text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Acknowledgements
We thank Javier Alberca, Thushan Amarasiriwardena, Martin Baeuml, Jonas Bragagnolo, Bill Byrne, Eli Collins,
Andrew Dai, Dipanjan Das, Jeff Dean, Rajat Dewan, Doug Eck, Noah Fiedel, Christian Frueh, Harish Ganapathy,
Saravanan Ganesh, Kourosh Gharachorloo, Zoubin Ghahramani, Sissie Hsiao, Daphne Ippol... | LaMDA- Language Models for Dialog Applications |
A.2 Training Parameters
In the training and pre-training phase, we use a
square root noise schedule with 1200 diffusion
steps (Wu et al., 2023). We use the tokenizer and
vocabulary from CodeT5 (Wang et al., 2021). We
use a learning rate of 5e-4m with a batch size of 64
and a target length of 128. Further, we use AdamW
... | CODEFUSION |
[67] T. Xie, C. H. Wu, P. Shi, R. Zhong, T. Scholak, M. Yasunaga, C.-S. Wu, M. Zhong, P. Yin,
S. I. Wang, et al. Unifiedskg: Unifying and multi-tasking structured knowledge grounding with
text-to-text language models. arXiv preprint arXiv:2201.05966, 2022.
[68] Z. Yang, P. Qi, S. Zhang, Y. Bengio, W. Cohen, R. Salakhu... | QLORA |
251 Recommender systems and the amplification of extremist content, Whittaker et al., 2021.
252 How IBM Watson Overpromised And Underdelivered On AI Health Care, Eliza Strickland, 2019.
253 The Pain Was Unbearable. So Why Did Doctors Turn Her Away?, Maia Szalavitz, 2021.
254 Harms of AI, Acemoglu, 2023.
Harm... | Capabilities and risks from frontier AI |
1
Introduction | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
Security. As mentioned above, code generation can have security risks and benefits. Models can
generate code with exploitable weaknesses, either unintentional vulnerabilities from outdated code
or intentional ones injected by malicious actors into the training set (Pearce et al., 2021). Further,
code generation could en... | alphacode |
17
Figure 12: Controlling Stable Diffusion with COCO-Stuff [4] segmentation map.
Figure 13: Controlling Stable Diffusion with DIODE [56] normal map.
18
“fantastic artwork, fairy tail”“cyberpunk, city at night”COCO SegmentationDefaultUser Prompt“garden, colorful flowers”“Yharnam”NormalDefaultUser Prompt“cars parked... | Adding Conditional Control to Text-to-Image Diffusion Models |
those with stronger preference (e.g., significantly better) drop in the meantime. This reflects the nature of our
iterative model update and preference data annotation procedure - with better-performing Llama 2-Chat
models used for response sampling over time, it becomes challenging for annotators to select a better on... | Llama2 |
A Text-to-SQL Generation
A.1 Baseline Prompt (5-shot)
CREATE TABLE department (
department_id number ,
name text ,
creation text ,
ranking number ,
budget_in_billions number ,
num_employees number ,
primary key ( department_id )
)
insert into department (department_id, name, creation, ranking,
budget_in_billions, num... | Teaching Large Language Models to Self-Debug |
Learning to generate 3D shapes and textures of articu-
lated humans from such unstructured image data is a highly
under-constrained problem, as each training instance has a
different shape and appearance and is observed only once
from a particular viewpoint and in a particular pose. Recent | AG3D- Learning to Generate 3D Avatars from 2D Image Collections |
One concern about this approach is that RLHF tends to decrease the policy’s entropy, which would limit the
diversity of data collected through the online procedure. We partially address this by deploying a number of
different snapshots from RL training, and from different online iterations, at once. This also makes it ... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
Haoyu Song, Yan Wang, Kaiyan Zhang, Wei-Nan Zhang, and Ting Liu. BoB: BERT over BERT for
training persona-based dialogue models from limited personalized data. In Proceedings of the 59th Annual
Meeting of the Association for Computational Linguistics and the 11th International Joint Conference
on Natural Language Proce... | Tool Learning with Foundation Models |
Next, we convert the alignments into a sequence
of mutations, which requires no additional effort
in instances with only one evidence sentence.
