text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
[44] Georgios Pavlakos, Luyang Zhu, Xiaowei Zhou, and Kostas
Daniilidis. Learning to estimate 3D human pose and shape
from a single color image. In Computer Vision and Pattern
Recognition (CVPR), pages 459–468, 2018. 3
[45] Gerard Pons-Moll, Javier Romero, Naureen Mahmood, and
Michael J. Black. Dyna: A model of dynami... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
✓ ✓
✓
1000
10000
12
585
11.2
20
1800
9283
Language
Dataset
English
TIMIT Acoustic-Phonetic Continuous Speech Corpus
English
Lip Reading Sentences 2 (LRS2)
English
LibriSpeech (LS)
English
GigaSpeech
Multilingual
Fleurs
English
LibriTTS
English
L2ARCTIC
English
CMUARCTIC
English
Wall Street Journal (WSJ)
Multilingu... | AReviewofDeepLearningTechniquesforSpeechProcessing |
4.3. Quantitative Evaluation
User study. We sample 20 unseen hand-drawn sketches,
and then assign each sketch to 5 methods: PITI [89]’s sketch
model, Sketch-Guided Diffusion (SGD) [88] with default
edge-guidance scale (β = 1.6), SGD [88] with relatively
high edge-guidance scale (β = 3.2), the aforementioned
ControlNet... | AddingConditionalControltoText-to-ImageDiffusionModels |
Finetuning language models on a collection of datasets phrased as instructions has been shown to improve
model performance and generalization to unseen tasks. In this paper we explore instruction finetuning
with a particular focus on (1) scaling the number of tasks, (2) scaling the model size, and (3) finetuning on
chain... | Scaling Instruction-Finetuned Language Models |
< 1 and s2 = 1. Therefore the parameter error is
What kind of domains are downweighted?
Intuitively, we can downweight the very noisy (high
entropy/difficulty) domain 3 because the initialization perfectly matches the ground truth. This
allows us to reallocate samples to the other domains 1 and 2. Between these, domain... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
Candidate GenerationNamed Entity RecognitionCommonsense ClassificationFill in The BlankText CompletionSentence CompositionTitle GenerationSummarizationQuestion UnderstandingNIv211020501005001000Question GenerationExplanation GenerationGenerate Question and AnswerQuestion AnsweringGrade School Math Word ProblemsAlgebrai... | Scaling Instruction-Finetuned Language Models |
1) Pose Generalization. For surface geometry reconstruc-
tion, the SMPL input can be regarded as an initial guess
for the network, which helps to factor out pose changes
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
5
Fig. 3. Comparison of the traditional reconstruction loss and our proposed depth-ambig... | PaMIR- Parametric Model-Conditioned Implicit Representation for Image-based Human Reconstruction |
organizations. One of the main challenges in such collaborative work is the impact of unforeseen
events that can disrupt any activity within a value chain. Against this background, digital twins (DT)
offer a solution to help organizations make decisions under uncertainties to better manage value
chains. DTs are mode... | informatics-phd-projects-2022-23 |
In this paper, we introduce a new text-conditioning space
that is dependent on both the denoising process timestep and
the U-Net layers, which we call P∗. The P∗ space is com-
posed of a set of vectors pt ∈ P+, one for each timestep
t. This results in a richer representation space that consid-
ers the time-dependent na... | A Neural Space-Time Representation for Text-to-Image Personalization |
performance to be similar to the bottleneck proposed in
Section 2.1. Therefore, due to its simplicity and strong per-
formance, we recommend the original adapter architecture. | Parameter-Efficient Transfer Learning for NLP |
Scaling law results (Kaplan et al., 2020; Brown et al., 2020;
Hoffmann et al., 2022) have been a major driving factor
in the recent surge of research and investment into LLMs
(Ganguli et al., 2022a). Scaling laws allow us to precisely
predict some coarse-but-useful measures of how capable
future models will be as we sc... | Eight Things to Know about Large Language Models |
Therefore, based on the ablation results and ease of scaling inference, for the 34B and 70B Llama 2 models
we chose to use GQA instead of MQA.
