DETAILED ACTION
This communication is in response to the Application filed on 02 October 2023. Claims 1-40 are pending and have been examined.
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 .
Priority
Applicant claims the benefit of US Provisional Application No. 63/454,605, filed 24 March 2023, 2021. Claims 1-40 have been afforded the benefit of this filing date.
Information Disclosure Statement
The IDS dated 05 January 2024 has been considered and placed in the application file.
The IDS dated 21 June 2024 has been considered and placed in the application file.
The IDS dated 31 December 2025 has been considered and placed in the application file.
The IDS dated 20 April 2026 has been considered and placed in the application file.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claims 1-4, 8-10, 12-17, 20-24, 28-30, 32-37, and 40 are provisionally rejected on the ground of nonstatutory double patenting as being unpatentable over claims 1 and 3-11 of co-pending Application No. 18/479,672 (reference application) in view of “Lossless Speedup of Autoregressive Translation with Generalized Aggressive Decoding” (Xia et al.). Although the claims at issue are not identical, they are not patentably distinct from each other because the claims of the reference application are narrower than the claims of the instant application. Please see the table below for the claim mappings.
Instant Application: 18/479,659
Reference Application: 18/479,672
Claim 1: A processing system, comprising:
a memory having executable instructions stored thereon; and
one or more processors configured to execute the executable instructions to cause the processing system to:
generate, based on an input query and a first generative model, a plurality of sets of tokens, each set of tokens in the plurality of sets of tokens corresponding to a candidate response to the input query;
output to a second generative model, the plurality of sets of tokens for verification;
receive, from the second generative model, an indication of a selected set of tokens from the plurality of sets of tokens based on the input query and the plurality of sets of tokens; and
output the selected set of tokens as a response to the input query.
Claim 1: A processing system comprising:
memory having executable instructions stored thereon; and
one or more processors configured to execute the executable instructions to cause the processing system to:
generate, based on an input query and using a first generative model, a first plurality of sets of tokens, each set of tokens in the first plurality of sets of tokens corresponding to a first portion of a candidate response to the input query;
output, from the first generative model to a second generative model, the first plurality of sets of tokens for verification;
after the first plurality of sets of tokens is outputted and while waiting to receive, from the second generative model, an indication of a selected set of tokens from the first plurality of sets of tokens, speculatively generate a second plurality of sets of tokens using the first generative model, each set of tokens in the second plurality of sets of tokens corresponding to a second portion of the candidate response to the input query;
after the second plurality of sets of tokens is speculatively generated, receive, at the first generative model from the second generative model, the indication of the selected set of tokens from the first plurality of sets of tokens;
output, from the first generative model to the second generative model, tokens from the second plurality of sets of tokens associated with the selected set of tokens for verification; and
output the selected set of tokens as a response to the input query.
Claim 1
Claim 1 of the reference application recites all of the limitations of claim 1 of the instant application except “an indication… based on the input query and the plurality of sets of tokens.”
However, Xia et al. disclose an indication of a selected set of tokens from the plurality of sets of tokens based on the input query and the plurality of sets of tokens (Xia et al. Figure 3, see row labelled "Output" which corresponds to verified tokens based on the set of drafted tokens and input sentence).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s indication based on the input query and the plurality of sets of tokens within the method of claim 1 of the reference application because such a modification is the result of simple substitution of one known element for another producing a predictable result. More specifically, the reference application’s indication and Xia et al.’s indication perform the same general and predictable function, the predictable function being indicating the selected tokens for outputting as a response to the input query. Since each individual element and its function are shown in the prior art, albeit shown in separate references, the difference between the claimed subject matter and the prior art rests not on any individual element or function but in the very combination itself - that is in the substitution of reference application’s indication by replacing it with Xia et al.’s indication. Thus, the simple substitution of one known element for another producing a predictable result renders the claim obvious.
Claim 2
Claim 3 of the reference application recites all of the limitations of claim 2 of the instant application.
Claim 3
Claim 4 of the reference application recites all of the limitations of claim 3 of the instant application.
Claim 4
Claim 5 of the reference application recites all of the limitations of claim 4 of the instant application.
Claim 8
Claim 6 of the reference application recites all of the limitations of claim 8 of the instant application.
Claim 9
Claims 1 and 7 of the reference application recites all of the limitations of claim 9 of the instant application.
Claim 10
Claim 8 of the reference application recites all of the limitations of claim 10 of the instant application.
Claim 12
Claim 9 of the reference application recites all of the limitations of claim 12 of the instant application.
Claim 13
Claim 10 of the reference application recites all of the limitations of claim 13 of the instant application.
Claim 14
Claim 11 of the reference application recites all of the limitations of claim 14 of the instant application.
Claim 15
Claim 1 of the reference application recites all of the limitations of claim 15 of the instant application except “comparing a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens”
However, Xia et al. disclose compare a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
" See Figure 3; "Verify (AT)" is considered analogous to a second generative model. Thus, taking the maximum probability distrubtion among drafted tokens
y
~
j
+
1
…
j
+
k
is considered analogous to comparing probability distributions generated by the second generative model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s autoregressive probability distribution comparisons within the method of claim 1 of the reference application because such a modification is the result of combining prior art elements according to known methods to yield predictable results. More specifically, claim 1 of the reference application as modified by Xia et al.’s autoregressive probability distribution comparisons can yield a predictable result of improving system accuracy since the system would be able dynamically compare probability distributions with previous generations, allowing for iterative improvement over time. Thus, a person of ordinary skill would have appreciated including in claim 1 of the reference application the ability to do Xia et al.’s autoregressive probability distribution since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
Claim 16
Claim 5 of the reference application recites all of the limitations of claim 16 of the instant application.
Claim 17
Claim 1 of the reference application recites all of the limitations of claim 17 of the instant application except “generating probability distributions for each respective set of tokens based on a single pass through the second generative model.”
However, Xia et al. disclose wherein to compare the probability distribution associated with each respective set of tokens in the plurality of sets of tokens to the corresponding probability distribution generated by the second generative model for the respective set of tokens, the one or more processors are configured to cause the processing system to generate probability distributions for each respective set of tokens based on a single pass through the second generative model (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
where ...
y
^
j
+
i
is the top-1 result verified by the AT model conditioning on the previously translated tokens
y
≤
j
and the drafted tokens
y
~
j
+
1
…
j
+
i
-
1
." The Verify (AT) model performing verification in parallel is considered analogosu to generating probability distributions for each set of tokens based on a single pass).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s autoregressive probability distribution comparisons within the method of claim 1 of the reference application.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 15.
Claim 20
Claim 9 of the reference application recites all of the limitations of claim 20 of the instant application.
Claim 21
Claim 1 of the reference application recites all of the limitations of claim 21 of the instant application except “an indication… based on the input query and the plurality of sets of tokens.”
