text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
limitations inherent in these techniques, pro-
viding a solid foundation for future research in
addressing hallucinations and related phenom-
ena within the realm of LLMs.
Introduction | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
17/18
02/05/2023, 07:05
Recent Comments
A brief history of LLaMA models - AGI Sphere
1. Andrew on Local ChatGPT on Mac – How to install Vicuna language model (2 ways)
2. Sadiq Khawaja on Local ChatGPT on Mac – How to install Vicuna language model (2 ways)
3. Sadiq Khawaja on Local ChatGPT on Mac – How to install V... | A brief history of LLaMA models - AGI Sphere |
-10.66), (-0.98, -9.22), (-0.98, -7.96), (-0.93, -6.74)]Object type: car, object id: 3, future waypoint coordinates in 3s: [(-25.19, -17.79), (-25.19, -17.79), (-25.18, -17.78), (-25.18, -17.78), (-25.18, -17.78), (-25.17, -17.78)] Figure 4. An example of the memory search process. Cont’d. | ALanguageAgentforAutonomousDriving |
4.1 Datasets and Evaluation Metrics
We evaluate our methods on eleven datasets across three categories of different reasoning tasks,
including (1) six arithmetic reasoning datasets: GSM8k (Cobbe et al., 2021a), AQuA (Ling et al.,
2017), AddSub (Hosseini et al., 2014), SingleEq (Koncel-Kedziorski et al., 2015), SVAMP (... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
http://arxiv.org/abs/1906.06669.
Taku Kudo. Subword Regularization: Improving Neural Network Translation Models with Mul-
In Proceedings of the 56th Annual Meeting of the Association
tiple Subword Candidates.
for Computational Linguistics (Volume 1: Long Papers), pp. 66–75, Melbourne, Australia,
July 2018. Association... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
The following discussion is broken into five parts: The first section explores
bots in the context of their general use online and then unpacks research that
examines their social use. The second looks into their political use and discusses
research on how to detect such use. The third details arguments on how bots can
a... | Social_Media_and_Democracy |
blocks within a 32-block distance;
• craftItem(bot, name, count = 1): Craft the item with a crafting table nearby;
• placeItem(bot, name, position): Place the block at the specified position;
• smeltItem(bot, itemName, fuelName, count = 1): Smelt the item with the
specified fuel. There must be a furnace nearby;
24
... | VOYAGER- An Open-Ended Embodied Agent with Large Language Models |
3.3. Additional Classification Tasks
To further validate that adapters yields compact, performant,
models, we test on additional, publicly available, text clas-
sification tasks. This suite contains a diverse set of tasks:
The number of training examples ranges from 900 to 330k,
the number of classes ranges from 2 to 15... | Parameter-Efficient Transfer Learning for NLP |
As an AI language model, I do not have personal preferences or opinions. However, based on
the given information, a customer can buy either the chocolate cake ($12, 400 calories) and
the vanilla cake ($10, 300 calories) or the strawberry cake ($8, 200 calories) and the vanilla
cake ($10, 300 calories). Both options wil... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
,
a
n
d
t
h
e
o
p
p
o
r
t
u
n
i
t
i
e
s
t
h
a
t
w
i
l
l
g
e
t
t
h
e
m
t
o
w
h
e
r
e
t
h
e
y
w
a
n
t
t
o
g
o
,
y
o
u
c
a
n
b
e
t
t
e
r
a
l
i
g
n
t
h
e
i
r
s
k
i
l
l
s
,
c
a
p
a
b
i
l
i
t
i
e
s
,
a
n
d
c
a
r
e
e
r
p
a
t
h
s
t
o
t
h
e
n
e
e
d
s
o
f
t
h
e
b
u
s
i
n
e
s
... | Senior Software Engineer, Machine Learning - Generative AI Job in Bellevue, WA at SeekOut |
166
Chloe Wittenberg & Adam J. Berinsky
misinformation is “information that is false, but not intended to cause harm”
(p. 5), whereas disinformation is “false information that is deliberately created
or disseminated with the express purpose to cause harm” (p. 4). Finally, a third
approach emphasizes the temporal natu... | Social_Media_and_Democracy |
we compute: (1) errors for SMPL-X meshes estimated from
linguistic shape attributes and/or anthropometric measure-
ments by A2S and its variations, and (2) errors for linguistic
shape attributes estimated from SMPL-X meshes by S2A.
