DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-13 are pending for examination in the instant application.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 08/20/2025 is in compliance
with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Priority
The applicant is claiming foreign priority based on foreign applications
EP25315007.2, filed on 01/08/2025. Receipt is acknowledged of certified copies of papers
required by 37 CFR 1.55.
Claim Rejections - 35 USC § 103
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.
Claims 1-13 is/are rejected under 35 U.S.C.103 as being unpatentable
over Tamvada al. (Pub. No.: US 20220386393 A1), hereinafter “Tam” in view of Wei et al. (Pub. No.: US 20140185606 A1), hereinafter “Wei” in further view of Galbi et al. (Pub. No.: US 20240370699 A1), hereinafter “Gal”.
As to claim 1, Tam discloses A mobile network architecture, for processing
requests based on Large Language Models, LLMs, issued to LLM applications by users, wherein the architecture comprises a 5G network with (Tam, [0023], The communication network 100 includes a radio-based network 103, which may correspond to a cellular network such as a fourth-generation (4G) Long-Term Evolution (LTE) network, a fifth-generation (5G) network, a 4G-5G hybrid core with both 4G and 5G RANs, or another network that provides wireless network access. [0001], 5G is the fifth-generation technology standard for broadband cellular networks, which is planned eventually to take the place of the fourth generation (4G) standard of Long-Term Evolution (LTE).
a radio access network, for radiofrequency communication with user equipment’s, UEs, the UEs accessing the network via 5G access nodes according to 3GPP, gNBs, including far-edge server calculation and storage resources (Tam, [0025], The radio-based network 103 can include a RAN that provides the wireless network access to the plurality of wireless devices 106 through a plurality of cells 109. Each of the cells 109 may be equipped with one or more antennas and one or more radio units (RUs) that send and receive wireless data signals to and from the wireless devices 106. [0024], The wireless devices 106 are sometimes referred to as user equipment (UE) or customer premises equipment (CPE). [0023], These components may be based on the 3GPP specifications by following an application architecture in which control plane and user plane processing is separated (CUPS Architecture). [0030], Various network functions to implement the radio-based network 103 may be deployed in distributed computing devices 112, which may correspond to general-purpose computing devices configured to perform the network functions. For example, the distributed computing devices 112 may execute one or more virtual machine instances that are configured in turn to execute one or more services that perform the network functions. In one embodiment, the distributed computing devices 112 are ruggedized machines that are deployed at each cell site.).;
a distributed network of programmable user plane functions, UPFs, also acting as a 5G data transport plane for routing data packets towards/from the UEs (Tam ,[0017], Various embodiments of the present disclosure introduce a distributed implementation of a UPF that places a user plane component of the UPF within the RAN in contrast to traditional implementations where the UPF is wholly implemented in the core network. [0027], The UPF can also handle packet routing and forwarding, including directing flows to specific data networks based on traffic matching filters. [0041], The UPF 146 forwards return network traffic to the radio access network 143 also via the network interface 154. [0124], The UPF controller 424 may determine to apportion the network traffic based at least in part on the UPF control configuration 475 (FIG. 4 ), which may include rule sets for assigning network traffic to types of user plane components.).;
and - a core network control plane comprising 5G functions according to 3GPP (Tam, [0026], The network functions of the core network can include a User Plane Function (UPF), Access and Mobility Management Function (AMF), and Session Management Function (SMF), described in more detail below. [0089], The DU/CUs 284 and the legacy DU/CU 322 may exchange control plane messages via the AMF 330 in the core network 118.).
Tam, however, is silent to discloses explicitly, for processing LLM requests, the LLM applications implement multi-layer regressive inference processes using key-value caches allocated to respective UEs.