However, Xia et al. disclose an indication of a selected set of tokens from the plurality of sets of tokens based on the input query and the plurality of sets of tokens (Xia et al. Figure 3, see row labelled "Output" which corresponds to verified tokens based on the set of drafted tokens and input sentence).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s indication based on the input query and the plurality of sets of tokens within the method of claim 1 of the reference application.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 1.
Claim 22
Claim 3 of the reference application recites all of the limitations of claim 22 of the instant application.
Claim 23
Claim 4 of the reference application recites all of the limitations of claim 23 of the instant application.
Claim 24
Claim 5 of the reference application recites all of the limitations of claim 24 of the instant application.
Claim 28
Claim 6 of the reference application recites all of the limitations of claim 28 of the instant application.
Claim 29
Claims 1 and 7 of the reference application recites all of the limitations of claim 29 of the instant application.
Claim 30
Claim 8 of the reference application recites all of the limitations of claim 30 of the instant application.
Claim 32
Claim 9 of the reference application recites all of the limitations of claim 32 of the instant application.
Claim 33
Claim 10 of the reference application recites all of the limitations of claim 33 of the instant application.
Claim 34
Claim 11 of the reference application recites all of the limitations of claim 34 of the instant application.
Claim 35
Claim 1 of the reference application recites all of the limitations of claim 35 of the instant application except “comparing a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens”
However, Xia et al. disclose compare a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
" See Figure 3; "Verify (AT)" is considered analogous to a second generative model. Thus, taking the maximum probability distrubtion among drafted tokens
y
~
j
+
1
…
j
+
k
is considered analogous to comparing probability distributions generated by the second generative model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s autoregressive probability distribution comparisons within the method of claim 1 of the reference application.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 15.
Claim 36
Claim 5 of the reference application recites all of the limitations of claim 36 of the instant application.
Claim 37
Claim 1 of the reference application recites all of the limitations of claim 37 of the instant application except “generating probability distributions for each respective set of tokens based on a single pass through the second generative model.”
However, Xia et al. disclose wherein to compare the probability distribution associated with each respective set of tokens in the plurality of sets of tokens to the corresponding probability distribution generated by the second generative model for the respective set of tokens, the one or more processors are configured to cause the processing system to generate probability distributions for each respective set of tokens based on a single pass through the second generative model (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
where ...
y
^
j
+
i
is the top-1 result verified by the AT model conditioning on the previously translated tokens
y
≤
j
and the drafted tokens
y
~
j
+
1
…
j
+
i
-
1
." The Verify (AT) model performing verification in parallel is considered analogosu to generating probability distributions for each set of tokens based on a single pass).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to include Xia et al.’s autoregressive probability distribution comparisons within the method of claim 1 of the reference application.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 15.
Claim 40
Claim 9 of the reference application recites all of the limitations of claim 40 of the instant application.
Claims 11, 19, 21, and 29 are provisionally rejected on the ground of nonstatutory double patenting as being unpatentable over claim 1 of co-pending Application No. 18/479,672 (reference application) in view of Xia et al. as applied to claim 1 of the instant application above, and further in view of “Accelerating Large Language Model Decoding with Speculative Sampling” (Chen et al.).
Claim 11
Claim 1 of the reference application recites all of the limitations of claim 11 of the instant application except “receiving a token generated by the second generative model based on the selected set of tokens; and outputting the received token as an additional token subsequent to the selected set of tokens.”
However, Chen et al. disclose receiving a token generated by the second generative model based on the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the received token as an additional token subsequent to the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Sampling an additional token in addition to the already-sampled tokens
x
n
+
1
,
…
,
x
n
+
K
is considered analogous to outputting the received token as an additional token subsequent to the selected set of tokens).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify claim 1 of the reference application to incorporate Chen et al.’s additional token sampling because such a modification is the result of combining prior art elements according to known methods to yield predictable results. More specifically, claim 1 of the reference application as modified by Chen et al.’s additional token sampling can yield a predictable result of increasing system efficiency since the method would always guarantee generating at least one additional token per loop, thus making further progress towards a final output. Thus, a person of ordinary skill would have appreciated including in claim 1 of the reference application the ability to do Chen et al.’s additional token sampling since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
Claim 19
Claim 1 of the reference application recites all of the limitations of claim 19 of the instant application except “generating an additional token based on the selected set of tokens using the second generative model; and outputting the additional token to the first generative model.”
However, Chen et al. disclose generating an additional token based on the selected set of tokens using the second generative model (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the additional token to the first generative model (Chen et al. pg. 4, Section "Modified Rejection Sampling", Paragraphs 4, "If the token is accepted, we set
x
n
+
1
←
x
~
n
+
1
and repeat the process for
x
~
n
+
2
until either a token is rejected or all tokens have been accepted"; pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Algorithm 2 illustrates the above step executing within a while loop. Thus, if the while loop's condition is not met, it can be inferred that the additional token is re-inputted into the draft model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify claim 1 of the reference application to incorporate Chen et al.’s additional token sampling.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 11.
Claim 31
Claim 1 of the reference application recites all of the limitations of claim 31 of the instant application except “receiving a token generated by the second generative model based on the selected set of tokens; and outputting the received token as an additional token subsequent to the selected set of tokens.”
However, Chen et al. disclose receiving a token generated by the second generative model based on the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the received token as an additional token subsequent to the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Sampling an additional token in addition to the already-sampled tokens
x
n
+
1
,
…
,
x
n
+
K
is considered analogous to outputting the received token as an additional token subsequent to the selected set of tokens).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify claim 1 of the reference application to incorporate Chen et al.’s additional token sampling.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 11.
Claim 39
Claim 1 of the reference application recites all of the limitations of claim 39 of the instant application except “generating an additional token based on the selected set of tokens using the second generative model; and outputting the additional token to the first generative model.”
However, Chen et al. disclose generating an additional token based on the selected set of tokens using the second generative model (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the additional token to the first generative model (Chen et al. pg. 4, Section "Modified Rejection Sampling", Paragraphs 4, "If the token is accepted, we set
x
n
+
1
←
x
~
n
+
1
and repeat the process for
x
~
n
+
2
until either a token is rejected or all tokens have been accepted"; pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Algorithm 2 illustrates the above step executing within a while loop. Thus, if the while loop's condition is not met, it can be inferred that the additional token is re-inputted into the draft model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify claim 1 of the reference application to incorporate Chen et al.’s additional token sampling
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 11.
Claim Rejections - 35 USC § 101
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 1-40 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. All of the claims are method claims (21-40), apparatus/machine claims (1-20) or manufacture claim under (Step 1), but under Step 2A all of these claims recite abstract ideas and specifically mental processes. These mental processes are more particularly recited in claims 1, 15, 21, and 35 as:
generating, based on an input query and a first generative model, a plurality of sets of tokens…
outputting, to a second generative model, the plurality of sets of tokens for verification…
receiving, from the second generative model, an indication of a selected set of tokens from the plurality of sets of tokens…
comparing a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens…
selecting a set of tokens from the plurality of sets of tokens based on the comparing…
Under Step 2A Prong One, claims 1, 15, 21, and 35 are directed to an abstract idea and specifically a mental process. As detailed above, the steps of generating, outputting, receiving, comparing, selecting, etc. may be practically performed in the human mind with the use of a physical aid such as a pen and paper. For example, a human could receive an input query, draft a plurality of sets of tokens based on the input query, estimate a confidence distribution for each set of tokens based on their assumption that the respective set of tokens respond accurately to the input query, select the set of tokens with the highest confidence distribution, and then respond to the input query with the selected set of tokens.