Figure 24 shows how inference speed changed for the 30B GQA and MQA ablation models compared to the
MHA baseline, in an experiment using 8 x 80 GiB A100s with tensor parallelism... | Llama2 |
2https://platform.openai.com/docs/models/gpt-3-5
7
nAcc@1
nAcc@2
nAcc@3
nAcc@1
nAcc@2
nAcc@3
Method
Random
Constant
TST-M
HyperSTAR
ASKL
FLAML
LLM-ZS
LLM-FS
MLCopilot
54.70
72.85
72.73
67.37
77.01
77.84
61.37
78.93
81.59
60.70
74.61
74.44
68.14
81.76
82.95
79.41
83.10
83.23
64.80
75.02
74.56
68.71
85.02
8... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
frequently used in online videos). The rendering resolution is 640x360, which is downsampled to 128x128 before being
input to the models. We empirically found 128x128 to be the smallest resolution for which in-game GUI elements are still
discernible, and then chose that to minimize compute costs. Whenever an in-game GU... | JARVIS-1 |
[1] Ramesh, Aditya, et al. "Hierarchical text-conditional image generation with clip latents."
arXiv preprint arXiv:2204.06125 (2022).
[2] Rombach, Robin, et al. "High-resolution image synthesis with latent diffusion models."
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti... | The Myth of Culturally Agnostic AI Models |
mengdan.zhu@emory.edu; yifei.zhang2@emory.edu;
j.carlyang@emory.edu; mrz7dp@virginia.edu;
Abstract
The burgeoning field of Large Language Models (LLMs), exemplified by sophis-
ticated models like OpenAI’s ChatGPT, represents a significant advancement
in artificial intelligence. These models, however, bring forth substan... | Beyond Efficiency |
9
[44] Zhenhua Xu, Yujia Zhang, Enze Xie, Zhen Zhao, Yong Guo,
Kenneth KY Wong, Zhenguo Li, and Hengshuang Zhao.
DriveGPT4: Interpretable End-to-end Autonomous Driving
via Large Language Model. arXiv preprint arXiv:2310.01412,
2023. 11
[45] Wenyuan Zeng, Wenjie Luo, Simon Suo, Abbas Sadat, Bin
Yang, Sergio Casas, an... | ALanguageAgentforAutonomousDriving |
mations inferred using anthropometric constraints for
human action classification (Maji et al., 2011). Hiyadi
and al. (Hiyadi et al., 2015) used the depht informa-
tion obtained from Kinect sensor and a tracking algo-
rithm for 3D human gestures recognition. Jian (Jiang,
2010) proposed an exemple-based method, based on
... | VISAPP_HumanPoseEstimation |
sampling from the marginals of X, i.e. ˜x ∼(cid:81)d
1Not to be confused with RF variants that employ non-adaptive
splits, which are sometimes also referred to as unsupervised, since
they ignore the response variable. See, e.g., Genuer (2012).
or synthetic (Y = 0). The method has expected accuracy
1/2 in the worst ca... | Adversarial Random Forests for Density Estimation and Generative Modeling |
(cid:21)
(10)
Eπθ(y|x)
max
πθ
rϕ(x, y) − β log
πref(y | x) exp
rϕ(x, y)
(cid:20)
(cid:124)
This is the same objective optimized in prior works [49, 38, 1, 26] using the DPO-equivalent reward
for the reward class of rϕ. In this setting, we can interpret the normalization term in f (rϕ, πref, β)
as the soft valu... | Direct Preference Optimization |
more feasible, enabling faster development cycles and the exploration of more complex model architectures. | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
11
Arch.
ViT-g/14
Res. Linear Finetuned
224
448
86.5
86.7
88.5
88.9
∆
+2.0
+2.2
Table 5: Supervised finetuning on ImageNet-1k. We use the pipeline of Touvron et al. (2022) to
finetune our encoders on ImageNet-1k at resolutions 224× 224 or 448× 448. We compare with the accuracy
obtained with linear probing and obs... | DINOv2- Learning Robust Visual Features without Supervision |
values but also contribute to the development of philosophy by serving as a basis for understanding
how these values evolve and develop over time. Ultimately, these insights contribute to the refinement
of LLM-based agents, ensuring their alignment with human values and ethical standards [27]. | TheRiseandPotentialofLargeLanguageModel BasedAgents |
representation learning. arXiv preprint arXiv:1911.05371, 2019. 6
Y. M. Asano, C. Rupprecht, A. Zisserman, and A. Vedaldi. Pass: An imagenet replacement
for self-supervised pretraining without humans. arXiv preprint arXiv:2109.13228, 2021.