To create an unseen mesh test set, we withhold 339 male
and 410 female CAESAR meshes fro... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
[111] Mireia Diez, Lukáš Burget, Federico Landini, Shuai Wang, and Honza Černock`y. 2020. Optimizing Bayesian HMM
based x-vector clustering for the second DIHARD speech diarization challenge. In ICASSP 2020-2020 IEEE International
Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 6519–6523.
A Revi... | AReviewofDeepLearningTechniquesforSpeechProcessing |
In Figure 4a, we evaluate each reward model by its best-of-500 selection. We
see that process supervision significantly outperforms both forms of outcome
supervision at all data collection scales.
In Figure 4b, we evaluate the best
reward model from each series by its best-of-N performance across different
values of N.... | Let’s Verify Step by Step |
5 Limitations and Risks
Limitations A lack of appropriate datasets for
evaluating the handling of extremely lengthy texts
has resulted in our model being validated solely
through manual verification. This method, how-
ever, is inadequate for evaluating different scenar-
ios comprehensively and objectively. Therefore,
we... | Unleashing Infinite-Length Input Capacity for Large-scale Language Models with Self-Controlled Memory System |
Training models with even lower precision. The best method we found to stabilize our models
without hurting (and sometimes improving) quality was the router z-loss. This is an auxiliary loss
that encourages the model logits to have values smaller in absolute magnitude. Given the max range
of numbers float32 and bfloat1... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
c2( f (s), f (t)) ≤ c2( f (s), f (u)) + c2( f (u), f (t)) ≤ c1(s, u) + c2( f (u), f (t)),
that is, c2 must also be a consistent heuristic for c1. Hence, there is no need to consider consistency explicitly.
Definition 46. An M↑ transformation τ = (cid:3) f , R, w1, w2(cid:4) can have the following metric properties:
A↓:... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Mocha and Match
Mocha is a young girl from high school. She has learned so much interesting
knowledge from her teachers, especially her math teacher. Recently, Mocha is
learning about binary system and very interested in bitwise operation.
This day, Mocha got a sequence 𝑎 of length 𝑛.
In each operation, she
can selec... | alphacode |
2. Fanfiction. Hundreds of GiB of fanfiction
has been written and put online, primarily
on the websites www.fanfiction.net
and www.https://archiveofourown.
org/. This represents a significant untapped
resource for language modeling as it is al-
most exclusively short-form fiction, a writing
style that is not represented in... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
In this section we investigate the influence of instruction finetuning on benchmarks measuring several
potential harms to end users, including toxic language harms, representational bias, and specific forms of
gender bias. We additionally investigate the impact of instruction finetuning on improving zero-shot and
few-shot ... | Scaling Instruction-Finetuned Language Models |
25
Perception from Neural NetsMeaningful Objects from Tool LibrarySensory DataNotable Objects from CoT. ReasoningTrajectory from Motion PlanningSensory DataPerception from Neural NetsMeaningful Objects from Tool LibraryNotable Objects from CoT. ReasoningTrajectory from Motion Planning Figure 11. Visualization of how A... | ALanguageAgentforAutonomousDriving |
Exploiting the scaling law. The scaling laws seem to bar us from making large gains via major
changes to the transformer size and type, as per-token performance is tightly coupled to model size.
As a result, we find no improvements when using a funnel-transformer architecture (Dai et al., 2020;
Nawrot et al., 2022), whe... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
instruction tuning. arXiv preprint arXiv:2301.13688, 2023.
[371] Wang, Y., Y. Kordi, S. Mishra, et al. Self-instruct: Aligning language model with self generated
instructions. arXiv preprint arXiv:2212.10560, 2022.
[372] Liang, J., W. Huang, F. Xia, et al. Code as policies: Language model programs for embodied
contr... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
[Nashid et al., 2023] Noor Nashid, Mifta Sintaha, and Ali
Mesbah. Retrieval-based prompt selection for code-related
few-shot learning. In 2023 IEEE/ACM 45th International
Conference on Software Engineering (ICSE), pages 2450–
2462, 2023.
[OpenAI, 2023] OpenAI. Gpt-4 technical report. https://cdn.
openai.com/papers/gp... | Retrieval-AugmentedGenerationforLargeLanguageModels-ASurvey |
26
The Efficiency Spectrum of Large Language Models: An Algorithmic Survey
Efficient LLM Algorithmic Survey, Nov, 2023, USA.