Under Step 2A Prong Two, this judicial exception is not integrated into a practical application because claims 1-40 do not recite additional elements that integrate the exception into a practical application. In particular, claims 1, 15, 21, and 35 recite the additional elements of a processor (¶ [0130]-[0132]), memory (¶ [0133]), and generative models (¶ [0025]-[0026]). These additional elements are recited at a high level of generality and merely equate to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). Further, claims 1, 15, 21, and 35 recite the additional elements of “receiving…” and “outputting…”, both of which amount to insignificant extra-solution activities which are not indicative of integration into a practical application as per MPEP 2106.05(g). Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
Under Step 2B, the claims do not recite additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional elements of using a computer is noted as a general computer {processor (¶ [0130]-[0132]); memory (¶ [0133])}. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Further, the additional limitations in the claims noted above are directed towards insignificant extra-solution activities. The claims are not patent eligible.
With respect to claims 2, 17, 22, and 37, the claim relates to each set of tokens comprising a group of tokens with the highest probabilities within a probability distribution. This relates to a human assigning individual confidence values to each group of tokens within a set of tokens corresponding to confidence distribution. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 3 and 23, the claim relates to selecting a set of tokens based on the sum of their probabilities exceeding a sum threshold. This relates to a human selecting the set of tokens with the highest summed confidence scores, comparing the sum of confidence scores to a predetermined threshold. No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 4-6, 16, 24-26, and 36 the claim relates to representing the plurality of sets of tokens as a tree data structure. This relates to a human drafting a plurality of sets of tokens according to a tree data structure, ensuring that root node of the tree data structure corresponds to the input query and each path corresponds to a set of tokens. The human could also ensure that the depth of the tree data structure corresponds to the maximum number of tokens drafted, and that the size of the tree is limited to their mental capacity to draft tokens. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 7 and 27, the claim relates to pruning the tree data structure to generate a subsequent plurality of sets of tokens and selecting a subsequent set of tokens for responding to the input query. This relates to a human pruning their tree data structure to only keep the most confident tokens, drafting additional sets of tokens based off of the pruned tree data structure, and then selecting a set of tokens from the additional plurality of sets of tokens for responding to the input query. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 8 and 28, the claim relates to the first generative model being a unique instance, and taking as input unique parameters. This relates to a human basing their token drafting on their current mental state, which is unique to the human in that moment. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 9-10 and 29-30, the claim relates to generating a subsequent plurality of sets of tokens while a second generative model verifies the first plurality of sets of tokens, generating a refined subsequent set of tokens, and then outputting the refined subsequent set of tokens to the second generative model for verification. This relates to a human relegating verification/selection of a set of tokens to a second human, drafting a refined plurality of sets of tokens while the second human verifies/selects the initial plurality of sets of tokens, selecting a refined set of tokens from the refined plurality of sets of tokens, and then handing the refined set of tokens to the second human for further verification/selection. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 11, 19, 31, and 39, the claim relates to generating an additional token based on the selected set of tokens subsequent to the selected set of tokens. This relates to a human sampling and concatenating an additional draft token to the already-selected set of tokens. The additional limitations of a “generative model” is recited at a high level of generality (¶ [0025]-[0026]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 12, 20, 32, and 40, the claim relates to the first and second generative models corresponding to a draft and target model in a speculative decoding pipeline, respectively. The limitations of a “model” and a “speculative decoding pipeline” is recited at a high level of generality (¶ [0025]-[0028]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 13 and 33, the claim relates to the draft model trained to have a probability distribution that approximates the target model’s probability distribution. This relates to a human adjusting their confidence distributions to align better with a second human’s confidence distributions of the plurality of sets of tokens. The limitations of a “model” is recited at a high level of generality (¶ [0025]-[0028]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 14 and 34, the claim relates to the draft model and target model executing on a local and remote system respectively. This relates to the human being a first human within vicinity of the user who created the input query, and the human relegating verification/selection duties to a second human who is not located within the vicinity of the user who created the input query. The limitations of a “model” and a “system” is recited at a high level of generality (¶ [0025]-[0028], ¶ [0060]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claims 18 and 38, the claim relates to the second generative model being trained on masked self-attention and positional encodings in a tree data structure. This relates to the human following the procedures of masked self-attention to generate the confidence distributions, as well as basing their verification/selection on the positions of each set of tokens within a tree data structure. The limitations of a “model” is recited at a high level of generality (¶ [0025]-[0028], ¶ [0060]) and merely equates to “apply it” or otherwise merely uses a generic computer as a tool to perform an abstract which are not indicative of integration into a practical application as per MPEP 2106.05(f). No additional limitations are present. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
For all of the above reasons, taken alone or in combination, claims 1-40 recite a non-statutory mental process.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-2, 8, 15, 17, 21-22, 28, 35, and 37 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by “Lossless Speedup of Autoregressive Translation with Generalized Aggressive Decoding” (Xia et al.).
Claim 1
Regarding claim 1, Xia et al. disclose a processing system, comprising:
a memory having executable instructions stored thereon (Xia et al. Appendix E, "GAD-base only costs about an additional 400MB GPU memory which is negligible for a modern GPU."); and
one or more processors configured to execute the executable instructions (Xia et al. Table 10, "The results are obtained with fp32 computation on a single Nvidia P100 GPU.") to cause the processing system to:
generate, based on an input query and a first generative model, a plurality of sets of tokens (Xia et al. pg. 3-4, Section 3, Paragraph 2, "Formally, given the source sentence
x
=
(
x
1
,
x
2
,
…
,
x
n
)
and the previously translated tokens
y
=
(
y
1
,
y
2
,
…
,
y
j
)
, GAD decodes the next
k
(drafted) tokens as a block in parallel:
y
~
j
+
1
…
j
+
k
=
a
r
g
m
a
x
y
~
j
+
1
…
j
+
k
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
" See Figure 3; "Draft (NAT)" is considered analogous to a first generative model), each set of tokens in the plurality of sets of tokens corresponding to a candidate response to the input query (Xia et al. Figure 3 illustrates a plurality of sets of tokens (e.g. "sind die phys@@") corresponding to a candidate response (see row labelled "Output"));
output to a second generative model, the plurality of sets of tokens for verification (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel" See Figure 3; "Verify (AT)" is considered analogous to a second generative model);
receive, from the second generative model, an indication of a selected set of tokens from the plurality of sets of tokens based on the input query and the plurality of sets of tokens (Xia et al. Figure 3, see row labelled "Output" which corresponds to verified tokens based on the set of drafted tokens and input sentence); and
output the selected set of tokens as a response to the input query (Xia et al. Figure 3, row "Output" corresponds to an indication of the selected set of tokens).