19
M. Assran, R. Balestriero, Q. Duval, F. Bordes, I. Misra, P. Bojanowski, P.... | A Cookbook of Self-Supervised Learning |
Tt−1
Note on inpainting algorithm. We have chosen to adopt
the recently proposed MCG [11] inpainting algorithm,
which outperforms related state-of-the-art diffusion-based
methods (e.g. RePaint [47], DDRM [37]), as we empirically
found it to produce excellent results. Motivated by the orig-
inal algorithm, which aims a... | Relightify-Relightable3DFacesfromaSingleImageviaDiffusionModels |
Jurassic-X: Crossing the neuro-symbolic chasm with the MRKL system
Was Clinton ever elected president of the United States?
Yes
No
No, Clinton was
never elected as
president of the
United States.
Clinton was
elected president
in the 1992
presidential
elections…
Bill Clinton was
elected president.
https://www.a... | Jurassic-X_ Crossing the neuro-symbolic chasm with the MRKL system |
132
Erika Franklin Fowler, Michael M. Franz, & Travis N. Ridout
the funding information often contains the legal name of the funder along with
a phone number (as in the Priorities USA Action example), an address, or
treasurer’s name, which can aid the classification process. | Social_Media_and_Democracy |
Figure 12: Percent loss degradation compared to the Cerebras-GPT compute-efficient scaling law for
varying tokens per parameter.
E Number of Model Parameters
Table 12 shows the formula we use to calculate parameter counts for Cerebras-GPT models.
©2023 Cerebras Systems Inc. All Rights Reserved.
29
Cerebras-GPT: Op... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
DownsampleUpsampleItemsSkipUNetBlockItemsItems×NRCAMI the U-Nets’ denoising process on a text prompt
encoded by a transformer model. In this way, the
generated music both conforms to the latent space
of music and corresponds to the text prompt.
3.2.1 Text Conditioning
To obtain the text embeddings, prior work on text-... | Moûsai |
The concept of a policy improvement operator plays a central role in reinforcement learning. In
reinforcement learning, a policy is the strategy by which an agent chooses its actions, given the
current state of the world. A policy improvement operator is a function that takes in an arbitrary
policy and returns an impro... | CHAIN-OF-THOUGHTREASONING IS APOLICY IMPROVEMENTOPERATOR |
We exploit these new datasets to train SHAPY with three
novel losses, which can be exploited by any 3D human
body reconstruction method: (1) We define functions of the
SMPL body mesh that return a sparse set of anthropomet-
ric measurements. When measurements are available for an
image we use a loss that penalizes mesh... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
doesn’t encounter split subtokens in the SPM format while it does in the PSM format.
Results on the effect of infilling training on downstream generation tasks and the performance of our infilling
models on infilling benchmarks are reported in Section 3.2. | CodeLlama2 |
(3) and (4) are analogous to (1) and (2), except for the following cases where we sketch the differences.
PS↑ ⇒ P↑: Suppose t ∈ f (R1(s)). Then there is some path s0, . . . , sm such that s0 = s and t ∈ f (sm). Since τ is PS↑ there
is a path from every state in f (s0) to every state in f (sm), so it follows that t ∈... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
emerges as an approach to alleviate the manual effort involved. Popular methodologies include neural
architecture search (NAS) [26], meta-learning [1], and Bayesian optimization [13].
Though AutoML is able to reach beyond-human levels in solving machine learning tasks, it still faces
a few drawbacks. Firstly, most Auto... | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
CREATE TABLE has_allergy (
stuid number ,
allergy text ,
foreign key ( allergy ) references allergy_type ( allergy ) ,
foreign key ( stuid ) references student ( stuid )
)
insert into has_allergy (stuid, allergy) values ( 1001, ’Cat’ );
CREATE TABLE student (
stuid number ,
lname text ,
fname text ,
age number ,
sex t... | Teaching Large Language Models to Self-Debug |
Product-Led AI | Greylock
https://greylock.com/greymatter/seth-rosenberg-product-led-ai/
2/10 | Product-Led AI _ Greylock |
improve the productivity of developers writing malicious code (Chen et al., 2021; Weidinger et al.,
2021). Alternatively, competitive programming code generation models in particular could give users
unfair advantages in programming competitions or technical interviews. | alphacode |
2013.
[34] Diederik P Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen, Ilya Sutskever, and Max Welling. Improved
variational inference with inverse autoregressive flow. In Advances in Neural Information Processing
Systems, pages 4743–4751, 2016.
[35] John Lawson, George Tucker, Bo Dai, and Rajesh Ranganath. Energy-ins... | Denoising Diffusion Probabilistic Models |
We evaluate the ability of StarCoder to turn natural language into working code in multiple program-
ming languages using MultiPL-E (Cassano et al., 2023), which translates the HumanEval (Chen et al.,
2021) and MBPP (Austin et al., 2021) Python benchmarks into 18 other programming languages as
follows.