[58] Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V Le, and Ruslan Salakhutdinov. 2019. Transformer-xl: Attentive language models beyond a
fixed-length context. arXiv preprint... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
trait taxonomy: History, measurement, and conceptual issues. (2008)
[62] American Educational Research Association, American Psychological Asso-
ciation, National Council on Measurement in Education (eds.): Standards
for Educational and Psychological Testing. American Educational Research
Association, Lanham, MD (2014... | PersonalityTraitsinLargeLanguageModels |
Table 17 shows the results of our different methods. Both the additive and multiplicative biases are
essentially free: cheap to compute, adds few new parameters, and incurs no additional communi-
cation costs with model and expert parallelism. When using our router z-loss from Section 3.1, we
observe no instabilities f... | ST-MOE- DESIGNING STABLE AND TRANSFERABLE SPARSE EXPERT MODELS |
3. We collect TEXT2MUSIC, a dataset of 50K
text-music pairs constituting 2,500 hours of
music.
4. Our model outperforms existing baselines by
clear margins on 11 different evaluation cri-
teria, demonstrating merits such as high ef-
ficiency, text-music relevance, music quality,
and long-context structure.
2 Related ... | Moûsai |
3.7 Evolution of performance during training
During training, we tracked the performance of our
models on a few question answering and common
sense benchmarks, and report them in Figure 2.
On most benchmarks, the performance improves
steadily, and correlates with the training perplexity
of the model (see Figure 1). The... | LLaMA- Open and Efficient Foundation Language Models |
21
these challenges. Their investigations confirm the presence of contextual sparsity and
its potential for precise prediction, enabling us to leverage it to hasten LLM infer-
ence without sacrificing model quality or learning abilities in context. To capitalize on
these findings, they introduce Deja Vu, a system profici... | Beyond Efficiency |
REVEAL: Retrieval-Augmented Visual-Language Pre-Training with
Multi-Source Multimodal Knowledge Memory
Ziniu Hu1*, Ahmet Iscen2, Chen Sun2, Zirui Wang2, Kai-Wei Chang1, Yizhou Sun1
Cordelia Schmid2, David A. Ross2, Alireza Fathi2
1University of California, Los Angeles, 2Google Research
3
2
0
2
r
p
A
3
]... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
shelf language models of sufficient scale simply via prompting. This prompting setup is important
because it allows for intermediate step reasoning without a large number of labeled annotations, and
because a single model can perform a range of reasoning tasks without any gradient updates. | Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |
(4) In-Context Learning: Utilizing mixed-modality models opens up possibilities for the devel-
opment of in-context learning approaches for a wide range of speech-related tasks. This
paradigm allows the tasks to be explicitly defined within the input, along with accompa-
nying examples. Remarkable progress has already ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
- Heading Angular Velocity (v_yaw): (0.00) - Acceleration (ax,ay): (-0.00,-0.50) - Can Bus: (-0.74,0.14) - Heading Speed: (0.95) - Steering: (-0.02)Historical Trajectory (last 2 seconds): [(-0.07,-6.43), (-0.05,-4.34), (-0.02,-2.32), (-0.01,-0.91)]Mission Goal: FORWARDFront object detections:Front object detected, obje... | ALanguageAgentforAutonomousDriving |
A.1 Hyperparameters for Fine-tuning
If not otherwise specified, we fine-tune BiomedGPT with 50 epochs and a learning rate of 7e-5 in terms of
a batch size of 128, 64, and 32 for small-, medium-, and base-size models, respectively. The input image
resolution is set to 256×256, dropout (Srivastava et al., 2014) rate is ... | BiomedGPT |
A
U
T
H
O
R
S
S
t
e
v
e
n
B
i
l
l
s
,
N
i
c
k
C
a
m
m
a
r
a
t
a
,
D
a
n
M
o
s
s
i
n
g
,
H
e
n
k
T
i
l
l
m
a
n
,
L
e
o
G
a
o
,
G
a
b
r
i
e
l
G
o
h
,
I
l
y
a
S
u
t
s
k
e
v
e
r
,
J
a
n
L
e
i
k
e
,
J
e
f
f
W
u
,
W
i
l
l
i
a
m
S
a
u
n
d
e
r
s
*
C
o
r
e
R
e
s
e
a
r
c
h
C
o
n
t
r
i
b
u
t
o
r
;
A
u
t
... | Language models can explain neurons in language models |
[10] Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi, Yeganeh Kordi, Amirreza Mirzaei,
Atharva Naik, Arjun Ashok, Arut Selvan Dhanasekaran, Anjana Arunkumar, David Stap, Eshaan
Pathak, Giannis Karamanolakis, Haizhi Gary Lai, Ishan Virendrabhai Purohit, Ishani Mondal,
Jacob William Anderson, Kirby C. Kuznia, Krima ... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Kushal Tirumala, Daniel Simig, Armen Aghajanyan,
and Ari S Morcos. 2023. D4: Improving llm pretrain-
ing via document de-duplication and diversification.
arXiv preprint arXiv:2308.12284.
Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier
Martinet, Marie-Anne Lachaux, Timothée Lacroix,
Baptiste Rozière, Naman Goyal,... | DataManagementForLargeLanguageModels-ASurvey |
[587] Choi, M., J. Pei, S. Kumar, et al. Do llms understand social knowledge? evaluating the
sociability of large language models with socket benchmark. CoRR, abs/2305.14938, 2023.
[588] Wilson, A. C., D. V. Bishop. " if you catch my drift...": ability to infer implied meaning is
distinct from vocabulary and grammar ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
• audio tokens: the SoundStream tokens representing au-
dio.
6
Likewise, the model outputs two types of tokens: visual
tokens and audio tokens. In addition to video and audio to-
kens along with text embeddings, we incorporate additional
special tokens enumerated as shown in Table 1.
Special Token
Usage
<bos>
<t... | VideoPoet |
Denny Zhou, Nathanael Sch¨arli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuur-
mans, Claire Cui, Olivier Bousquet, Quoc Le, and Ed Chi. Least-to-Most Prompting Enables
Complex Reasoning in Large Language Models, April 2023. URL http://arxiv.org/
abs/2205.10625. arXiv:2205.10625 [cs].
Hattie Zhou, Azade No... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
The goal of this section is to model a number of different abstraction and abstraction-like methods from the literature
within our framework. We note that even though the methods are quite different, they can all be modelled in a highly
uniform and reasonably succinct way. For instance, labels wil... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
13The complete results are presented in Tables 7, 8, 9 and 10 in the Appendix.
14The complete results on unseen entities are presented in Tables 11 and 12 in the Appendix.
15The results in WNUT 2017 are almost the same, because almost all entities are unseen.
9
withoutwith0.00.51.0F1-score0.60.740.650.73CoNLLeswithou... | MULTI HASH EMBEDDINGS IN SPACY |
02/05/2023, 15:33
Google AI updates: Bard and new AI features in Search
Feb 06, 2023 · min read
4
Sundar Pichai
AI is the most profound technology we are working on today. Whether it’s helping doctors detect diseases earlier
or enabling people to access information in their own language, AI helps people, business... | Google AI updates_ Bard and new AI features in Search |
but this does not reflect true generalization ability. The
model trained on the large dataset combination achieves
very strong scores across the board, confirming that using
many datasets makes a difference. | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
language models. CoRR, abs/2308.11339, 2023.
[408] Nair, V., E. Schumacher, G. J. Tso, et al. DERA: enhancing large language model completions
with dialog-enabled resolving agents. CoRR, abs/2303.17071, 2023.
[409] Talebirad, Y., A. Nadiri. Multi-agent collaboration: Harnessing the power of intelligent LLM
agents. ... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol. Extracting and composing robust
features with denoising autoencoders. In Proceedings of the 25th international conference
on Machine learning, pages 1096–1103, 2008. 4, 6, 14
P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, P.-A. Manzagol, and L. Bottou. Stacke... | A Cookbook of Self-Supervised Learning |
Scholars have also identified analytical thinking, or a person’s capacity to
override gut feelings and intuitions, as another determinant of their responses to
misinformation. In this sense, individuals who are more prone to careful,
deliberate processing of information (or “cognitive reflection”) seem to be less
suscept... | Social_Media_and_Democracy |
Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems 33 (2020), 1877–1901.
[3] Paul-Christian Bürkner. 2017. brms: An R package for Bayesian multilevel models using Stan. Journal of statistical software 80 (2017), 1–28.
[4] Bob Carpenter, Andrew Gelman, Ma... | Adoptionand AppropriationofLLMs |
In order for abstraction to be useful, the abstract instance must be easier to solve and the total time spent should be
less than without using abstraction. This is a reasonable requirement, yet it has turned out very difficult to guarantee. Ab-
straction refinement can give huge savings in solution time under ideal circ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
13
Efficient LLM Algorithmic Survey, Nov, 2023, USA.