Claim 2
Regarding claim 2, the rejection of claim 1 is incorporated.
Xia et al. further disclose wherein each set of tokens in the plurality of sets of tokens comprises a group of tokens having the highest probabilities within a probability distribution associated with the first generative model over a universe of tokens (Xia et al. pg. 4, Section 3, Paragraph 3, "
y
^
j
+
i
is the top-1 result verified by the AT model conditioning on the previously translated tokens
y
≤
j
and the drafted tokens
y
~
j
+
1
…
j
+
i
-
1
." See Equation 5, which illustrates selecting
y
^
j
+
i
based on the maximum probability within a probability distribution of drafted tokens. The set of drafted tokens
y
~
j
+
1
…
j
+
i
-
1
is considered analogous to a universe of tokens).
Claim 8
Regarding claim 8, the rejection of claim 1 is incorporated.
Xia et al. further disclose wherein each respective set of tokens in the plurality of sets of tokens is generated using a unique instance of the first generative model and unique parameters as inputs into the unique instance of the first generative model (Xia et al. pg. 3-4, Section 3, Paragraph 2, "GAD decodes the next
k
(drafted) tokens as a block in parallel:
y
~
j
+
1
…
j
+
k
=
a
r
g
m
a
x
y
~
j
+
1
…
j
+
k
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
" The input of previously translated tokens
y
≤
j
causes every inference of
y
~
j
+
1
…
j
+
k
to be a unique instance.
θ
N
A
T
is considered analogous to unique parameters).
Claim 15
Regarding claim 15, Xia et al. disclose a processing system, comprising:
a memory having executable instructions stored thereon (Xia et al. Appendix E, "GAD-base only costs about an additional 400MB GPU memory which is negligible for a modern GPU."); and
one or more processors configured to execute the executable instructions (Xia et al. Table 10, "The results are obtained with fp32 computation on a single Nvidia P100 GPU.") to cause the processing system to:
receive, from a system on which a first generative model operates, an input query and a plurality of sets of tokens (Xia et al. pg. 3-4, Section 3, Paragraph 2, "Formally, given the source sentence
x
=
(
x
1
,
x
2
,
…
,
x
n
)
and the previously translated tokens
y
=
(
y
1
,
y
2
,
…
,
y
j
)
, GAD decodes the next
k
(drafted) tokens as a block in parallel:
y
~
j
+
1
…
j
+
k
=
a
r
g
m
a
x
y
~
j
+
1
…
j
+
k
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
" See Figure 3; "Draft (NAT)" is considered analogous to a first generative model), each respective set of tokens in the plurality of sets of tokens corresponding to a respective candidate response to the input query (Xia et al. Figure 3 illustrates a plurality of sets of tokens (e.g. "sind die phys@@") corresponding to a candidate response (see row labelled "Output"));
compare a probability distribution associated with each respective set of tokens in the plurality of sets of tokens to a corresponding probability distribution generated by a second generative model for the respective set of tokens (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
" See Figure 3; "Verify (AT)" is considered analogous to a second generative model. Thus, taking the maximum probability distrubtion among drafted tokens
y
~
j
+
1
…
j
+
k
is considered analogous to comparing probability distributions generated by the second generative model);
select a set of tokens from the plurality of sets of tokens based on the comparing (Xia et al. pg. 4, Section 3, Paragraph 3, "
y
^
j
+
i
is the top-1 result verified by the AT model conditioning on the previously translated tokens
y
≤
j
and the drafted tokens
y
~
j
+
1
…
j
+
i
-
1
."); and
output, to the first generative model, an indication of the selected set of tokens (Xia et al. Figure 3, row "Output" corresponds to an indication of the selected set of tokens).
Claim 17
Regarding claim 17, the rejection of claim 15 is incorporated.
Xia et al. further disclose wherein to compare the probability distribution associated with each respective set of tokens in the plurality of sets of tokens to the corresponding probability distribution generated by the second generative model for the respective set of tokens, the one or more processors are configured to cause the processing system to generate probability distributions for each respective set of tokens based on a single pass through the second generative model (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel. As the original IAD, we find the bifurcation position
c
by comparing the drafted tokens with the autoregressive decoding results conditioning on the draft as Figure 2 shows: ...
y
^
j
+
i
=
a
r
g
m
a
x
y
^
j
+
i
log
P
(
y
^
j
+
i
|
y
≤
j
,
y
~
j
+
1
…
j
+
i
-
1
;
θ
A
T
)
where ...
y
^
j
+
i
is the top-1 result verified by the AT model conditioning on the previously translated tokens
y
≤
j
and the drafted tokens
y
~
j
+
1
…
j
+
i
-
1
." The Verify (AT) model performing verification in parallel is considered analogous to generating probability distributions for each set of tokens based on a single pass).
Claim 21
Regarding claim 21, the limitations of claim 21 are similar to that of claim 1 and therefore are rejected for similar reasons as described above.
Claim 22
Regarding claim 22, the rejection of claim 21 is incorporated. The limitations of claim 22 are similar to that of claim 2 and therefore are rejected for similar reasons as described above.
Claim 28
Regarding claim 28, the rejection of claim 21 is incorporated. The limitations of claim 28 are similar to that of claim 8 and therefore are rejected for similar reasons as described above.
Claim 35
Regarding claim 35, the limitations of claim 35 are similar to that of claim 15 and therefore are rejected for similar reasons as described above.
Claim 37
Regarding claim 37, the rejection of claim 35 is incorporated. The limitations of claim 37 are similar to that of claim 17 and therefore are rejected for similar reasons as described above.
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 3, 11-13, 19-21, 23, 31-33, and 39-40 are rejected under 35 U.S.C. 103 as obvious over Xia et al. as applied to claims 1, 15, 21, and 35 above, and further in view of “Accelerating Large Language Model Decoding with Speculative Sampling” (Chen et al.).
Claim 3
Regarding claim 3, the rejection of claim 1 is incorporated.
Xia et al. further disclose wherein each set of tokens in the plurality of sets of tokens comprises a group of tokens selected based on a sum of probabilities associated with tokens in the group of tokens (Xia et al. pg. 3-4, Section 3, Paragraph 2, "Formally, given the source sentence
x
=
(
x
1
,
x
2
,
…
,
x
n
)
and the previously translated tokens
y
=
(
y
1
,
y
2
,
…
,
y
j
)
, GAD decodes the next
k
(drafted) tokens as a block in parallel:
y
~
j
+
1
…
j
+
k
=
a
r
g
m
a
x
y
~
j
+
1
…
j
+
k
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
"
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
is considered analogous to a sum of probabilities associated with tokens in a group of tokens), [the sum exceeding a threshold probability].