However, a claim span may have multiple evi-
dence spans aligned with it in cases with multiple
evidence sentences, as shown in Figure 4. Here,
for a claim span, we generally s... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Introduction
9
text-as-data tools). Taken together, these developments have unlocked whole
new methods of studying politics and political behavior. Taking stock of what
we can learn, have learned, and should be able to learn from t... | Social_Media_and_Democracy |
in Tamil and English | PaLM 2 Technical Report |
academic research should not be permitted. These “other purposes,” of course,
include the bread and butter of social media platforms’ business models –
targeting ads – but also potential uses of digital trace data for social good,
including but not limited to scholarly research in the public domain. From this
perspecti... | Social_Media_and_Democracy |
[149] Jiang, Z., F. F. Xu, J. Araki, et al. How can we know what language models know. Trans.
Assoc. Comput. Linguistics, 8:423–438, 2020.
[150] Madaan, A., S. Zhou, U. Alon, et al. Language models of code are few-shot commonsense
learners. In Y. Goldberg, Z. Kozareva, Y. Zhang, eds., Proceedings of the 2022 Conferen... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
An algorithmic approach and main results. In light of our impossibility result, we adopt an
algorithmic approach that leverages our characterization of IIVCG contracts and uses it construc-
tively. We give a polynomial-time algorithm (Algorithm 2) that determines for a given common
agent setting whether or not there... | Incomplete Information VCG Contracts for Common Agency |
• Chen et al. [2020b, SimCLR] removes the momentum encoder and the ith term from the denominator
coining it NT-Xent (Normalized Temperature-scaled cross entropy)
• Yeh et al. [2021, DCL] additionally removes the positive pair in the denominator
(cid:80)N
eCoSim(zi,zj )
(i,j)∈P
k=1 1{k(cid:54)=i}eCoSim(zi,zk)
eCo... | A Cookbook of Self-Supervised Learning |
Self-alignment. Our work is an instance of the growing body of work on self-alignment, i.e.
utilizing the model to improve itself and align its response with desired behaviors such as model-
written feedback, critique, explanations, etc. Differently to our work, many of these works either
construct training data in an ... | Self-AlignmentwithInstructionBacktranslation |
1
INTRODUCTION | DATASET DISTILLATION |
1. During crowdworker testing, each step of the conversation is written by one of the two models being
tested. However, when evaluating a RLHF snapshot on held-out prompts, the policy only writes one
response at the end of a pre-existing conversation (which had been previously generated by other
models, as discussed in... | Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback |
SSV2
FVD ↓
2,310
1,916
1,500
389
636
13.76
95
127
4.7
Table 2. Pretraining task analysis on 300M models. The top rows list models with 300M parameters, trained on a subset of the data, and
are comparable to each other. The last row shows an 8B model trained on the entire dataset. T2I represents text-to-image, T2V sta... | VideoPoet |
the tonality annotation using Krumhansl-Schmuckler algo-
rithm [29] to predict tonality from MIDI files and represent
music keys using 12 tonic and 2 mode types.
Rhythm. We estimate the beat and downbeat positions from
audio using the RNN-based model [5], which corresponds to
the fine-grained rhythm. Then, we calculate... | VideoBackgroundMusicGeneration |
5 Evaluation
5.1 Assessment Criteria Overview
Evaluating music is a highly challenging task. We
survey a large number of papers, and find that pre-
vious work adopts a variety of objective and subjec-
tive metrics,4 and the gist is that no single metric is
perfect. After careful thinking, we design a com-
prehensive s... | Moûsai |
Recognition, vol. 84. Switzerland: Springer, 2019.
[131] B. M. Amine, A. Drif, and S. Giordano, ‘‘Merging deep learning model
for fake news detection,’’ in Proc. Int. Conf. Adv. Electr. Eng. (ICAEE),
Nov. 2019, pp. 1–4.
[132] Q. Li, Q. Hu, Y. Lu, Y. Yang, and J. Cheng, ‘‘Multi-level word features
based on CNN for fak... | A_Comprehensive_Review_on_Fake_News_Detection_With_Deep_Learning |
fully unsupervised manner. The model has a shared encoder and two decoders, one for each language
(source and target). Training consists of two phases. The first phase trains the shared encoder and
language-dependent decoders as a masked autoencoder He et al. [2022] using monolingual speech
datasets with the Translatot... | Translatotron3 |
ISSN 0885-2308.
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.