Ding, Chen, et al.
capture the cyclical patterns in token relationships. By diminishing attention between distant positions, these methods ensure
the model’s focus remaining on the more immediate and contextually relevant tokens rather than the tokens that are fa... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
introduces an active re-
trieval approach, triggered by the LM’s generation of low-
probability words. It creates a temporary sentence for doc-
ument retrieval, then regenerates the sentence with the re-
trieved context to predict subsequent sentences. RETRO uses
the previous chunk to retrieve the nearest neighbor at t... | RAG forLargeLanguageModels-ASurvey |
Overreliance occurs when users excessively trust and depend on the model, potentially leading
to unnoticed mistakes and inadequate oversight. This can happen in various ways: users may not be
vigilant for errors due to trust in the model; they may fail to provide appropriate oversight based on
the use case and context;... | gpt-4-system-card |
3 x + b3)
4 EXPERIMENTAL SETUP
The main goal of our experiments is to benchmark our hash embedding implementation,
MultiHashEmbed, on different settings and scenarios against traditional word embeddings. This
section outlines the datasets we used as well as our model architecture. We tested on a variety of
named enti... | MULTI HASH EMBEDDINGS IN SPACY |
This sort of dynamic applies to very few of the technologies we’re familiar with (disciplines like
computer security, which involve actively anticipating the strategies available to adversaries, may
be the closest analog). That is: planes, rockets, nuclear plants, and so forth may be dangerous and
complicated—but they ... | Is Power-Seeking AI an Existential Risk? |
Keller and Leerssen warn of high rates of false positives in both filtering and human
review of content. Moreover, in the face of vague legal directives, platforms tend to
overcensor to avoid liability, a finding that takes on added urgency in view of
President Trump’s May 2020 Executive Order on Preventing Online Censor... | Social_Media_and_Democracy |
To investigate the appropriate insertion strategy for LoRA, we conduct three sets of associable instruction tuning experi-
ments using Oogiri-GO I2T data. LoRA is inserted separately into the textual, visual, and both textual and visual modules of
Qwen-VL. Experimental results indicate that, based on the 3T1 metric, th... | Let’sThinkOutsidetheBox |
crime risk operations by leveraging generative AI and LLMs. Genpact is accelerating efficiencies and impact
for their clients by integrating their proprietary cloud-based financial crime suite with Amazon Bedrock. | AMZN-Q3-2023-Earnings-Release |
Angela Fan, Thibaut Lavril, Edouard Grave, Armand Joulin, and Sainbayar Sukhbaatar. Addressing some
limitations of transformers with feedback memory. arXiv preprint arXiv:2002.09402, 2020.
Alex Graves, Greg Wayne, and Ivo Danihelka. Neural turing machines. arXiv preprint arXiv:1410.5401, 2014.
Alex Graves, Greg Wayn... | Scaling Transformer to 1M tokens and beyond with RMT |
[34] Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen.
Progressive growing of gans for improved quality, stability,
and variation. In International Conference on Learning Rep-
resentations, 2018.
[35] Tero Karras, Samuli Laine, and Timo Aila. A style-based
generator architecture for generative adversarial net... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
h
t
o
c
o
n
s
t
r
u
c
t
m
u
l
t
i
-
e
p
i
s
o
d
e
h
i
s
t
o
r
y
.
M
u
l
t
i
-
e
p
i
s
o
d
i
c
c
o
n
t
e
x
t
s
o
f
2
-
4
e
p
i
s
o
d
e
s
a
r
e
n
e
c
e
s
s
a
r
y
t
o
l
e
a
r
n
a
n
e
a
r
-
o
p
t
i
m
a
l
i
n
-
c
o
n
t
e
x
t
R
L
a
l
g
o
r
i
t
h
m
.
T
h
e
e
m
e
r
g
e
n
c
e
o
f
i
n
-
c
... | LLM Powered Autonomous Agents _ Lil'Log |
i
n
t
e
r
p
r
e
t
i
n
g
n
e
u
r
a
l
n
l
p
:
T
h
e
c
a
s
e
o
f
g
e
n
d
e
r
b
i
a
s
V
i
g
,
J
.
,
G
e
h
r
m
a
n
n
,
S
.
,
B
e
l
i
n
k
o
v
,
Y
.
,
Q
i
a
n
,
S
.
,
N
e
v
o
,
D
.
,
S
a
k
e
n
i
s
,
S
.
,
H
u
a
n
g
,
J
.
,
S
i
n
g
e
r
,
Y
.
a
n
d
S
h
i
e
b
e
r
,
S
.
,
2
0
2
... | Language models can explain neurons in language models |
fθ0(x; ϕ(t)) ≈f lin
θ0
(x; ϕ(t)) = fθ0 (x; ϕ(0))
+ ∇ϕfθ0(x; ϕ(0))T (ϕ(t) − ϕ(0)).