Xia et al. do not explicitly disclose all of a sum exceeding a threshold probability.
However, Chen et al. disclose wherein each set of tokens in the plurality of sets of tokens comprises a group of tokens selected based on [a sum of] probabilities associated with tokens in the group of tokens, the [sum] probabilities exceeding a threshold probability (Chen et al. pg. 4, Section "Modified Rejection Sampling", Paragraphs 1-3, "We require a method to recover the distribution of the target model from samples from the draft model, and logits of said tokens from both models. To achieve this, we introduce the following rejection sampling scheme of the drafted tokens. Given a sequence of tokens
x
1
,
…
,
x
n
, and
K
draft tokens
x
~
n
+
1
,
…
,
x
~
n
+
K
generated from
P
(
.
|
.
)
, we accept
x
~
n
+
1
with probability:
min
1
,
q
(
x
~
n
+
1
|
x
1
,
…
,
x
n
)
p
(
x
~
n
+
1
|
x
1
,
…
,
x
n
)
where
q
(
x
~
n
+
1
|
x
1
,
…
,
x
n
)
and
p
x
~
n
+
1
x
1
,
…
,
x
n
are the probability of
x
~
n
+
1
according to the target and draft models respectively, conditioned on the context so far."
min
1
,
q
(
x
~
n
+
1
|
x
1
,
…
,
x
n
)
p
(
x
~
n
+
1
|
x
1
,
…
,
x
n
)
is considered analogous to a threshold probability).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al.’s token selection to incorporate Chen et al.’s threshold probability.
The suggestion/motivation for doing so would have been that, “By applying this sequentially, we recover the distribution of the target model for the accepted tokens (see proof in Theorem 1) within hardware numerics,” as noted by the Chen et al. disclosure in pg. 4, Section “Modified Rejection Sampling,” Paragraph 7.
Claim 11
Regarding claim 11, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of a sampling an additional token.
However, Chen et al. disclose receiving a token generated by the second generative model based on the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the received token as an additional token subsequent to the selected set of tokens (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Sampling an additional token in addition to the already-sampled tokens
x
n
+
1
,
…
,
x
n
+
K
is considered analogous to outputting the received token as an additional token subsequent to the selected set of tokens).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al.’s token selection to incorporate Chen et al.’s additional token sampling.
The suggestion/motivation for doing so would have been that, “Since the final token of the draft gives us the logits for the next token, if every drafted token is accepted, we can sample from it normally. This gives us a maximum of
K
+
1
tokens per loop, over the naive implementation which would only return
K
tokens … Because we do not interact with the body of the transformer itself, this method can be used in conjunction many other techniques for accelerating or optimising the memory use of sampling, such as quantisation and multi-query attention,” as noted by the Chen et al. disclosure in pg. 4, Section “Modified Rejection Sampling,” Paragraph 7-9.
Claim 12
Regarding claim 12, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of a speculative decoding pipeline.
However, Chen et al. disclose wherein:
the first generative model corresponds to a draft model in a speculative decoding pipeline (Chen et al. pg. 1, Section "Introduction", Paragraph 3, "We present an algorithm to accelerate transformer sampling for latency critical applications, which we call speculative sampling (SpS). This is achieved by: 1. Generating a short draft of length
K
. This can be attained with either a parallel model (Stern et al., 2018) or by calling a faster, auto-regressive model
K
times. We shall refer to this model as the draft model, and focus on the case where it is auto-regressive."), and
the second generative model corresponds to a target model in the speculative decoding pipeline (Chen et al. pg. 1, Section "Introduction", Paragraph 3, "We present an algorithm to accelerate transformer sampling for latency critical applications, which we call speculative sampling (SpS). This is achieved by: ... 2. Scoring the draft using the larger, more powerful model from we wish to sample from. We shall refer to this model as the target model.").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. to incorporate Chen et al.’s speculative sampling.
The suggestion/motivation for doing so would have been, “to accelerate transformer sampling for latency critical applications,” as noted by the Chen et al. disclosure in pg. 1, Section "Introduction", Paragraph 3.
Claim 13
Regarding claim 13, the rejection of claim 12 is incorporated.
Xia et al. further disclose wherein the [draft] first generative model comprises a model trained to have a probability distribution that approximates a corresponding probability distribution for the target model (Xia et al. pg. 4, Section 3.1, Paragraph 1, "Formally, given the source sentence
x
=
(
x
1
,
…
,
x
n
)
and the randomly sampled prefix
y
≤
p
(
0
≤
p
<
m
)
of the target sentence
y
=
(
y
1
,
…
,
y
m
)
, the model is trained to predict the next k tokens, as shown in Figure 2:
L
N
A
T
=
∑
i
=
p
+
1
p
+
k
log
P
(
y
i
|
y
≤
p
,
x
;
θ
N
A
T
)
").
Chen et al. further disclose a draft model (Chen et al. pg. 1, Section "Introduction", Paragraph 3, "We present an algorithm to accelerate transformer sampling for latency critical applications, which we call speculative sampling (SpS). This is achieved by: 1. Generating a short draft of length
K
. This can be attained with either a parallel model (Stern et al., 2018) or by calling a faster, auto-regressive model
K
times. We shall refer to this model as the draft model, and focus on the case where it is auto-regressive.").
Claim 19
Regarding claim 19, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of a sampling an additional token.
However, Chen et al. disclose generating an additional token based on the selected set of tokens using the second generative model (Chen et al. pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
"); and
outputting the additional token to the first generative model (Chen et al. pg. 4, Section "Modified Rejection Sampling", Paragraphs 4, "If the token is accepted, we set
x
n
+
1
←
x
~
n
+
1
and repeat the process for
x
~
n
+
2
until either a token is rejected or all tokens have been accepted"; pg. 3, Algorithm 2, lines 19-20, "If all tokens
x
n
+
1
,
…
,
x
n
+
K
are accepted, sample extra token
x
n
+
K
+
1
~
q
(
x
|
,
x
1
,
…
,
x
n
,
x
n
+
K
)
and set
n
←
n
+
1
" Algorithm 2 illustrates the above step executing within a while loop. Thus, if the while loop's condition is not met, it can be inferred that the additional token is re-inputted into the draft model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al.’s token selection to incorporate Chen et al.’s additional token sampling.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 11.
Claim 20
Regarding claim 20, the rejection of claim 15 is incorporated. The limitations of claim 20 are similar in scope to that of claim 12 and therefore are rejected for similar reasons as described above.
Claim 23
Regarding claim 23, the rejection of claim 21 is incorporated. The limitations of claim 23 are similar in scope to that of claim 3 and therefore are rejected for similar reasons as described above.
Claim 31
Regarding claim 31, the rejection of claim 21 is incorporated. The limitations of claim 31 are similar in scope to that of claim 11 and therefore are rejected for similar reasons as described above.
Claim 32
Regarding claim 32, the rejection of claim 21 is incorporated. The limitations of claim 32 are similar in scope to that of claim 12 and therefore are rejected for similar reasons as described above.