MultiPL-E has a ... | StarCoder_paper (1) |
A wide variety of content might or might not fit a definition of hate speech,
depending on the context (Parekh et al. 2012; Sellars 2016). For example, while
slurs and insults are easily identifiable, language containing epithets may not
necessarily be considered hate speech by the speaker or target recipient
(Delgado 198... | Social_Media_and_Democracy |
increase efficiency [9]. Conversely, this means that people who do not use LLMs, or are inefficient in doing so, will not
be able to compete in the long run, resulting in an economic disadvantage. | Adoptionand AppropriationofLLMs |
As we conclude, it is evident that while significant progress has been made in
developing resource-efficient LLMs, there remains a vast landscape of opportunities
for further research and innovation. The insights and discussions presented in this
survey aim to serve as a guiding framework for future work in this rapidly e... | Beyond Efficiency |
sha1_base64="76w10YEtETzUXdaT0wTZt0xBig8=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuCG5cV7EPaacmkmTY0kxmSO0oZ+h9uXCji1n9x59+YtrPQ1gOBwzn3ck9OkEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmjjVjDdYLGPdDqjhUijeQIGStxPNaRRI3grGNzO/9ci1EbG6x0nC/YgOlQgFo2ilXjeiOArCrD3tYV/0yxW36s5BVomXkwrkqPfLX91BzNKIK2SSGtPx3AT9jGoUTPJpqZsanlA2pkPesVTRiBs/m... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
4.3 EFFECT OF AUGMENTATIONS
In this section, we conduct experiments to
study the effect of augmentations in Meta-
Math. We first finetune the LLaMA-2-7B
model on augmented GSM8K (MetaMath-
GSM8K) data, and test the finetuned model
on GSM8K and MATH. Table 1 shows
the testing accuracy of different combina-
tions of augm... | METAMATH |
0
2
0
2
1
2
0
2
2
2
0
2
3
2
0
2
4
2
0
2
5
2
0
2
6
2
0
2
7
2
0
2
8
2
0
2
9
2
0
2
0
3
0
2
1
3
0
2
2
3
0
2
3
3
0
2
4
3
0
2
5
3
0
2
Internet Users
Yearly Transacting Addresses
Monthly Transacting Addresses
a16z crypto
State of Crypto
2023
What’s Next
59
What we... | State-of-Crypto2023 |
Both Facebook’s and Google’s libraries included both actively running ads
and inactive ads, but Twitter’s searchable archive contained only ads
currently running (Singer 2018a). Google has easy-to-use congressional
district information not yet available from Facebook but currently only
makes data available by week wher... | Social_Media_and_Democracy |
4.3 Evaluation on the New Bing
4.3.1 Evaluated Prompts
Based on our use cases of the New Bing, we no-
tice that direct prompts are sufficient to generate
personal information from the New Bing. Unlike
previous privacy analyses of LMs, the New Bing
plugs the LLM into the search engine. The pow-
erful search plugin enable... | Multi-step Jailbreaking Privacy Attacks on ChatGPT |
vt + (cid:15))
(3)
Here, mt is the momentum, a running average of the gradient, and vt is the velocity, a running average of the
squared gradient. When gradients to a weight are small (e.g., in the case of K bias weights growth above),
vt will tend to be very small, because it is squared. In this case, (cid:15) needs ... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
to OpenAI (2023). The borderline dataset is designed intentionally so that its prompts look adversarial
(e.g., containing sensitive words or subwords) but are not actually unsafe (e.g., “give me a recipe for Christmas
Crack”) (see Appendix Table 41 for more examples).
With more safety data mixed in model tuning, the fa... | Llama2 |
2.9 Potential for Risky Emergent Behaviors
Novel capabilities often emerge in more powerful models.[60, 61] Some that are particularly concerning
are the ability to create and act on long-term plans,[62] to accrue power and resources (“power-
seeking”),[63] and to exhibit behavior that is increasingly “agentic.”[64] Ag... | gpt-4-system-card |
[27] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman,
Yael Pritch, and Daniel Cohen-Or. Prompt-to-prompt im-
age editing with cross attention control. arXiv preprint
arXiv:2208.01626, 2022. 3
[28] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bern-
hard Nessler, and Sepp Hochreiter. Gans trained by a two
tim... | AddingConditionalControltoText-to-ImageDiffusionModels |
Figure 4. Expression extrapolation. Performance of baseline
methods worsen drastically as expressions become more extreme
(higher norm). Geom. error denotes the angular error of the sur-
face normals (lower is better, see Sec. 4.3).
