], Ln 1-14). The combination of Chen and Persia does not teach and wherein the capacity profile of the edge device comprises operating system constraints. In the same field of edge computing, Sindwani teaches and wherein the capacity profile of the edge device comprises operating system constraints(Col 12, Ln 44-60, execute respective ones of the plurality of useable versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations of the useable versions of the trained ML model and the plurality of different compute instance types. Compute instance types may include features and/or capabilities of specific or particular compute resources for running the trained ML model. In some aspects, different compute instance types may include different virtual machine instance types specified based on an operating system type ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Persia with the edge computing system of Sindwani, as it helps minimize cost(Col 2, Ln 25-30). Regarding Claim 9: The combination of Chen, Persia and Sindwani teaches the system of claim 8, and Chen teaches wherein the text-based service request comprises a description of an AI task, a description of an AI model architecture, a description of an input to the AI model, a description of an output of the AI model, an example of a deployment scenario of the AI model, an example of a specific use-case for the AI model, an example of a re-use of the AI model, a list of performance requirements of the AI model, or a list of generative prompts to the AI model(Pg 5, Fig 3: Application 1, Application 2). Regarding Claim 10: The combination of Chen, Persia and Sindwani teaches the system of claim 8, and Chen teaches wherein the operations to generate the model and data descriptions using the text-based service request further comprise: providing the text-based service request to a pre-trained large language model as input(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); and generating the model and data descriptions using results received from the pre-trained large language model(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Also See Pg 5 Fig 3, library of expert knowledge & task data -> PFM). Regarding Claim 15: Chen teaches receiving a text-based service request for an artificial intelligence (AI) model for an edge device(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); generating model and data descriptions using the text-based service request, wherein generating the model and data descriptions comprises translating requirements, prompts, and other requested features in the text-based service request into the model and data descriptions(Pg 4, Col 1, Para 2, Ln 1-1, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Then, an expert knowledge graph can be established with a tree structure based on the existing structured data knowledge, as illustrated in Fig. 2. See Fig 2 . Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Pg 4, Col 2, Para 3, Adaptor Tuning, Ln 1-8, The adaptor tuning method entails the insertion of additional trainable parameters at each layer of the PFM. These parameters are proficient in modifying the outputs to better align with specific tasks without altering the fundamental structure or weights of the model. This strategy aims to retain the majority of pre-trained knowledge while permitting a degree of fine-tuning to enhance performance on specific tasks .); generating an AI task capacity profile, wherein the AI task capacity profile is generated based on a capacity profile of the edge device, and wherein the capacity profile of the edge device comprises ……..and at least one of supported AI model formats of the edge device and available inference engines of the edge device(Pg 4, Col 2, Para 5, Ln 1-6, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices . See Fig 3, Edge, traditional algs toolkit, tool AI models toolkit); and selecting a resource-optimal AI model for deployment on the edge device based on the AI task capacity profile(Pg 4, Col 2, Para 5, Ln 1-15, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices……..resource scheduling and task orchestration for a cell have shifted from its edge server to the cloud-based intent aware PFM within this framework . See Pg 5, Fig 3, Edge AI models toolkit. Pg 5, Col 2, Para 1, Ln 1-15, AI models stored in the algorithm toolkits are orchestrated by the well trained PFM to handle tasks across multiple edge/end-user devices……Through intent recognition and unified orchestration, these models are assigned to tasks that align with their capabilities, enabling them to effectively leverage the relationships between tasks ). Chen does not explicitly teach a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform operations comprising. Persia teaches a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform operations comprising(Para [0083], Ln 1-14, Processor 620 is implemented in hardware, firmware, or a combination of hardware and software. In some implementations, processor 620 includes one or more processors capable of being programmed to perform a function. Memory 630 includes a random access memory ): It would have been obvious for one skilled in the art, at the effective time of filing, to modify Chen with the computer components of Persia, as it provides an environment for the system to be realized(Para [0082], Ln 1-13, Para [0083], Ln 1-14). The combination of Chen and Persia does not teach and wherein the capacity profile of the edge device comprises operating system constraints. In the same field of edge computing, Sindwani teaches and wherein the capacity profile of the edge device comprises operating system constraints(Col 12, Ln 44-60, execute respective ones of the plurality of useable versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations of the useable versions of the trained ML model and the plurality of different compute instance types. Compute instance types may include features and/or capabilities of specific or particular compute resources for running the trained ML model. In some aspects, different compute instance types may include different virtual machine instance types specified based on an operating system type ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Persia with the edge computing system of Sindwani, as it helps minimize cost(Col 2, Ln 25-30). Regarding Claim 16: Claim 16 contains similar limitations as Claim 9, and is therefore rejected for the same reasons. Regarding Claim 17: Claim 17 contains similar limitations as Claim 10, and is therefore rejected for the same reasons . 