Claim 33
Regarding claim 33, the rejection of claim 22 is incorporated. The limitations of claim 33 are similar in scope to that of claim 13 and therefore are rejected for similar reasons as described above.
Claim 39
Regarding claim 39, the rejection of claim 35 is incorporated. The limitations of claim 39 are similar in scope to that of claim 19 and therefore are rejected for similar reasons as described above.
Claim 40
Regarding claim 40, the rejection of claim 35 is incorporated. The limitations of claim 40 are similar in scope to that of claim 20 and therefore are rejected for similar reasons as described above.
Claims 4, 16, 24, and 36 are rejected under 35 U.S.C. 103 as obvious over Xia et al. as applied to claims 1, 15, 21, and 35 above, and further in view of “Planning with Large Language Models for Code Generation” (Zhang et al.).
Claim 4
Regarding claim 4, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of a tree data structure.
However, Zhang et al. disclose wherein:
the plurality of sets of tokens are represented as a tree data structure (Zhang et al. pg. 4, Section 3.1, Paragraphs 4, "a state
s
is the concatenation of the problem description and a partial or complete program, where a complete program ends with a special terminal token. An action
a
is a token in the vocabulary set of the Transformer. ... In this paper, we consider a tree search-based planning algorithm inspired by Monte-Carlo tree search (MCTS), illustrated in Fig. 2. Intuitively, the tree search algorithm maintains a tree structure where nodes correspond to states and edges correspond to actions."),
a root node of the tree data structure corresponds to the input query (Zhang et al. pg. 5, Figure 2, "<PD> stands for problem description." See Figure 2, which illustrates the tree data structure with <PD> as the tree's root node. See Figure 1 for an exemplary problem description. A problem description is considered analogous to an input query), and
each path through the tree data structure corresponds to a set of tokens from the plurality of sets of tokens (Zhang et al. pg. 4, Section 3.1, Paragraphs 4, "a state
s
is the concatenation of the problem description and a partial or complete program, where a complete program ends with a special terminal token. An action
a
is a token in the vocabulary set of the Transformer" Figure 2 illustrates tree data structures wherein each path of each tree data structure corresponds to a set of tokens (e.g. "a=") from a plurality of a sets of tokens (e.g. "a="; "a,"; "x,")).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. to incorporate Zhang et al.’s tree data structure for searching and sampling tokens.
The suggestion/motivation for doing so would have been that, “[Monte-Carlo tree search] is more efficient and applicable to large domains when combined with deep learning or with a default policy as prior knowledge. We consider the same overall recipe in our work,” as noted by the Zhang et al. disclosure in pg. 3, Section 2 Paragraph 4.
Claim 16
Regarding claim 16, the rejection of claim 15 is incorporated. The limitations of claim 16 are similar in scope to that of claim 4 and therefore are rejected for similar reasons as described above.
Claim 24
Regarding claim 24, the rejection of claim 21 is incorporated. The limitations of claim 24 are similar in scope to that of claim 4 and therefore are rejected for similar reasons as described above.
Claim 36
Regarding claim 36, the rejection of claim 35 is incorporated. The limitations of claim 36 are similar in scope to that of claim 4 and therefore are rejected for similar reasons as described above.
Claims 5-7, and 25-27 are rejected under 35 U.S.C. 103 as obvious over Xia et al. in view of Zhang et al. as applied to claims 4 and 24 above, and further in view of “The Gumbel-Top-k Trick for Sampling Sequences Without Replacement” (Kool et al.).
Claim 5
Regarding claim 5, the rejection of claim 4 is incorporated.
Xia et al. in view of Zhang et al. do not explicitly disclose all of a tree data structure whose depth corresponds to a maximum number of tokens generated.
However, Kool et al. disclose wherein a depth of the tree data structure corresponds to a maximum number of tokens generated (Kool et al. pg. 2, Section 2.5, Paragraphs 3, "A sequence model defines a valid probability distribution over both partial and complete sequences. ... If the model is additionally conditioned on a context
x
(e.g., a source sentence), we write
p
θ
(
y
|
x
)
." pg. 3, Section 3.1, Paragraphs 1-2, "We represent the sequence model (7) as a tree (as in Figure 1), where internal nodes at level
t
represent partial sequences
y
1
:
t
, and leaf nodes represent completed sequences. ... As instantiating the complete probability tree is computationally prohibitive, we construct an equivalent process that only requires computation linear in the number of samples
k
and the sequence length." See Figure 1, which illustrates a tree of depth=3 and sample size
k
=3. Sample size
k
is considered analogous to a maximum number of generated tokens) by a single pass through the first generative model (Kool et al. pg. 5, Section 3.3, Paragraphs 1, "Stochastic Beam Search executes in a single pass, and requires computation linear in the sample size
k
and the sequence length").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. in view of Zhang et al. to incorporate Kool et al.’s tree data structure because such a modification is the result of simple substitution of one known element for another producing a predictable result. More specifically, Zhang et al.’s tree data structure and Kool et al.’s tree data structure perform the same general and predictable function, the predictable function being using tree-based searching in order to plan the best response to an input query. Since each individual element and its function are shown in the prior art, albeit shown in separate references, the difference between the claimed subject matter and the prior art rests not on any individual element or function but in the very combination itself - that is in the substitution of Zhang et al.’s tree data structure by replacing it with Kool et al.’s tree data structure. Thus, the simple substitution of one known element for another producing a predictable result renders the claim obvious.
Claim 6
Regarding claim 6, the rejection of claim 4 is incorporated.
Xia et al. in view of Zhang et al. do not explicitly disclose all of a tree data structure whose size is based a computational complexity metric.
However, Kool et al. disclose wherein a maximum size of the tree data structure is set (Kool et al. pg. 4, Section 3.1 "Stochastic Beam Search", Paragraph 1, "Formally, the
k
-th highest log-probability of the nodes at level t provides a lower bound required to be among the top
k
leaves, while
G
ϕ
S
is an upper bound for the set of leaves
S
such that it can be discarded or pruned if it is lower than the lower bound, so if
G
ϕ
S
is not among the top
k
." See Figure 1, which illustrates how the maximum size of a tree data structure is influenced by the value of
k
) based on a computational complexity metric associated with generating a target set of tokens by the second generative model (Kool et al. pg. 5, Section 3.3, Paragraphs 1, "Stochastic Beam Search ... requires computation linear in the sample size
k
and the sequence length, which (except for the beam search overhead) is equal to the computational requirement for sampling with replacement." The computational requirement for sampling with replacement is considered analogous to a computational complexity metric. Sample size
k
is considered analogous to generating a target set of tokens by a generative model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. in view of Zhang et al. to incorporate Kool et al.’s tree data structure.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 5.
Claim 7
Regarding claim 7, the rejection of claim 4 is incorporated.