(15)
D. Hybrid Fine-Tuning
Hybrid fine-tuning approaches aim to combine various
PEFT approaches, such as adapter, prefix-tuning, and LoRA,
to leverage the strengths of each method and mitigate their
weaknesses. By integrating differ... | Parameter-EfficientFine-TuningMethods |
gradient values have a tendency to underflow in FP16. Underflow can cause weights to receive either no
gradient or low-precision, eccentric gradients, which can further exacerbate dynamic loss scale and underflow.
Underflows and Weight Growth: We detect underflows by observing any significant increase in the
number of identi... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
Language models can explain neurons in language models
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
11/32 | Language models can explain neurons in language models |
e
s
u
l
t
s
,
a
n
d
p
r
i
o
r
i
t
i
z
i
n
g
t
a
s
k
s
i
n
r
e
a
l
-
t
i
m
e
.
2
.
2
P
I
N
E
C
O
N
E
12/04/2023, 14:50 | Task-driven Autonomous Agent Utilizing GPT-4, Pinecone, and LangChain for Diverse Applications – Yohei Nakajima |
Then, we summarize the success and failure cases of LLMs in different tasks. Finally, we shed light on several | ASurveyonEvaluationofLargeLanguageModels |
Code Llama - Instruct
Code Llama - Python
Size
Multi-lingual Human-Eval
TS
C#
PHP
C++ Java
Bash Average
16B 21.0% 22.2% 8.4% 20.1% 8.2% 0.6% 13.4%
13B 16.9% 19.1% 13.5% 10.1% 8.5% 2.8% 11.8%
12B 30.6% 31.9% 28.9% 31.3% 22.1% 11.7% 26.1%
15.5B 30.6% 28.5% 26.8% 32.2% 20.6% 11.0% 25.0%
15.5B 31.6% 30.2% 26.1% 32.... | CodeLlama2 |
"""
def small_nnum(lst,n):
lst = sorted(lst)
lst = lst[:n]
return lst
Feedback: With the above function, small_nnum([10, 20, 50, 70, 90, 20, 50,
40, 60, 80, 100],2)==[10,20]. The assertion is "small_nnum([10, 20, 50, 70,
90, 20, 50, 40, 60, 80, 100],2)==[10,20]". So the code passes the assertion.
The code above is co... | Teaching Large Language Models to Self-Debug |
Given K training poses with J joints{Pk∈ RJ×3}K
k=1
Lreconstr+ λsparseLsparse
Wenc∈RL×J , Wdec∈RJ×L
∥Pk− WdecWencPk∥
+∥Wdec∥
s. t. Wenc1J= 1L, Wdec1L= 1J ,
Lreconstr= 1
K∑
k=1
Lsparse=∥Wenc∥
minimize
(3)
K
,
1
1
1
where 1a is a vector of dimension a filled with ones and
λsparse controls the strength of the spa... | Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats |
3. In domains other than TYREWORLD, LLM-AS-P fails in the same way with or without the
example plan as context. In particular, in the BLOCKSWORLD domain, LLM-AS-P cannot
keep track of properties like ON and CLEAR. In the GRIPPERS domain, the robot can only
pick up balls when they are in the same room, but most of the L... | LLM+P- Empowering Large Language Models with Optimal Planning Proficiency |
11
solve-rate is an additional indication that it has not encountered such problems
via test set contamination. Our generalization results from Section 5 further
strengthen our claim that test set contamination has not significantly impacted
this work, since we observe qualitatively similar results on problems that a... | Let’s Verify Step by Step |
References
[1] Sameer Agarwal, Yasutaka Furukawa, Noah Snavely, Ian Si-
mon, Brian Curless, Steven M Seitz, and Richard Szeliski.
Building rome in a day. Communications of the ACM, 2011.
1
[2] Marc Badger, Yufu Wang, Adarsh Modh, Ammon Perkes,
Nikos Kolotouros, Bernd Pfrommer, Marc Schmidt, and
Kostas Daniilidis. 3D b... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
3
2
0
2
r
p
A
7
1
]
L
C
.
s
c
[
1
v
4
5
3
8
0
.
4
0
3
2
:
v
i
X
r
a
Tool Learning with Foundation Models
Yujia Qin1, Shengding Hu1, Yankai Lin2∗, Weize Chen1, Ning Ding1, Ganqu Cui1, Zheni Zeng1,
Yufei Huang1, Chaojun Xiao1, Chi Han3, Yi Ren Fung3, Yusheng Su1, Huadong Wang1,
Cheng Qian1, Runchu T... | Tool Learning with Foundation Models |
• Agent-Driver integrates a tool library for dynamic per-
ception and prediction, a cognitive memory for human
knowledge, and a reasoning engine that emulates human
decision-making, all orchestrated by LLMs to enable a
more anthropomorphic autonomous driving process.