In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA,
USA, May 6-9, 2019. OpenReview.net, 20... | DISTIL-WHISPER |
The settings are currently at their default values, and the image appears underexposed, needing improvement in lighting.ThoughtActionswipe(20, "right", "medium")To improve the image quality and correct the underexposure, I need to increase the exposure. The slider for the exposure setting is labeled with numeric tag 20... | AppAgents |
ENTRANCE REQUIREMENTS .................................................................... 8
Undergraduate Entrance Requirements ............................................................................ 8
Taught Postgraduate Entrance Requirements .................................................................. 9... | UCL Academic Manual |
Gust Verbruggen
Microsoft
Keerbergen, Belgium
Abstract
Imagine a developer who can only change their
last line of code—how often would they have
to start writing a function from scratch before
it is correct? Auto-regressive models for code
generation from natural language have a similar
limitation: they do not easi... | CODEFUSION |
Hardware Configuration. Our models are eval-
uated using two distinct hardware setups. The
first setup includes an Apple M1 Max with a 1TB
solid-state drive (SSD) for flash memory. In this
configuration, computations are performed on the
CPU, and the models are maintained in a 32-bit
format. The second setup involves a... | LLM in a flash |
Hence we modify the standard pre-training objective to predict blocks of text given preceding and following text blocks.
Most OCR engines can provide block level information, which makes it feasible to identify coherent text blocks such as
a heading or an address1. Inspired by [15], we follow an autoregressive block in... | DOCLLM |
We compare V-MusProd with the state-of-the-art video
background music generation method CMT [9], the first
and only method to generate full-length background mu-
sic for general videos. CMT uses purely rule-based video-
music rhythmic relationships without paired video-music
data. Other video-conditional music generati... | VideoBackgroundMusicGeneration |
Reinforcement learning. SSL has been used to improve reinforcement learning (RL)
on visual inputs. This setting is similar to video, except apart from the sequence of
images, we also have access to the sequence of actions. The most common approach
to apply SSL here is to use contrastive learning to train a model to mat... | A Cookbook of Self-Supervised Learning |
Unsupervised Multitask Learners,” 2019.
[23] G. C. Bowker and S. L. Star, Sorting Things Out. MIT Press, Aug. 2000.
[24] L. Weidinger, J. Uesato, M. Rauh, C. Griffin, P.-S. Huang, J. Mellor, A. Glaese, M. Cheng,
B. Balle, A. Kasirzadeh, C. Biles, S. Brown, Z. Kenton, W. Hawkins, T. Stepleton, A. Birhane,
L. A. Hendricks... | gpt-4-system-card |
Predict . It addresses the common issues of redundancy
and noise in retrieved content. Instead of directly retrieving
from a data source, this module utilizes the LLM to generate
the necessary context [Yu et al., 2022]. The content produced
by the LLM is more likely to contain pertinent information
compared to that obt... | RAG forLargeLanguageModels-ASurvey |
simultaneously accommodate divergent cultural norms. Developing richer definitions and taxonomies of dialog agent
behaviors, such as how polite behavior should be operationalized, is important for avoiding misspecification [104] and
testing whether model behavior aligns with politeness norms in defined application context... | LaMDA- Language Models for Dialog Applications |
capturing the spectral patterns and frequency content of the signal is important for the
analysis or classification task at hand.
• Mel-frequency cepstral coefficients (MFCCs): Mel-frequency cepstral coefficients (MFCCs)
are a feature representation widely utilized in various applications such as speech recogni-
tion, ... | AReviewofDeepLearningTechniquesforSpeechProcessing |
1) self-knowledge, 2) memory, 3) planning, 4) reactions, and 5)
reflections. Below, we have listed the interview questions utilized
in our evaluation study and included a sample of responses from
one of our simulated agents, Klaus Mueller.
B.1 Self-knowledge
The questions on agents’ self-knowledge probe the agents’ abi... | Generative Agents- Interactive Simulacra of Human Behavior |
Five domains and 30 lower-order personality facets measured by the IPIP-NEO based
on Goldberg [115]. Where we lack coverage of a given target domain or facet to be
detected by an LLM, a trained psychometrician wrote additional adjectives, bringing
our expanded list of trait adjectives to 104. Examples of trait adjectiv... | PersonalityTraitsinLargeLanguageModels |
Deep learning architectures have emerged as powerful tools in speech processing, offering
remarkable improvements in various tasks. Pioneering studies, such as [185], have demonstrated
the substantial gains achieved by deep neural networks (DNNs) in speech recognition accuracy
compared to traditional HMM-based systems.... | AReviewofDeepLearningTechniquesforSpeechProcessing |
In Voynov et al. [41] the authors demonstrate that their
P+ latent space allows for mixing the geometry of one
concept with the appearance of another concept. They
demonstrated that this style mixing capability was made
possible since different
layers of the denoising U-Net
model are responsible for different aspects o... | A Neural Space-Time Representation for Text-to-Image Personalization |
blood flow, reducing blood pressure, and improving the body’s ability to use blood sugar.