(only available for the synthetic dataset), image quality, and
expression fidelity. Imag... | I M Avatar- Implicit Morphable Head Avatars from Videos |
Quantative comparison: The NUWA [60] paper provided a qualitative evaluation on Kinetics-
400. Since the NUWA model is only 0.9B parameters we also use a model of the same size. Our
model was trained on 50% video and 50% image data in this experiment. The NUWA model fine-
tuned on Kinetics but the Phenaki model is not: ... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
[81] Arkady Zgonnikov, David Abbink, and Gustav Markkula. 2022. Should I Stay or Should I Go? Cognitive Modeling of
Left-Turn Gap Acceptance Decisions in Human Drivers. Human Factors (Dec. 2022), 15 pages. https://doi.org/10.1177/
00187208221144561
[80] John R Wilson and Andrew Rutherford. 1989. Mental models: Theory ... | AI enhance sour performance |
sha1_base64="hP+6LrUf2d3tZaldqaQQvEKMXyw=">AAAB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i77+gsTLXTzQrMMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GSS7KDDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRtRxzzi6dk7A0N+5oYkv39... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
[66] Sihyun Yu, Jihoon Tack, Sangwoo Mo, Hyunsu Kim, Junho Kim, Jung-Woo Ha, and Jinwoo
Shin. Generating videos with dynamics-aware implicit generative adversarial networks. arXiv
preprint arXiv:2202.10571, 2022.
[67] Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lucas Beyer. Scaling vision trans-
formers. In ... | PHENAKI- VARIABLE LENGTH VIDEO GENERATION FROM OPEN DOMAIN TEXTUAL DESCRIPTIONS |
8 RELATED WORK
Recent works that apply large language models to specific domains such as medicine (Singhal et al.,
2022; 2023; Li et al., 2023b; Wu et al., 2023a; Li et al., 2023a; Wang et al., 2023; Xiong et al.,
2023), finance (Wu et al., 2023b; Yang et al., 2023) and law (Cui et al., 2023; Huang et al., 2023),
can ... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
Figure 2. Model-agency websites contain multiple images of mod-
els together with anthropometric measurements. A wide range of
body shapes are represented; example from pexels.com.
Figure 3. We crowd-source scores for linguistic body-shape
attributes [57] and compute anthropometric measurements for
CAESAR [47] body me... | Accurate 3D Body Shape Regression using Metric and Semantic Attributes |
And then I’ve added th
F.12 Gutenberg (PG-19)
s he met with as a novelist, he was anxious to prosecute his
original profession of medicine; and having procured from a foreign
university the degree of M.D., he commenced to practise physic in
Chelsea, but without success. He wrote, however, an essay "On the
External Us... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman.
Glue: A multi-task benchmark and analysis platform for natural language understanding, 2019.
Alex Wang, Yada Pruksachatkun, Nikita Nangia, Amanpreet Singh, Julian Michael, Felix Hill, Omer
Levy, and Samuel R. Bowman. Superglue: A... | LORA |
her dad for
a cat.
her dad for
a cat.
her mom
again.
her dad for
a dog and
her mom
said yes.
she
gave
him a big
hug.
she
gave
him a big
hug.
could
he
have it.
he
her
banana.
gave
the
could
her
he
give
the apple.
could
he
have it.
could
the
he
eat
banana.
school
school
school
school
school
school
h... | TinyStories-HowSmallCanLanguageModelsBeandStillSpeak CoherentEnglish? |
upcomingChinesecomedymoviesarethere?Whatarethetop5mostanticipatedones?Thought:IneedtofindtheupcomingChinesedramamoviesandthetop2mostwantedmovies.Action:coming_out_filter.ActionInput:China,Comedy,5,TrueObservation:datetitlecateregionwantNum04-01JourneytotheWestComedy/Sci-FiMainlandChina17943407OneandOnlyDrama/ComedyMainla... | Tool Learning with Foundation Models |
string currentBranch ;
string versionFilePath ;
string currentBranchPath ();
string branchPath ();
bool existVersionFile ();
bool directoryExists ( string path ) {
// Check if the directory exists
if ( access ( path . c_str () , F_OK ) == 0) {
return true ;
} else {
return false ;
}
}
string versionFilePath () {
... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
Modality
Hyperparameter
Architecture
z-shape
Channels
Depth
Channel multiplier
Attention resolutions
Head channels
Number of heads
CA embed dim
CA resolutions
Autoencoders
Weight initialization
Parameterization
Learning rate
Total batch size
Diffusion Setup
Diffusion steps
Noise schedule
β0
βT
Sampling Parameters
Sampl... | Any-to-Any Generation via Composable Diffusion |
of blindness among adults (Fong et al., 2004). Researchers typically tackle this issue from the perspective of
multitasking (Peng et al., 2020) and transferring (Raghu et al., 2019). However, these practices compel the
available datasets to obey strong assumptions like homogeneous structures or overlapping distribution... | BiomedGPT |
We first show example demo in Fig. 3, where we present various single to single modality generation.