• Agent-Driver significantly outperforms the state-... | ALanguageAgentforAutonomousDriving |
2
Figure 2: Evolution of performance when scaling in parameters. We show performance on eight
types of vision tasks, as presented in Sec. 7, and average metrics with each type. Features are extracted
from our self-supervised encoders, DINOv2 (dark blue), and we compare them with self-supervised methods
(pale orange),... | DINOv2- Learning Robust Visual Features without Supervision |
11.2.2 Model-Based Metrics.
Auxiliary Decoder. “Faithfulness” refers to the amount of source meaning that is faithfully
expressed in the translation, and it is used interchangeably with the term “adequacy” [49, 186]. Feng
et al. [49] propose adding another “evaluation decoder” apart from the standard translation decod... | SurveyofHallucinationinNatural Language Generation |
with Attributes Database’’ (AADB), which contains aesthetic
scores and high-level visual attributes assigned to each image
by multiple human raters. The original AlexNet softmax clas-
sification layer is replaced with an Euclidean Loss regression
layer and attribute prediction branches are added on top of
the second ful... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
To train an ASR model in Section 5.5, we extract 80-dimensional log Mel features with a 25ms
window and a 10ms frame shift, and then apply global mean-variance normalization. The ASR
model is an RNN-T with a Conformer-based encoder [Gulati et al., 2020]. The conformer applies
time scale reduction to the input features ... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
Success
Rate
Eval
Times
Language Instruction
0.1772
0.2584
0.2469
0.0159
0.0759
0.2278
0.2239
79
89
81
63
79
79
67
Mine redstone and make piston.
Mine redstone and make redstone_torch.
Mine redstone and make redstone_block.
Mine redstone and make activator_rail.
Mine redstone and make compass.
Mine redstone and ma... | JARVIS-1 |
tasks and domains, could be combined with subsymbolic approaches and their ability to deal with large amounts of data, to
handle noise, and to capture the richness of perceptual data. In this sense, it is natural to hypothesise that a neuro-symbolic
integration could also support explainable systems to be more explai... | Knowledge graphs as tools for explainable machine learning: A survey |
MultiHashEmbed
MultiEmbed
Precision
0.59±0.02
0.61±0.02
0.21±0.04
0.33±0.01
0.26±0.03
0.54±0.02
Recall
0.60±0.01
0.57±0.00
0.10±0.01
0.38±0.00
0.18±0.02
0.58±0.01
F1-score
0.60±0.01
0.59±0.02
0.14±0.01
0.35±0.03
0.21±0.02
0.56±0.01
Precision
0.64±0.02
0.63±0.03
0.23±0.02
0.30±0.01
0.32±0.02
0.62±0.01
Recall
0.64±... | MULTI HASH EMBEDDINGS IN SPACY |
DominikS (Stammbach, 2021)
focuses primar-
ily on sentence-level evidence retrieval, scoring
individual tokens from a given Wikipedia doc-
ument, and then selecting the highest scoring
sentences by averaging token scores. It uses a
fine-tuned document level BigBird model (Zaheer
et al., 2020) for this purpose. For clai... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
A. Owens, J. Wu, J. H. McDermott, W. T. Freeman, and A. Torralba. Ambient sound
provides supervision for visual learning. In Computer Vision–ECCV 2016: 14th European
Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14,
pages 801–816. Springer, 2016. 5
A. Painsky, M. Feder, and N. Tishby... | A Cookbook of Self-Supervised Learning |
Going beyond online news consumption, Barnidge (2017) offers a useful
comparison of how US adults report being exposed to political disagreement in
different settings. His study relies on survey data which, at the expense of
potential reporting biases, has the advantage of allowing a comparison of
offline interactions a... | Social_Media_and_Democracy |
Aakanksha Chowdhery, Sharan Narang, Jacob Devlin,
Maarten Bosma, Gaurav Mishra, Adam Roberts,
Paul Barham, Hyung Won Chung, Charles Sutton,
Sebastian Gehrmann, Parker Schuh, Kensen Shi,
Sasha Tsvyashchenko, Joshua Maynez, Abhishek
Rao, Parker Barnes, Yi Tay, Noam Shazeer, Vin-
odkumar Prabhakaran, Emily Reif, Nan Du, B... | LaMini-LM- A Diverse Herd of Distilled Models from Large-Scale Instructions |
Constructing a well-suited training dataset,
which we define as data management, is vitally
important and challenging in both the pretraining
and supervised fine-tuning (SFT) stages of LLMs.