Xia et al. further disclose generating a [subsequent] plurality of sets of tokens based on [the pruned tree data structure and] the input query (Xia et al. pg. 3-4, Section 3, Paragraph 2, "Formally, given the source sentence
x
=
(
x
1
,
x
2
,
…
,
x
n
)
and the previously translated tokens
y
=
(
y
1
,
y
2
,
…
,
y
j
)
, GAD decodes the next
k
(drafted) tokens as a block in parallel:
y
~
j
+
1
…
j
+
k
=
a
r
g
m
a
x
y
~
j
+
1
…
j
+
k
∑
i
=
1
k
log
P
(
y
~
j
+
1
|
y
≤
j
,
x
;
θ
N
A
T
)
" See Figure 3; "Draft (NAT)" is considered analogous to a first generative model;
outputting, to the second generative model, the [subsequent] plurality of sets of tokens for verification (Xia et al. pg. 4, Section 3, Paragraph 3, "Then, the drafted tokens
y
~
j
+
1
…
j
+
k
are verified with an AT model in the autoregressive manner, which performs in parallel" See Figure 3; "Verify (AT)" is considered analogous to a second generative model); and
receiving, from the second generative model, an indication of a [subsequent] selected set of tokens from the [subsequent] plurality of sets of tokens based on the input query, [the pruned tree data structure,] and the [subsequent] plurality of sets of tokens (Xia et al. Figure 3, see row labelled "Output" which corresponds to verified tokens based on the set of drafted tokens and input sentence), wherein outputting the selected set of tokens as the response to the input query comprises outputting the selected set of tokens [and the subsequent selected set of tokens] as the response to the input query (Xia et al. Figure 3, see row labelled "Output" which corresponds to verified tokens based on the set of drafted tokens and input sentence).
Zhang et al. further disclose generating a [subsequent] plurality of sets of tokens based on the pruned tree data structure and the input query (Zhang et al. pg. 21, Appendix D.1, Paragraph 2-4, "The algorithm first selects a node for expansion. It starts from the root node and selects subtrees recursively until it reaches a node that has not been expanded before (with no children). ... In the following steps, the algorithm expands the node by adding its children to the tree and evaluates the node using
B
E
A
M
_
S
E
A
R
C
H
, as we described in the main paper." Expanding a node by adding its children to the tree is considered analogous to generating a subsequent plurality of sets of tokens based on a tree data structure and input query. See Figures 12-13 for an exemplary expansion);
outputting, to the second generative model, the [subsequent] plurality of sets of tokens for verification (Zhang et al. pg. 6, Section 3.2, Paragraph 5, "In the evaluation step, we need to evaluate the selected node. Note that node may still be a partial program. ... Here, we use the Transformer again by calling the
B
E
A
M
_
S
E
A
R
C
H
function to generate a complete program from the current node, where
B
E
A
M
_
S
E
A
R
C
H
(
s
,
b
)
generates a sequence using the Transformer beam search algorithm with the prefix
s
and beam size
b
." The generated sequence using Transformer beam search is considered analogous to a the subsequent plurality of sets of tokens. Thus, evaluating a node is considered analogous to verifying a plurality of sets of tokens); and
receiving, from the second generative model, an indication of a [subsequent] selected set of tokens from the [subsequent] plurality of sets of tokens based on the input query, the [pruned] tree data structure, and the [subsequent] plurality of sets of tokens (Zhang et al. pg. 6, Section 3.2, Paragraph 5, "
B
E
A
M
_
S
E
A
R
C
H
(
s
,
b
)
generates a sequence using the Transformer beam search algorithm with the prefix
s
and beam size
b
. We run the generated program on the public test cases to get its reward, and set it to be the value of node (Line 16-17). This value is backpropagated up in the tree so that the values of its ancestors are updated (Line 20)." Assigning and backpropagating a reward value based on the evaluation of a node is considered analogous to an indication), wherein outputting the selected set of tokens as the response to the input query comprises outputting the selected set of tokens [and the subsequent selected set of tokens] as the response to the input query (Zhang et al. pg. 5, Algorithm 1, line 22, "return program in
p
r
o
g
r
a
m
_
d
i
c
t
with the highest reward").
Xia et al. in view of Zhang et al. do not explicitly disclose all of pruning a data tree structure for subsequent generations.
However, Kool et al. disclose pruning the tree data structure based on the selected set of tokens (Kool et al. pg. 4, Section 3.1 "Stochastic Beam Search", Paragraph 1, "Formally, the
k
-th highest log-probability of the nodes at level
t
provides a lower bound required to be among the top
k
leaves, while
G
ϕ
S
is an upper bound for the set of leaves
S
such that it can be discarded or pruned if it is lower than the lower bound, so if
G
ϕ
S
is not among the top
k
." See Figure 1, which illustrates a tree data structure with pruned nodes (dotted nodes) based on the selected set of tokens (shaded nodes)); and
generating a subsequent plurality of sets of tokens based on the pruned tree data structure (Kool et al. pg. 4, Section 3.1 "Top-down sampling of perturbed log-probabilities", Paragraph 2, "Note that for the root
N
(since it contains all leaves
N
), it holds that
ϕ
N
=
log
∑
i
ϵ
N
exp
ϕ
i
=
0
, so we can let
G
ϕ
N
~
G
u
m
b
e
l
(
0
)
2
. Starting with
S
=
N
, we can recursively sample the children conditionally on the parent variable
G
ϕ
S
. For
S
'
ϵ
C
h
i
l
d
r
e
n
(
S
)
it holds that
ϕ
S
'
=
log
p
θ
(
y
S
'
)
and we can sample
G
ϕ
S
'
~
G
u
m
b
e
l
(
ϕ
S
'
)
conditionally on [Equation 10], e.g. with their maximum equal to
G
ϕ
S
." Recursively sampling a plurality of children tokens is considered analogous to generating a subsequent plurality of sets of tokens based on the pruned tree data structure) [and the input query];
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. in view of Zhang et al. to incorporate Kool et al.’s tree data structure.
The suggestion/motivation for doing so is similar to the suggestion/motivation described above with respect to claim 5.
Claim 25
Regarding claim 25, the rejection of claim 21 is incorporated. The limitations of claim 25 are similar in scope to that of claim 5 and therefore are rejected for similar reasons as described above.
Claim 26
Regarding claim 26, the rejection of claim 21 is incorporated. The limitations of claim 26 are similar in scope to that of claim 6 and therefore are rejected for similar reasons as described above.
Claim 27
Regarding claim 27, the rejection of claim 21 is incorporated. The limitations of claim 27 are similar in scope to that of claim 7 and therefore are rejected for similar reasons as described above.
Claims 9-10 and 29-30 are rejected under 35 U.S.C. 103 as obvious over Xia et al. as applied to claims 1 and 21 above, and further in view of “Blockwise Parallel Decoding for Deep Autoregressive Models” (Stern et al.).
Claim 9
Regarding claim 9, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of verifying and generating plurality of sets of tokens in parallel.