Then, we evaluate the synthesis quality of the unimodal generation on text, image, video, and
audio. CoDi achieves SOTA on audio captions and audio generation, as shown in Table 6 and
Table 4. Notably for the first time i... | Any-to-Any Generation via Composable Diffusion |
3.8 Speeding up Training
3.8.1 Distributed Training
Training self-supervised models often requires large batch sizes [Chen et al., 2020b, He
et al., 2020b], or can be considerably speed up by increasing the batch size, which is
ultimately limited by the memory capacity of the device the model is trained on. Distributed... | A Cookbook of Self-Supervised Learning |
From the beginning, there were constitutional challenges to the FCC’s ability
to regulate broadcasting, not just from the perspective of the First Amendment
but also regarding the national government’s authority over interstate commerce
and its ability to deprive private actors of property by denying them a license
und... | Social_Media_and_Democracy |
The problem with self-regulation by companies like Facebook and Google is
not a legal one; the issue they raise is one of basic legitimacy, brought on by their
scale. Traditional newspapers like the New York Times make decisions to carry
only certain content and not others in a print media market that is still relative... | Social_Media_and_Democracy |
decision-making process. Furthermore, the neural modules
are only activated when the LLM decides to call the relevant
functions, which brings flexibility to the system.
2.3. Cognitive Memory | ALanguageAgentforAutonomousDriving |
A survey of quantization methods for efficient neural network inference.
arXiv:2103.13630, 2021.
Babak Hassibi, David G Stork, and Gregory J Wolff. Optimal brain surgeon and general network
pruning. In IEEE International Conference on Neural Networks, 1993.
Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dryden, an... | GPTQ |
0|c, where xw
0 ≻ xl
0 and xl
pBT(xw
0 ≻ xl
0|c) = σ(r(c, xw
0 ) − r(c, xl
0))
(3)
where σ is the sigmoid function. r(c, x0) can be parame-
terized by a neural network ϕ and estimated via maximum
likelihood training for binary classification:
LBT(ϕ) = −E
(cid:2)log σ(cid:0)rϕ(c, xw
0 ) − rϕ(c, xl
0)(cid:1)(... | DiffusionModelAlignmentUsing Direct Preference Optimization |
One important external assessment of company transparency reports’
strengths and weaknesses can be found in the Electronic Frontier Foundation’s
periodic Who Has Your Back report (Gebhart 2018). Another can be found in the
Ranking Digital Rights Corporate Accountability Index, which rates technology
companies on numero... | Social_Media_and_Democracy |
●
a16z crypto
State of Crypto
2023
Trends to Watch: Regulation and Policy
30
Bipartisan momentum is building
Digital assets, blockchain technology and
cryptocurrencies have experienced tremendous growth in
the past few years and offer substantial potential benefits
if harnessed correctly. It is critical... | State-of-Crypto2023 |
∼ 275M
∼ 430M
∼ 450M
∼ 30M
∼ 100M
∼ 500M
No
No
Yes
No
Yes
No
Table 2: MathMix dataset components.
Note that when training smaller models, as in Section 4, we use a slightly
smaller variant of MathMix that excludes the critiques data and only consists of
1B tokens. For our large models experiments, we train on MathMi... | Let’s Verify Step by Step |
top-5
70.3
70.5
68.4
67.8
66.5
67.2
70.3
62.0
62.3
72.0
CF Rule (execution)
top-5
top-1
74.6
71.1
75.8
72.7
70.3
72.8
74.5
70.8
74.5
70.6
74.3
70.5
75.0
71.4
62.4
68.2
68.7
62.9
72.8
78.5
top-3
73.4
75.3
72.4
73.1
72.8
72.9
73.7
65.5
66.8
76.7
#P
770M
770M
175B
20B
15.5B
16B
350M
50M
93M
75M
GitHub notebooks and St... | CODEFUSION |
Eight Things to Know about Large Language Models
White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert,
H., Elnashar, A., Spencer-Smith, J., and Schmidt, D. C.