In the pretraining stage, constructing datasets with
high-quality and the most useful data is essential
for efficient training (J... | DataManagementForLargeLanguageModels-ASurvey |
line of best fit with and without active learning, we estimate that this form
of active learning is approximately 2.6x more data efficient than uniform data
labelling. We note that the model trained on the largest active learning dataset
(200 samples per problem) appears to slightly underperform the expected trend
line... | Let’s Verify Step by Step |
If the perfect neural network were to descend on us, we might discover through
extensive testing that it worked; it would take still another stage of scientific discovery
to understand how it worked. If we discover some neural network that succeeds and it
turns out that its constituents should happen to map perfectl... | The Next Decade in AI- |
Accuracy
Skip to Primary Navigation
Skip to Main Content
undefined
Sign In
We evaluate the model performance via a set of zero-shot classification tasks. The model is a CLIP Vision
model ([2103.00020] Learning Transferable Visual Models From Natural Language Supervision (arxiv.org) ) that
learns a matching b... | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
7 Conclusion
In this work, we presented SELF-DEBUGGING, which enables a large language model to debug
code generated by itself. In particular, we demonstrate that SELF-DEBUGGING empowers the
model to perform rubber duck debugging, so that the model can identify and fix the bugs without
human instructions. SELF-DEBUGGIN... | Teaching Large Language Models to Self-Debug |
To evaluate the robustness of fact verification
systems against the impact of superfluous informa-
tion from the retriever, we propose a new metric,
Stability Error Rate (SER), which measures the
proportion of instances where superfluous in-
formation changes the decision of the model.
ProoFVer achieves a SER of 5.73%,... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas
Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit,
and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In
ICLR, 2020.
Dheeru Dua, Yizhong... | gemini_1_report |
Indeed, I think that one of the central reasons we should expect to see practically PS-misaligned AI
systems getting used/deployed is precisely that they will demonstrate a high degree of usefulness
during training/testing—and consequently, it will be increasingly difficult to resist deploying them,
especially in the co... | Is Power-Seeking AI an Existential Risk? |
methods. IEEE Transactions on Intelligent Vehicles, 6(2):195–209, 2020.
Andrea Madotto, Zhaojiang Lin, Chien-Sheng Wu, and Pascale Fung. Personalizing dialogue agents via meta-
learning. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp.
5454–5459, Florence, Italy, 2019. As... | Tool Learning with Foundation Models |
responsible for integrating and organizing responses from all agents, thus updating the final answer
[447]. However, consolidating a large amount of feedback data and extracting valuable insights poses
a significant challenge for the coordinating agent.
Furthermore, majority voting can also serve as an effective approa... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
This research is not intended as a survey of the whole field of eXplainable AI and Knowledge Representation, but has
a particular focus on the advantages and limitations of using knowledge graphs as support and background knowledge for
explainable systems. In particular, we present the following contributions:
• we pr... | Knowledge graphs as tools for explainable machine learning: A survey |
[236] Naoyuki Kanda, Jian Wu, Yu Wu, Xiong Xiao, Zhong Meng, Xiaofei Wang, Yashesh Gaur, Zhuo Chen, Jinyu Li, and
Takuya Yoshioka. 2022. Streaming Speaker-Attributed ASR with Token-Level Speaker Embeddings. arXiv preprint
arXiv:2203.16685 (2022).
[237] Naoyuki Kanda, Xiong Xiao, Yashesh Gaur, Xiaofei Wang, Zhong Meng,... | AReviewofDeepLearningTechniquesforSpeechProcessing |
In addition to AI-focused providers, traditional software
and cloud service providers are expanding their offerings to
include RAG-centric services. Verba13 from Weaviate is de-
signed for personal assistant applications, while Amazon’s
Kendra14 provides an intelligent enterprise search service, al-
lowing users to nav... | RAG forLargeLanguageModels-ASurvey |
n
i
n
g
w
i
t
h
A
l
g
o
r
i
t
h
m
D
i
s
t
i
l
l
a
t
i
o
n
”
I
C
L
R
2
0
2
3
.
[
1
0
]
K
a
r
p
a
s
e
t
a
l
.
“
M
R
K
L
S
y
s
t
e
m
s
A
m
o
d
u
l
a
r
,
n
e
u
r
o
-
s
y
m
b
o
l
i
c
a
r
c
h
i
t
e
c
t
u
r
e
t
h
a
t
c
o
m
b
i
n
e
s
l
a
r
g
e
l
a
n
g
u
a
g
e
m
o
d
e
l
s
,
e
x
t
e
r
n
a
l
... | LLM Powered Autonomous Agents _ Lil'Log |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.