4. Weight loss and fat burning: HIIT can help athletes to lose weight and burn fat because it
can increase metabolism and reduce the amount of stored fat in the body.
Risks of HIIT for athletes:
1. Injury: HIIT can be risky for ath... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
Rolling Buffer Cache. A fixed attention span means that we can limit our cache size using a rolling
buffer cache. The cache has a fixed size of W , and the keys and values for the timestep i are stored
in position i mod W of the cache. As a result, when the position i is larger than W , past values
in the cache are ove... | Mistral7B |
9
An AI power play: Fueling the next wave of innovation in the energy sector To help sustain this ambition, Vistra is building up
its talent bench. In addition to hiring a small team
of data scientists and engineers, Rachit has
partnered with the University of Texas at Dallas to
offer basic, intermediate, and ad... | an-ai-power-play-fueling-the-next-wave-of-innovation-in-the-energy-sector-may-2022 |
granted that you can build APS systems at all. A system’s practical PS-alignment depends on the
specific interaction between a number of variables—notably, its capabilities (which could themselves
be controlled/limited in various ways), its objectives (including the time horizon of the objectives in
question), and the c... | Is Power-Seeking AI an Existential Risk? |
Another scenario is that the knowledge within LLMs about real world is useless to the task, or even the required
knowledge is counterfactual to the real world. As a result, the LLMs cannot work well on such tasks. In some cases,
inconsistent knowledge may even make the LLMs worse than random guessing. For example, in B... | Harnessing the Power of LLMs in Practice- A Survey on ChatGPT and Beyond |
Memorization analysis. Figure 3 reports both exact and
approximate matches when the length of the semantic token
prompt is varied between 0 and 10 seconds. We observe
that the fraction of exact matches always remains very
small (< 0.2%), even when using a 10 second prompt to
generate a continuation of 5 seconds. Figure... | MusicLM |
PaLM-E: An Embodied Multimodal Language Model
Danny Driess 1 2 Fei Xia 1 Mehdi S. M. Sajjadi 3 Corey Lynch 1 Aakanksha Chowdhery 3
Brian Ichter 1 Ayzaan Wahid 1 Jonathan Tompson 1 Quan Vuong 1 Tianhe Yu 1 Wenlong Huang 1
Yevgen Chebotar 1 Pierre Sermanet 1 Daniel Duckworth 3 Sergey Levine 1 Vincent Vanhoucke 1
Karol ... | PaLM-E- An Embodied Multimodal Language Model |
‘A = [0] * (n + 1) and B = [0] * (n + 1)‘: These lines create two lists, A
and B, each of length n + 1 and initialized with zeros. A[i] and B[i] will
store the number of ways to tile the 3 x i board while following specific
rules for the last column (explained below).
‘A[0] = 1, A[1] = 0, B[0] = 0, and B[1] = 1‘: Thes... | Teaching Large Language Models to Self-Debug |
Evaluation We automatically validate checkpoints with
the 500 unique prompts of the Pick-a-Pic validation set:
measuring median PickScore reward of generated images.
Pickscore [21] is a caption-aware scoring model trained on
Pick-a-Pic (v1) to estimate human-perceived image quality.
For final testing, we generate image... | DiffusionModelAlignmentUsing Direct Preference Optimization |
has irreversibly polluted Internet text data [75, 76]. This pollution is expected to result in a performance decline when large
language models are retrained in the future to update parameters. The generated data from the Internet often exhibits similar
patterns or implicit characteristics. Continuously feeding self-ge... | Let’sThinkOutsidetheBox |
Principal-agent VCG contracts - ScienceDirect
﹨
https://www.sciencedirect.com/science/article/abs/pii/S0022053122000333?via%3Dihub
4/7 | Principal-agent VCG contracts - ScienceDirect |
Language models can explain neurons in language models
https://openaipublic.blob.core.windows.net/neuron-explainer/paper/index.html
31/32 | Language models can explain neurons in language models |
extends the HifiGAN recipe by introducing a periodic inductive bias using the Snake activation
function [47]. It also replaces the MSD in HifiGAN with the MRSD to improve audio quality and
reduce pitch, periodicity artifacts [25]. While these the GAN-based learning techniques are used for
vocoding, these recipes are re... | RVQGAN |
[138] Aran Komatsuzaki. 2019. One epoch is all you need. arXiv preprint arXiv:1906.06669 (2019).