07-22-aia AIA Claim (s) 11-12 and 18-19 is/are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Chen, Persia and Sindwani as applied to claim 8 above, and further in view of Xu . Regarding Claim 11: The combination of Chen, Persia and Sindwani teaches the system of claim 8, and Chen teaches wherein the operations to generate the AI task capacity profile further comprises: retrieving a capacity profile of the edge device(Pg 4, Col 2, Para 5, Ln 1-15, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices ); The combination of Chen, Persia and Sindwani does not specifically teach identifying performance and resource parameters by comparing the capacity profile of the edge device to an AI model requirements mapping; and generating the AI task capacity profile using the performance and resource parameters. In the same field of AI Edge Computing, Xu teaches identifying performance and resource parameters by comparing the capacity profile of the edge device to an AI model requirements mapping(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1- 5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n ); and generating the AI task capacity profile using the performance and resource parameters(Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen, Persia and Sindwani with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12). Regarding Claim 12: The combination of Chen, Persia and Sindwani teaches the system of claim 8, but does not specifically teach wherein the AI task capacity profile comprises a compatibility list that comprises hardware and software mismatches between a potential AI model and edge device or potential bottlenecks in memory, CPU, GPU, or software infrastructure of the edge device. In the same field of AI Edge Computing, Xu teaches wherein the AI task capacity profile comprises a compatibility list that comprises hardware and software mismatches between a potential AI model and edge device or potential bottlenecks in memory, CPU, GPU, or software infrastructure of the edge device(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen, Persia and Sindwani with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12). Regarding Claim 18: Claim 18 contains similar limitations as Claim 11, and is therefore rejected for the same reasons. Regarding Claim 19: Claim 19 contains similar limitations as Claim 12, and is therefore rejected for the same reasons . 07-22-aia AIA Claim (s) 13-14 and 20-21 is/are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Chen, Persia and Sindwani as applied to claim 8 above, and further in view of Zawish . Regarding Claim 13: The combination of Chen, Persia and Sindwani teaches the system of claim 8, and Chen teaches wherein the operations to select the resource-optimal AI model for deployment on the edge device based on the AI task capacity profile further comprise: identifying an AI model family using the model and data descriptions and the AI task capacity profile(Pg 5, Col 2, Para 1, Ln 1-15, AI models stored in the algorithm toolkits are orchestrated by the well trained PFM to handle tasks across multiple edge/end-user devices……Through intent recognition and unified orchestration, these models are assigned to tasks that align with their capabilities, enabling them to effectively leverage the relationships between tasks. Pg 4, Col 2, Para 5, Ln 1-6, intelligent edge within this framework pools the majority of its local resources. These resources are represented as part of the edge’s status information, which is then transmitted to the cloud along with requests from end-user devices. Pg 4, Col 1, Para 2, Ln 1-8, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods ); The combination of Chen, Persia and Sindwani does not teach and selecting a model variant of the AI model family based on the AI task capacity profile and resources of the edge device. In the same field of AI Edge Computing, Zawish teaches and selecting a model variant of the AI model family based on the AI task capacity profile and resources of the edge device(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13, reduction in memory of a CNN is critical when the aim is to achieve both computation and energy efficiency ….. Moreover, in order for the deep models to run at the edge, they must fit within the target device’s RAM without disrupting the IoT application at the runtime. To achieve this, the memory-based complexity of each convolutional layer k can be calculated using. Pg 3894, Col 1, Para 1, Ln 1-2, model M must be either less than or equal to the desired complexityCr). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen, Persia and Sindwani with the Pruning methods of Zawish, as it can improve computation and energy efficiency of the model(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13). Regarding Claim 14: The combination of Chen, Persia, Sindwani and Zawish teaches the system of claim 13, but does not teach wherein the model variant is a compressed, pruned, or quantized AI model to correspond to resources of the edge device. In the same field of AI Edge Computing, Zawish teaches wherein the model variant is a compressed, pruned, or quantized AI model to correspond to resources of the edge device(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13, reduction in memory of a CNN is critical when the aim is to achieve both computation and energy efficiency ….. Moreover, in order for the deep models to run at the edge, they must fit within the target device’s RAM without disrupting the IoT application at the runtime. To achieve this, the memory-based complexity of each convolutional layer k can be calculated using. Pg 3894, Col 1, Para 1, Ln 1-2, model M must be either less than or equal to the desired complexityCr). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen, Persia, Sindwani and Zawish with the Pruning methods of Zawish, as it can improve computation and energy efficiency of the model(Pg 3892, Col 2, Para 2, Memory Aware Pruning, Ln 1-13). Regarding Claim 20: Claim 20 contains similar limitations as Claim 13, and is therefore rejected for the same reasons. Regarding Claim 21: Claim 21 contains similar limitations as Claim 14, and is therefore rejected for the same reasons . 07-21-aia AIA Claim (s) 22-23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen, and further in view of Xu, and further in view of Sindwani . Regarding Claim 22: Chen teaches a computer-implemented method comprising: receiving a service request for an artificial intelligence (AI) model for an edge device(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); generating model and data specifications by using automated generative translations of the service request, wherein using the automated generative translations comprises translating requirements, prompts, and other requested features in the service request into the model and data specifications ; (Pg 4, Col 1, Para 2, Ln 1-1, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Then, an expert knowledge graph can be established with a tree structure based on the existing structured data knowledge, as illustrated in Fig. 2. See Fig 2 . Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Pg 4, Col 2, Para 3, Adaptor Tuning, Ln 1-8, The adaptor tuning method entails the insertion of additional trainable parameters at each layer of the PFM. These parameters are proficient in modifying the outputs to better align with specific tasks without altering the fundamental structure or weights of the model. This strategy aims to retain the majority of pre-trained knowledge while permitting a degree of fine-tuning to enhance performance on specific tasks ); Chen does not teach performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model; and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter. In the same field of AI Edge Computing, Xu teaches performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model, wherein the capacity profile of the edge device comprises…..and at least one of supported AI model formats of the edge device and available inference engines of the edge device(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n. Pg 4, Col 2, Para 4, Ln 1 – Pg 5, Col 1, Para 1, Ln 7, When additional GPU memory is required for loading an uncached requested PFM, the LC algorithm counts the number of examples in context, calculates them, and removes the cached PFM with the fewest effective examples in context. Therefore, at each timeslot t, the model caching decisions can be obtained by solving the maximization problem of the number of effective examples for the cached models, which can be represented as (eq 13(a-c)).); and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter( . Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories . Pg 4, Col 2, Para 4, Ln 1 – Pg 5, Col 1, Para 1, Ln 7, When additional GPU memory is required for loading an uncached requested PFM, the LC algorithm counts the number of examples in context, calculates them, and removes the cached PFM with the fewest effective examples in context. Therefore, at each timeslot t, the model caching decisions can be obtained by solving the maximization problem of the number of effective examples for the cached models, which can be represented as (eq 13(a-c)). It would have been obvious for one skilled in the art, at the effective time of filling, to modify Chen with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12). The combination of Chen and Xu does not teach wherein the capacity profile of the edge device comprises operating system constraints. In the same field of edge computing, Sindwani teaches and wherein the capacity profile of the edge device comprises operating system constraints(Col 12, Ln 44-60, execute respective ones of the plurality of useable versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations of the useable versions of the trained ML model and the plurality of different compute instance types. Compute instance types may include features and/or capabilities of specific or particular compute resources for running the trained ML model. In some aspects, different compute instance types may include different virtual machine instance types specified based on an operating system type ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Xu with the edge computing system of Sindwani, as it helps minimize cost(Col 2, Ln 25-30). Regarding Claim 23: The combination of Chen, Xu and Sindwani teaches the computer-implemented method of claim 22, and Chen teaches wherein generating the model and data specifications by using the automated generative translations of the service request further comprise: providing the service request to a pre-trained large language model as input(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); and generating the model and data specifications using results generated by the pre-trained large language model(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Also See Pg 5 Fig 3, library of expert knowledge & task data -> PFM) . 