Gal discloses the similar concept in same field of endeavor for processing LLM requests, the LLM applications implement multi-layer regressive inference processes using key-value caches allocated to respective UEs (Gal, [0001], A transformer is made up of multiple transformer blocks, also known as layers. A transformer layer has a self-attention kernel, a feed-forward kernel, and normalization kernels. [0013], During the token decode phase, instead of processing an extensive sequence of prompts, the most recently generated token is input into the LLM. The LLM performs a series of vector-matrix multiplications, and the speed of this phase can be constrained by available bandwidth, rather than computational capacity. The LLM uses the context stored in the KV cache to generate new tokens. The LLM adds a new token to the KV cache to update the context. [0014], The LLM can reference the KV cache to determine a next token, rather than having to process the entire input again. [0025], In some cases, KV values for multiple streams can be unique for different streams. For example, if an integer N number of streams are batched together, a size of KV cache data can be increased by N times.).
Therefore, before the effective filing date of the instant application it would have
been obvious to one of the ordinary skilled in the art to incorporate the teachings of
“Gal” in those of Tam for reducing latency, network traffic and improving the efficiency and responsiveness of data processing.
The combined system of Tam and Gal discloses the invention as applied above. Tam and Gal, however, are silent to discloses explicitly, the calculation and storage resources of the gNBs are configured to form a peer-to-peer network implementing key-value caches distributed within the 5G network, adapted to store the LLM requests from the UEs, wherein the gNBs serve the respective UEs; the key-value caches are ephemeral caches with a limited lifetime; the architecture further comprises a cache scheduler adapted to interact, on the one hand, with the LLM applications and, on the other hand, with the gNBs via the core network control plane; and - the cache scheduler comprises means adapted to create the key-value caches, assign the created key-value caches to the respective UEs, determine the lifetime of the key-value caches created, and update the key-value caches.
Wei however discloses a similar concept in the same field of endeavor including, the calculation and storage resources of the gNBs are configured to form a peer-to-peer network implementing key-value caches distributed within the 5G network, adapted to store the LLM requests from the UEs, wherein the gNBs serve the respective UEs (Wei, [0039], A cache configured to store hot content is deployed in a distributed manner on multiple NodeBs managed by the RNC. [0040], Physical links are newly established and corresponding signaling is added between the NodeBs subordinate to the RNC to form a network that includes the RNC and the subordinate NodeBs, so as to mutually transmit cached files and storage locations of the cached files on a UTRAN side. A connection between one NodeB and another NodeB may be an IP connection, a physical link between them may be in a wired or wireless form, and a network topology structure between them may be star, ring, or the like. [0154], NodeBs on which a Cache is deployed and an RNC to which the NodeBs belong together form a meshed network, and this is equivalent to expansion of a storage space. [0134], In this embodiment, the user equipment 61 is within coverage of the first base station 62, and requests accessing a packet domain service (content of target.html).;
the key-value caches are ephemeral caches with a limited lifetime (Wei, [0144], Such a part of content may be cached without reaching a popularity threshold and is always cached on the first NodeB within a period signed with an SP or before the SP requests an update to the subscribed content. [0152], the RNC selects content (such as blue.html shown in the figure) whose popularity is relatively low from content cached on the first NodeB, instructs the first NodeB to delete the content from the cache and store the new content target.html. In the lists maintained by the first NodeB and the RNC, entries of the content need to be deleted accordingly.).;
the architecture further comprises a cache scheduler adapted to interact, on the one hand, with the LLM applications and, on the other hand, with the gNBs via the core network control plane (Wei, [0044], Classification and a search sequence of the three lists may enable the RNC to schedule the Cache network at the fastest speed. [0154], interworking information is uniformly scheduled through the RNC, and a hit rate of the Cache can be ensured even in a mobility scenario. [0156], According to an idea of a distributed network, the following problem is solved in a manner of cooperative scheduling by the RNC: Content that reaches a cache threshold cannot be cached due to a limited storage space of a single network element.).;
and - the cache scheduler comprises means adapted to create the key-value caches (Wei, [0148], the RNC adds an entry to the cached-content list to record popularity and a storage address of the content and deletes the record of the content from the pre-popular content list. [0078], sending the found information of the location in which the packet domain service is cached to the base station for caching, adding the found packet domain service and location information of a base station on which the packet domain service is cached to the cached-content list, and marking the current popularity of the packet domain service as 0.).