However, Stern et al. disclose wherein:
while the second generative model verifies the plurality of sets of tokens, generate a subsequent plurality of sets of tokens based on the input query and the plurality of sets of tokens (Stern et al. pg. 3, Section 4, Paragraph 3, "suppose we have a single Transformer model which during the verification substep computes
p
i
(
y
j
+
i
'
+
i
|
y
^
≤
j
+
i
'
,
x
)
for all
i
=
1
,
…
,
k
and
i
'
=
1
,
…
,
k
in a constant number of operations. ... Invoking the model after plugging in the
k
future predictions from the prediction substep yields the desired outputs." See Figure 2, which illustrates performing the verification step and subsequent predictions in parallel);
based on the subsequent plurality of sets of tokens and the selected set of tokens, generate a refined subsequent set of tokens (Stern et al. Figure 2 illustrates a refined subsequent set of tokens; see last "Predict" step); and
output, to the second generative model, the refined subsequent set of tokens for verification (Stern et al. pg. 2, Section 3, Paragraph 2, "we start with an empty prediction ^y and set j = 0. Then we repeat the following three substeps until the termination condition is met: Predict... Verify... [and] Accept" It can be inferred that by repeatedly performing the above steps, the output of the last "Predict" step in Figure 2 will be output to the Verify model).
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al. to incorporate Stern et al.’s parallel verification and prediction.
The suggestion/motivation for doing so would have been that, “Our approach exhibits relatively little loss in quality compared to prior work. We achieve a BLEU score within 0.29 of the original Transformer with a real-time speedup over our baseline exceeding 3x,” as noted by the Stern et al. in the description of Table 4. See Table 4, which illustrates how parallel decoding achieves a wall-clock speedup compared to the traditional Transformer method.
Claim 10
Regarding claim 9, the rejection of claim 1 is incorporated.
Xia et al. further disclose wherein sets of tokens in the subsequent plurality of sets of tokens include padding accounting for a number of tokens in the selected set of tokens being less than a maximum number of tokens (Xia et al. pg. 5, Section 4, Paragraph 2, "GAD++ loosens the criterion to trust NAT’s draft more, by only requiring the drafted tokens to fall in
t
o
p
-
β
candidates with a tolerable (log-likelihood) score gap
τ
(away from the top-1 result)" See Figure 3, which illustrates in the verification of drafted tokens that sets of tokens with less than 3 tokens are filled with
[
b
l
a
n
k
]
symbols.
[
b
l
a
n
k
]
is considered analogous to padding).
Claim 29
Regarding claim 29, the rejection of claim 21 is incorporated. The limitations of claim 29 are similar in scope to that of claim 9 and therefore are rejected for similar reasons as described above.
Claim 30
Regarding claim 30, the rejection of claim 29 is incorporated. The limitations of claim 30 are similar in scope to that of claim 10 and therefore are rejected for similar reasons as described above.
Claims 14 and 34 are rejected under 35 U.S.C. 103 as obvious over Xia et al. as applied to claims 1 and 21 above, and further in view of US Patent Publication 20130253908 A1 (Zhai et al.).
Claim 14
Regarding claim 14, the rejection of claim 1 is incorporated.
Xia et al. do not explicitly disclose all of the first and second generative models being local and remote respectively.
However, Zhai et al. disclose wherein:
the first generative model comprises a model executing on a local system (Zhai et al. ¶ [0051]-[0053], "A local computing device may store a local language model. ... A local computing device may analyze 212 the context using the local language model and may determine 214 one or more suggested words based on the analysis."), and
the second generative model comprises a model executing on a remote system (Zhai et al. ¶ [0045], "In an embodiment, the remote computing device may receive 204 at least a portion of a context and/or context-related information from a local computing device. The remote computing device may analyze 206 the context and/or the context-related information using the remote language model and may determine 208 one or more remote suggested words based on the analysis.").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al.’s method to incorporate Zhai et al.’s separation of generative language models.
The suggestion/motivation for doing so would have been that, “Although more memory and processing power generally increases the accuracy of input prediction and power of correction, mobile devices often have limited memory with which to host very large language models. A server-based solution may alleviate the memory and processing power constraints, but the time delay that occurs in waiting for a response from a remote server often makes a pure server-based approach an unreliable text input solution,” as noted by the Zhai et al. disclosure in paragraph [0002].
Claim 34
Regarding claim 34, the rejection of claim 21 is incorporated. The limitations of claim 34 are similar in scope to that of claim 14 and therefore are rejected for similar reasons as described above.
Claims 18 and 38 are rejected under 35 U.S.C. 103 as obvious over Xia et al. as applied to claims 17 and 37 above, and further in view of “QueryFormer: A Tree Transformer Model for Query Plan Representation” (Zhao et al.).
Claim 18
Regarding claim 18, the rejection of claim 17 is incorporated.
Xia et al. further disclose wherein the single pass through the second generative model is performed based on masked self-attention (Xia et al. pg. 4, Section 3.2, Paragraph 1, "We use the conventional autoregressive Transformer (see Section 2.1) as our AT verifier, which is the key to guaranteeing the translation quality." The use of Transformer architecture is considered analogous to masked self-attention).
Xia et al. do not explicitly disclose all of incorporating position encodings in a tree data structure.
However, Zhao et al. disclose masked self-attention (Zhao et al. pg. 1664, Section 5.2.2, Paragraph 2, "we introduce Tree-bias Attention as a modification to the original self-attention module. The goal is to use the tree structure to control the information flow in the self-attention module. Specifically, the idea is to mask out illegal information paths, and to account for the relative positions for legal information paths in self-attention module.") and positional encodings in a tree data structure (Zhao et al. pg. 1664, Section 5.2.1, Paragraph 2, "Inspired by the Positional Encoding method, we incorporate the height information into the node embedding. As illustrated in Figure 3, the learned height embedding for each height
E
h
is added to the node features before they are fed into the attention block.").
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify Xia et al.’s method to incorporate Zhao et al.’s positional encodings in a tree data structure.
The suggestion/motivation for doing so would have been that, “it automatically determines the best way to model the information flow from bottom to top with just a few parameters (a scalar value for each distance),” as noted by the Zhao et al. disclosure in pg. 1664, Section 5.2.1, Paragraph 2.
Claim 38
Regarding claim 38, the rejection of claim 37 is incorporated. The limitations of claim 38 are similar in scope to that of claim 18 and therefore are rejected for similar reasons as described above.
Reference Cited
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
“Fast Inference from Transformers via Speculative Decoding” to Leviathan et al. discloses a similar strategy to Chen et al. of implementing speculative decoding in Transformers.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JACOB B VOGT whose telephone number is (571)272-7028. The examiner can normally be reached Monday - Friday, 11am - 8pm 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, PARAS D SHAH can be reached at (571)270-1650. 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.
/JACOB B VOGT/ Examiner, Art Unit 2653
/Paras D Shah/ Supervisory Patent Examiner, Art Unit 2653
07/02/2026
Prosecution Insights