07-21-aia AIA Claim (s) 24-25 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen, and further in view of Xu and further in view of Sindwani . Regarding Claim 24: Chen teaches receiving a service request for an artificial intelligence (AI) model for an edge device(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); generating model and data specifications by using automated generative translations of the service request, wherein using the automated generative translations comprises translating requirements, prompts, and other requested features in the service request into the model and data specifications ; (Pg 4, Col 1, Para 2, Ln 1-1, two prerequisites, namely, the dataset and fine-tuning scheme for the PFM. For the former, we endeavor to construct an expert knowledge library by aggregating non-private data from multiple clouds. To fully exploit the standardized characteristics of wireless communication processes, we categorize user requests based on various criteria, such as the type/target of the task, processing workflow, and signal processing methodologies. Then, an expert knowledge graph can be established with a tree structure based on the existing structured data knowledge, as illustrated in Fig. 2. See Fig 2 . Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Pg 4, Col 2, Para 3, Adaptor Tuning, Ln 1-8, The adaptor tuning method entails the insertion of additional trainable parameters at each layer of the PFM. These parameters are proficient in modifying the outputs to better align with specific tasks without altering the fundamental structure or weights of the model. This strategy aims to retain the majority of pre-trained knowledge while permitting a degree of fine-tuning to enhance performance on specific tasks ); Chen does not teach performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model; and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter. In the same field of AI Edge Computing, Xu teaches performing an AI task capacity profiling using the model and data specifications and a capacity profile of the edge device to identify a key performance parameter and a key resource parameter of the AI model, wherein the capacity profile of the edge device comprises…..and at least one of supported AI model formats of the edge device and available inference engines of the edge device(Pg 2, Col 2, Para 1, Ln 1-13, we consider an edge intelligence system model……cloud data center and edge servers can serve generative AI services. The cloud data center is represented by 0 and the set of edge servers is represented by N = {1,2,...,N}. In this system, edge servers and the cloud center provide generic AI services such as AIGC, depending on different PFMs. Pg 3, Col 1, Para 1, Ln 1-16, To offer AI services based on PFMs, we propose a joint foundation model caching and inference framework. Edge servers need to make model caching and request offloading decisions to utilize the existing edge computing resources for accommodating generative AI service requests of mobile users…….the binary variable indicating whether model m of application i is cached at edge server n. Pg 3, Col 1, Para 2, Ln 1- 5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories. Let Gn denote the capacity of GPU memory of edge server n. Pg 4, Col 2, Para 4, Ln 1 – Pg 5, Col 1, Para 1, Ln 7, When additional GPU memory is required for loading an uncached requested PFM, the LC algorithm counts the number of examples in context, calculates them, and removes the cached PFM with the fewest effective examples in context. Therefore, at each timeslot t, the model caching decisions can be obtained by solving the maximization problem of the number of effective examples for the cached models, which can be represented as (eq 13(a-c)).); and selecting the AI model for deployment on the edge device based on the key performance parameter and the key resource parameter( . Pg 3, Col 1, Para 2, Ln 1-5, The generative AI service requests of users can be executed at edge servers if the required components of models are loaded at the GPU memories . Pg 4, Col 2, Para 4, Ln 1 – Pg 5, Col 1, Para 1, Ln 7, When additional GPU memory is required for loading an uncached requested PFM, the LC algorithm counts the number of examples in context, calculates them, and removes the cached PFM with the fewest effective examples in context. Therefore, at each timeslot t, the model caching decisions can be obtained by solving the maximization problem of the number of effective examples for the cached models, which can be represented as (eq 13(a-c)). It would have been obvious for one skilled in the art, at the effective time of filling, to modify Chen with the Edge Computing system of Xu, as it can help improve model performance(Pg 6, Col 1, Para 3, Ln 1-12). The combination of Chen and Xu does not teach A system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:………wherein the capacity profile of the edge device comprises operating system constraints. In the same field of edge computing, Sindwani teaches a system comprising: a memory having computer readable instructions; and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising(Col 22, Ln 29-40, System memory 720 may be configured to store instructions and data accessible by processor(s) 710 . In at least some embodiments, the system memory 720 may comprise both volatile and non-volatile portions; in other embodiments, only volatile memory may be used. In various embodiments, the volatile portion of system memory 720 may be implemented using any suitable memory technology, such as static random access