assign the created key-value caches to the respective UEs (Wei, [0086], selecting a base station that has a relatively large remaining cache space among all the base station and is relatively close to the base station that has sent a request for accessing the packet domain service, so as to cache the packet domain service, and sending a location of the selected base station; and adding a record of the location of the base station to the cached-content list. [0044], As a hub network element of the meshed Cache network, the RNC only needs to maintain three lists to manage Cache content of the entire network. Classification and a search sequence of the three lists may enable the RNC to schedule the Cache network at the fastest speed. [0129], he apparatus for synchronizing a popularity value of cache data is configured to receive a request for accessing a packet domain service sent by a user equipment, respond to the packet domain service request, and if it is determined that the packet domain service is locally cached, send the packet domain service to the user equipment and increment a current popularity value of the packet domain service in a list of locally cached content according to a preset function; and when the current popularity value of the packet domain service is accumulated to a preset threshold, send the current popularity value of the packet domain service to a popularity information synchronizing node on a radio access network for synchronization, and clear the current popularity value locally cached.).
determine the lifetime of the key-value caches created and update the key-value caches (Wei, [0144], is always cached on the first NodeB within a period signed with an SP or before the SP requests an update to the subscribed content. [0071], The popularity information synchronizing node updates a popularity value that corresponds to the packet domain service in a cached-content list according to the request.).
Therefore, before the effective filing date of the instant application it would have
been obvious to one of the ordinary skilled in the art to incorporate the teachings of
“Wei” in those of Tam-Gal for synchronizing a popularity value of distributed cache data and a system for cooperative and distributed caching.
As to claim 2, Tam-Gal-Wei discloses the cache scheduler is adapted to
establish a concatenation scheme for a plurality of existing key-value caches into a single merged key-value cache, allocated to a plurality of UEs (Gal, [0025], In a mixed token batching stream, compute engines 202 can process multiple streams at the same time. Constant weight values (e.g., W values) of multiple streams can repeat and compute engines 202 can process constant weight values of multiple streams from constant weight values (e.g., K or V values) of a single stream in cache 204. In some cases, KV values for multiple streams can be unique for different streams. For example, if an integer N number of streams are batched together, a size of KV cache data can be increased by N times. [0027], In some examples, prefetcher 208 can prefetch data 222-0 to 222-A, as an entirety of a KV cache for an entire transformer model, into layer buffer 206. (Wei, [0156], According to an idea of a distributed network, the following problem is solved in a manner of cooperative scheduling by the RNC: Content that reaches a cache threshold cannot be cached due to a limited storage space of a single network element.).
As to claim 3, Tam-Gal-Wei discloses point-to-point, P2P, communication
means between gNBs to execute within the radio access network a concatenation of the plurality of existing key-value caches into a single merged key-value cache, according to a concatenation scheme established by the cache scheduler (Wei, [0040], Physical links are newly established and corresponding signaling is added between the NodeBs subordinate to the RNC to form a network that includes the RNC and the subordinate NodeBs, so as to mutually transmit cached files and storage locations of the cached files on a UTRAN side. [0150], The first NodeB sends, through the private interface between NodeBs, the content acquired from the SP to the second NodeB for caching.).
As to claim 4, Tam-Gal-Wei discloses the cache scheduler is adapted,
previously to the merging of the plurality of key-value caches, to determine, among a plurality of gNBs of the radio access network, a gNB determined as being optimum based on latency, capacity and/or bandwidth constraints, and to store the merged key-value cache at the so-determined gNB (Wei, [0086], acquiring remaining cache space of all base stations except for a base station that has sent a request for accessing the packet domain service, selecting a base station that has a relatively large remaining cache space among all the base station and is relatively close to the base station that has sent a request for accessing the packet domain service, so as to cache the packet domain service, and sending a location of the selected base station. [0150], The first NodeB sends, through the private interface between NodeBs, the content acquired from the SP to the second NodeB for caching.). (Tam, [0117], network traffic flows having a higher QoS parameter or requirement may be routed so as to minimize latency and improve throughput.).