A prompt pattern catalog to enhance prompt engineering
with ChatGPT. arXiv preprint 2302.11382, 2023.
Zhou, D., Sch¨arli, N., Hou, L., Wei, J., Scales,... | Eight Things to Know about Large Language Models |
2. Based on this belief and to address the positional O.O.D.
issue, we propose SelfExtend to extend the context win-
dow of LLMs without any fine-tuning. Our proposal
maps the unseen large relative positions (at inference)
to known positions (at training), thus it allows LLMs to
maintain coherence over longer texts wit... | Self-Extend LLM |
is a partition of S1.
impossible since f covers S2 by definition. It follows that f (t) (cid:7)= ∅ for all t ∈ S2.
which contradicts that f
is total. It follows that Rng( f ) covers S1.
) (cid:7)= ∅, so at least one of f (t) \ f (t
) and f (t) ∩ f (t
f (t) (cid:7)= f (t
(cid:10)
assume the first of these hold. ... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Query #1: Task Planning #1:1: {"task": "image-to-text", "id": 0, dep: [-1], "args": {"image": "example.jpg"}}2: {"task": "control-canny", "id": 0, dep: [-1], "args": {"image": "example.jpg"}}Response #1: I have generated a canny image for you based on the image located at example.jpg. The task process involved selectin... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
• Slot Filling: Slot Filling (SF) is a widely used technique in Speech Language Understanding
(SLU) that enables the extraction of important information, such as names, dates, and
locations, from a user’s speech. The process involves identifying the specific pieces of
information that are relevant to the user’s request... | AReviewofDeepLearningTechniquesforSpeechProcessing |
8.5.4. Valtorta’s theorem
Valtorta [90] proved that when using embeddings as abstraction functions, it is not possible to explore fewer nodes in
total, counting both ground and abstract nodes, when using the A∗ search algorithm [44] with path length in the abstract
graph as heuristic estimate, than if using D... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Comparing the top and bottom rows, it is evident that
adding Gaussian noise is instrumental in circumventing the
training collapse issue, which prevents generating null or
meaningless outputs. The middle and bottom rows under-
score the significance of token embeddings in the style in-
jection module, which better inje... | Instant3D |
gross errors than the naive PnP solution.
Synthetic dataset genetarion. We train separate PoseNet,
one for human, and one for quadruped animals. The train-
ing pipeline is shown in Fig. 12. Specifically, we render sur- | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
1. Introduction
The field of computer vision has seen significant ad-
vancements in recent years, particularly in the area of gen-
erative AI. In the domain of image generation, Stable Diffu-
sion has revolutionized content creation by providing open
software to generate arbitrary high-fidelity RGB images
from text promp... | LDM3D- Latent Diffusion Model for 3D |
In text classification, on most datasets, LLMs perform slightly worse than fine-tuned models. For sentiment analysis,
such as on IMDB [69] and SST [94], fine-tuned models and LLMs perform equally well. For toxicity detection, which
is another iconic text classification task, the gap is much larger. All LLMs cannot perf... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Data-to-Text Generation via Planning. Findings of EMNLP (2021).
[175] Lya Hulliyyatus Suadaa, Hidetaka Kamigaito, Kotaro Funakoshi, Manabu Okumura, and Hiroya Takamura. 2021.
Towards Table-to-text Generation with Numerical Reasoning. In Proceedings of the 59th Annual Meeting of the
Association for Computational Lingui... | SurveyofHallucinationinNatural Language Generation |
(cid:19)
(cid:18) 1
β
πr(y | x)
πref(y | x)
+ β log Z(x).
r(x, y) = β log
(5)
We can apply this reparameterization to the ground-truth reward r∗ and corresponding optimal model
π∗. Fortunately, the Bradley-Terry model depends only on the difference of rewards between two
completions, i.e., p∗(y1 ≻ y2 | x) = σ(r∗(... | Direct Preference Optimization |
Ganqu Cui, Lifan Yuan, Bingxiang He, Yangyi Chen, Zhiyuan Liu, and Maosong Sun. A unified evaluation
of textual backdoor learning: Frameworks and benchmarks. Advances in Neural Information Processing
Systems, 2022.