memory ): and wherein the capacity profile of the edge device comprises operating system constraints(Col 12, Ln 44-60, execute respective ones of the plurality of useable versions of the trained ML models on a plurality of different compute instance types using the model evaluation data to produce model evaluation results for a plurality of different combinations of the useable versions of the trained ML model and the plurality of different compute instance types. Compute instance types may include features and/or capabilities of specific or particular compute resources for running the trained ML model. In some aspects, different compute instance types may include different virtual machine instance types specified based on an operating system type ). It would have been obvious for one skilled in the art, at the effective time of filling, to modify the combination of Chen and Xu with the edge computing system of Sindwani, as it provides an environment for the system to be realized(Col 22, Ln 29-40), helps minimize cost(Col 2, Ln 25-30). Regarding Claim 25: The combination of Chen, Xu and Sindwani teaches the system of claim 24, and Chen teaches wherein the operations to generate the model and data specifications by using the automated generative translations of the service request further comprise: providing the service request to a pre-trained large language model as input(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device ); and generating the model and data specifications using results generated by the pre-trained large language model(Pg 3, Col 2, Para 2, Ln 1-6, The cloud identifies user intent by processing queries from the end-user device using the intent-aware PFM. It then manages task orchestration and resource allocation for the edge or end-user device. Pg 4, Col 1, Para 3, Ln 1-5, The latter requires a meticulously designed fine-tuning method due to the massive scale and resource requirements of the majority of foundation models. Thus, we implement the native intent-aware PFM by fine-tuning the existing PFM with parameter-efficient fine-tuning methods. Examples available in Pg 4, Col 1, Para 4 - Col 2 Para 3: Prompt tuning, Prefix Tuning, LoRA and Adaptor tuning. Also See Pg 5 Fig 3, library of expert knowledge & task data -> PFM). Conclusion 07-40 AIA Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL . See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALEXANDER G MARLOW whose telephone number is (571)272-4536. The examiner can normally be reached Monday - Thursday 10:00 am - 8:00 pm EST. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Richmond Dorvil can be reached at (571)272-7602. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /ALEXANDER G MARLOW/Assistant Examiner, Art Unit 2658 /RICHEMOND DORVIL/Supervisory Patent Examiner, Art Unit 2658 Application/Control Number: 18/623,195 Page 2 Art Unit: 2658 Application/Control Number: 18/623,195 Page 3 Art Unit: 2658 Application/Control Number: 18/623,195 Page 4 Art Unit: 2658 Application/Control Number: 18/623,195 Page 5 Art Unit: 2658 Application/Control Number: 18/623,195 Page 6 Art Unit: 2658 Application/Control Number: 18/623,195 Page 7 Art Unit: 2658 Application/Control Number: 18/623,195 Page 8 Art Unit: 2658 Application/Control Number: 18/623,195 Page 9 Art Unit: 2658 Application/Control Number: 18/623,195 Page 10 Art Unit: 2658 Application/Control Number: 18/623,195 Page 11 Art Unit: 2658 Application/Control Number: 18/623,195 Page 12 Art Unit: 2658 Application/Control Number: 18/623,195 Page 13 Art Unit: 2658 Application/Control Number: 18/623,195 Page 14 Art Unit: 2658 Application/Control Number: 18/623,195 Page 15 Art Unit: 2658 Application/Control Number: 18/623,195 Page 16 Art Unit: 2658 Application/Control Number: 18/623,195 Page 17 Art Unit: 2658 Application/Control Number: 18/623,195 Page 18 Art Unit: 2658 Application/Control Number: 18/623,195 Page 19 Art Unit: 2658 Application/Control Number: 18/623,195 Page 20 Art Unit: 2658 Application/Control Number: 18/623,195 Page 21 Art Unit: 2658 Application/Control Number: 18/623,195 Page 22 Art Unit: 2658 Application/Control Number: 18/623,195 Page 23 Art Unit: 2658 Application/Control Number: 18/623,195 Page 24 Art Unit: 2658 Application/Control Number: 18/623,195 Page 25 Art Unit: 2658 Application/Control Number: 18/623,195 Page 26 Art Unit: 2658 Application/Control Number: 18/623,195 Page 27 Art Unit: 2658 Application/Control Number: 18/623,195 Page 28 Art Unit: 2658 Application/Control Number: 18/623,195 Page 29 Art Unit: 2658 Application/Control Number: 18/623,195 Page 30 Art Unit: 2658 Application/Control Number: 18/623,195 Page 31 Art Unit: 2658 Application/Control Number: 18/623,195 Page 32 Art Unit: 2658 Application/Control Number: 18/623,195 Page 33 Art Unit: 2658 Application/Control Number: 18/623,195 Page 34 Art Unit: 2658 Application/Control Number: 18/623,195 Page 35 Art Unit: 2658 Application/Control Number: 18/623,195 Page 36 Art Unit: 2658 Application/Control Number: 18/623,195 Page 37 Art Unit: 2658 Application/Control Number: 18/623,195 Page 38 Art Unit: 2658 Application/Control Number: 18/623,195 Page 39 Art Unit: 2658 Application/Control Number: 18/623,195 Page 40 Art Unit: 2658 Application/Control Number: 18/623,195 Page 41 Art Unit: 2658 Application/Control Number: 18/623,195 Page 42 Art Unit: 2658 Application/Control Number: 18/623,195 Page 43 Art Unit: 2658 Application/Control Number: 18/623,195 Page 44 Art Unit: 2658 Application/Control Number: 18/623,195 Page 45 Art Unit: 2658 Application/Control Number: 18/623,195 Page 46 Art Unit: 2658 Application/Control Number: 18/623,195 Page 47 Art Unit: 2658 Application/Control Number: 18/623,195 Page 48 Art Unit: 2658 Application/Control Number: 18/623,195 Page 49 Art Unit: 2658
Prosecution Insights