As to claim 5, Tam-Gal-Wei discloses the cache scheduler is adapted, after
completion of the LLM request processing and in the absence of subsequent LLM request, to suppress the merged key-value cache (Wei, [0152], the RNC selects content (such as blue.html shown in the figure) whose popularity is relatively low from content cached on the first NodeB, instructs the first NodeB to delete the content from the cache and store the new content target.html. In the lists maintained by the first NodeB and the RNC, entries of the content need to be deleted accordingly. [0088], selecting cached content whose popularity is relatively low from the list of locally cached content, instructing the base station to delete the cached content whose popularity is relatively low from the list of locally cached content.).
assign unitary key-value caches to the UEs of the merged cache, and determine respective lifetimes for each of these unitary key-value caches (Gal, [0025], KV values for multiple streams can be unique for different streams. For example, if an integer N number of streams are batched together, a size of KV cache data can be increased by N times but an amount of memory to store the constants weight data can be unchanged. (Wei, [0144], is always cached on the first NodeB within a period signed with an SP or before the SP requests an update to the subscribed content.)
As to claim 6, Tam-Gal-Wei discloses wherein the lifetime of each key-value
cache is a lifetime calculated at the creation of the key-value cache by the cache scheduler based on metrics determined by the 5G network (Wei, [0144], In this embodiment, subscribed content may be configured through a configuration and maintenance terminal on the RNC side. Such a part of content may be cached without reaching a popularity threshold, and is always cached on the first NodeB within a period signed with an SP. [0148], the RNC adds an entry to the cached-content list to record popularity and a storage address of the content.).
As to claim 7, Tam-Gal-Wei discloses the cache scheduler is adapted to
automatically clear the key-value caches after expiry of the limited lifetime (Wei, [0144], In this embodiment, subscribed content may be configured through a configuration and maintenance terminal on the RNC side. Such a part of content may be cached without reaching a popularity threshold, and is always cached on the first NodeB within a period signed with an SP. [0042], After the synchronization is performed, the local popularity on the NodeB is cleared.).
As to claim 8, Tam-Gal-Wei discloses the cache scheduler is adapted to
assign an increased lifetime to the most frequently accessed key-value caches (Wei, [0152], the RNC selects content (such as blue.html shown in the figure) whose popularity is relatively low from content cached on the first NodeB, instructs the first NodeB to delete the content from the cache and store the new content target.html. In the lists maintained by the first NodeB and the RNC, entries of the content need to be deleted accordingly. [0148], if the pre-popular content list has the record and the Cache threshold is reached after a popularity value of current access is accumulated, the RNC instructs the first NodeB to cache content returned by the SP. [0088], selecting cached content whose popularity is relatively low from the list of locally cached content, instructing the base station to delete the cached content whose popularity is relatively low.).
As to claim 9, Tam-Gal-Wei discloses the lifetime calculation metrics include
5G network operational metrics, comprising: - rate of use of the gNB; - storage capacity available in the gNBs (Tam, [0110], performance indicators, such as processor utilization, memory utilization, latency metrics, bandwidth utilization, and so forth.).;
- distribution/allocation of the network resources between uplink and downlink
(Tam, [0027], Another feature of the UPF includes per-flow or per-application QoS handling, including transport level packet marking for uplink (UL) and downlink (DL), and rate limiting.).;
- latency; - bandwidth (Tam, [0104], Such RBN metrics 451 may include bandwidth metrics, dropped packet metrics, signal strength metrics, latency metrics, and so on. The RBN metrics 451 may be aggregated on a per-device basis, a per-cell basis, a per-customer basis, etc.).;
and/or - maximum number of tokens per second (Wei, [0086], acquiring remaining cache space of all base stations except for a base station that has sent a request for accessing the packet domain service).