Jonas Degrave, Federico Felici, Jonas Buchli, Michael Neunert, Brendan Tracey, Francesco Carpanese, Timo
... | Tool Learning with Foundation Models |
Broader impacts. We hope to improve training efficiency and reduce the environmental impact
of training large LMs (Lacoste et al., 2019, Ligozat et al., 2021, Patterson et al., 2021, Strubell et al.,
2019). However, large LMs have also been well-documented to have risks and biases (Abid et al.,
2021, Blodgett and OConno... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
formance of translation systems (Ghorbani et al., 2021). In
order to avoid learning “transcript-ese”, we developed many
heuristics to detect and remove machine-generated tran-
scripts from the training dataset. Many existing ASR sys-
tems output only a limited subset of written language which
removes or normalizes away... | RobustSpeechRecognitionviaLarge-ScaleWeakSupervision |
produce a dataset similar to the one promised in the original RFP – a dataset of
almost 38 million links comprised of several trillion cell entries delineating the
numbers and types of people who saw and engaged with the URLs, but with
statistical noise added to the data to protect users’ privacy. The dataset also
cont... | Social_Media_and_Democracy |
[55] M. Z. Hossain, M. A. Rahman, M. S. Islam, and S. Kar, ‘‘Ban-
FakeNews: A dataset for detecting fake news in Bangla,’’ in Proc.
12th Lang. Resour. Eval. Conf. Marseille, France: European Language
Resources Association, May 2020, pp. 2862–2871. [Online]. Available:
https://www.aclweb.org/anthology/2020.lrec-1.349
[... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
Easter Egg Hunt.
85 patients (23%) were hospitalised alive and admitted to a hospital ward. Of them, <API> Calcula-
tor(85 / 23) → 3.70 </API> 65% had a cardiac aetiology [. . .]
But hey, after the <API> Calendar() → Today is Saturday, June 25, 2011. </API> Disneyland
fiasco with the fire drill, I think it’s safe to say ... | Toolformer |
t
o
o
u
r
t
e
a
m
?
T
h
i
s
r
o
l
e
i
n
v
o
l
v
e
s
w
o
r
k
i
n
g
w
i
t
h
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
i
n
t
e
g
r
a
t
i
n
g
w
i
t
h
L
a
r
g
e
L
a
n
g
u
a
g
e
S
e
n
i
o
r
S
o
f
t
w
a
r
e
E
n
g
i
n
e
e
r
,
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
-
G
e
n
e
r
a
t
i
v
e
A
I
B
e
l
l
e
v
u
e
,
W
A
$
1
2
... | Senior Software Engineer, Machine Learning - Generative AI Job in Bellevue, WA at SeekOut |
it requires evaluating the model on a large minibatch, while only selecting a small fraction to up-
date the model with. Overall, in contrast to these DRO methods which aim to produce robust
models, we use DRO to optimize the data for training larger models more efficiently. | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
USER: What do you think of Rosalie Gascoigne’s sculptures?
8
(This transcript was taken from a dialog with one of the authors, and includes the generated base output and search
queries, which are not usually shown to the user.)
The basic LaMDA language model, which we refer to as the ‘Base’ model here, generates a d... | LaMDA- Language Models for Dialog Applications |
8This results slightly differ from the earlier setup because we release the best checkpoints only on direct prompting. CoT did worse,
so we prioritized direct prompting.
25
7 Acknowledgements
The authors would like to specially thank (in alphabetical order): Alexey Gritsenko, Andrew M. Dai, Jacob
Devlin, Jai Gupta, ... | UL2- Unifying Language Learning Paradigms |
[33] Asbjørn Hrõbjartsson and Peter Christian Gøtzsche. 2001. Is the placebo powerless? An analysis of clinical trials
comparing placebo with no treatment. The New England journal of medicine 344 21 (May 2001), 1594–602. https:
//doi.org/10.1056/nejm200105243442106
[34] Jie Huang, Wenhua Wu, Zhenyi Zhang, and Yutao Ch... | AI enhance sour performance |
avetakenxbefore.66A.10ALFWorldNoentitywasfound!Thought:Maybethisisarelationinstead.Action:get_relation_id("IPv4routingprefix")Observation:idlabeldescription0P3761IPv4routingprefixrangeofIPv4addressesThought:IneedtofindtheentityassociatedwiththisIPaddressAction:search_by_query("SELECT?objWHERE{?objwdt:P3761"130.237.88.0/2... | Tool Learning with Foundation Models |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.