As to claim 10, Tam-Gal-Wei discloses the lifetime calculation metrics
include metrics specific to the key-value cache user, comprising: - frequency of use of the LLM applications by the user (Wei, [0134], target.html may be directly returned to the user equipment 61, with a popularity value of this resource being simultaneously accumulated by 1. i.e. per-access counting = frequency of use.).;
- belonging of the user to a predefined domain gathering a plurality of users (Wei, [0041], a maintenance list of subscribed Cache content, that is, a list of content specified by a subscription SP for caching.).;
- privilege level assigned to user (Tam, [0103], This device data 448 includes corresponding users, account information, billing information, data plan, permitted applications or uses, an indication of whether the wireless device 106 is mobile or fixed, a location, a current cell, a network address.).;
- QoS degree guaranteed to the user; and/or - subscription by the user to LLM inference services (Tam, [0117], network traffic flows having a higher QoS parameter or requirement may be routed so as to minimize latency and improve throughput.).
As to claim 11, Tam-Gal-Wei discloses wherein the lifetime calculation
metrics include metrics of similarity between simultaneous users of the key-value caches (Wei, [0005], user equipment covered by different NodeBs may access same hot content, causing the same hot content to be cached on multiple NodeBs and causing same content to occupy cache resources of the multiple NodeBs.).
comprising: - belonging or not to a same predefined domain grouping a plurality of users (Tam, [0074], The larger control plane can be divided into cells in order to reduce the likelihood that large scale errors will affect a wide range of customers, for example by having one or more cells per customer, per network, or per region that operate independently.).;
and/or - identical or close geographic location of the users (Tam ,[0103], a location, a current cell, a network address.).
As to claim 12, Tam-Gal-Wei discloses point-to- point communication means
between gNBs for a direct exchange of information relating to the lifetime calculation metrics (Wei, [0040], Physical links are newly established and corresponding signaling is added between the NodeBs subordinate to the RNC to form a network that includes the RNC and the subordinate NodeBs, so as to mutually transmit cached files and storage locations of the cached files on a UTRAN side. [0150], Then broadcast message is sent to all NodeBs except for the first NodeB, requesting each of the NodeBs to send information about a remaining cache space. [0093], The NodeBs may also exchange data with each other through the HUB NodeB.).
As to claim 13, Tam-Gal-Wei discloses the UEs are devices of the group
comprising smartphones, autonomous robots, and/or video surveillance cameras (Tam, [0024], In various examples, the wireless devices 106 may include smartphones, connected vehicles, IoT devices, sensors, machinery (such as in a manufacturing facility), hotspots, and other devices.).
including a circuit for connection to the 5G network (Tam, [0025], Each of the cells 109 may be equipped with one or more antennas and one or more radio units (RUs) that send and receive wireless data signals to and from the wireless devices 106.).
and whose profile has already been entered into a 5G-network core network user database (Tam ,[0103], The device data 448 corresponds to data describing wireless devices 106 that are permitted to connect to the radio-based network 103. [0077], a unified data management (UDM) function that manages subscriber data for authorization, registration, and mobility management.).
Conclusion
The prior art made of record and not relied upon is considered pertinent to
applicant’s disclosure:
Yuan et. al. (US 20260119893 A1) is one of the pertinent art discloses techniques for direct knowledge injection into large language models using a key-value cache network layer. A Key-Value cache (KV-cache) network layer to a Large Language Model (LLM). A twin data distribution corresponding to the KV-cache network layer is updated by adding new data points. A Gaussian Mixture Model (GMM) is updated with the new data points based on the updated twin data distribution. The KV-cache network layer is updated with the new data points based on the Gaussian Mixture Model. A question is issued to the LLM, where the LLM generates an answer using the new data points of the updated KV-cache network layer. The answer to the question is received from the LLM.
Ramachandran et. al. (US 20260119893 A1) is one of the pertinent art discloses an apparatus including: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: receive, from a plurality of network nodes, information related to at least one parameter of models of the plurality of network nodes; determine at least one cluster of the plurality of network nodes based on at least one similarity criterion and the information related to the least one parameter of models of the plurality of network nodes; and determine at least one global model for the at least one cluster using local models of network nodes that belong to the at least one cluster.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MUSADDIQUE WAHID whose telephone number is (571)270-0865. The examiner can normally be reached Monday Friday, 8 a.m. 5 p.m. ET..
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, Umar Cheema can be reached at 5712703037. 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.
/M.W./ Examiner, Art Unit 2458
/UMAR CHEEMA/Supervisory Patent Examiner, Art Unit 2458