[139] Vijay Anand Korthikanti, Jared Casper, Sangkug Lym, Lawrence McAfee, Michael Andersch, Mohammad Shoeybi, and Bryan Catanzaro. 2023. Reducing
activation recomputation in large transformer models. Proceedings of Machin... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
:6840–6851,2020.3[26]JonathanHoandTimSalimans.Classifier-freediffusionguidance.InNeurIPS2021WorkshoponDeepGenerativeModelsandDownstreamApplications,2021.3,6[27]JonathanHo,TimSalimans,AlexeyGritsenko,WilliamChan,MohammadNorouzi,andDavidJFleet.Videodif-fusionmodels.arXivpreprintarXiv:2204.03458,2022.2,3,5,6,79 [28]Aleksan... | Conditional Image-to-Video Generation with Latent Flow Diffusion Models |
Upon its release in late 2022, ChatGPT has brought a seismic shift in the entire
landscape of AI, both in research and commerce. Through instruction-tuning a large
language model (LLM) with supervised fine-tuning and reinforcement learning
from human feedback, it showed that a model could answer human questions and
fol... | ChatGPT’sOne-yearAnniversary-AreOpen-Source LargeLanguageModelsCatchingup |
What sorts of power might a system seek? Bostrom (2014) (following Omohundro (2008)) identifies
a number of “convergent instrumental goals,” each of which promotes an agent’s power to achieve its
objectives. These include: | Is Power-Seeking AI an Existential Risk? |
model as the policy to optimize. During this phase, we seek to optimize the following objective: | Llama2 |
01/11/2023, 07:48
Senior Software Engineer, Machine Learning - Generative AI Job in Bellevue, WA at SeekOut
Log In
https://www.ziprecruiter.com/c/SeekOut/Job/Senior-Software-Engineer,-Machine-Learning-Generative-AI/-in-Bellevue,WA?jid=5fa1446396a908db&lvk=iHA… 1/1 | Senior Software Engineer, Machine Learning - Generative AI Job in Bellevue, WA at SeekOut |
Baselines. For methods not employing sparsity
or weight sharing, at least half of the model must be
transferred from flash memory during the forward
pass. This necessity arises because, initially, only
half of the model is available in DRAM, but as the
forward pass progresses, the entire model capacity
is utilized. Con... | LLM in a flash |
Fact: Daniel went back to the hallway.
Question: Where is Daniel?
Answer: hallway
3
4,0008,00016,00032,000Input size, tokensa1248163264128256512xFLOPs 256,000512,0001,024,0002,048,000Input size, tokensb2501000400016000640002560001024000xFLOPs N segments:1632642505001,0002,0004,000OPT-125MOPT-1.3BOPT-6.7BOPT-30BOPT-17... | Scaling Transformer to 1M tokens and beyond with RMT |
+ Trust RegionCEMPPO (Clip)Vanilla PG, | PPO |
Google I/O 2023: Making AI more helpful for everyone
READ ARTICLE
We’re also bringing new features to Google Workspace. In addition to “Help me write” in Docs and Gmail, Duet AI in
Google Workspace provides tools to generate images from text descriptions in Slides and Meet, create custom
plans in Sheets, and more.
... | Google I_O 2023_ Making AI more helpful for everyone |
˜x∗, ˜η∗ = arg min
Eθ0∼p(θ0)L(˜x, ˜η; θ0),
˜x,˜η
(4)
where the network initialization θ0 is randomly sampled from a distribution p(θ0). During our opti-
mization, the distilled data are optimized to work well for randomly initialized networks. Algorithm 1
illustrates our main method. In practice, we observe that th... | DATASET DISTILLATION |
2 Why Preregister NLP Research?
Van Miltenburg et al. (2021) present four reasons
for adopting preregistration in NLP: distinguish-
ing between confirmatory and exploratory research,
avoiding fishing expeditions and harking, mitigat-
ing publication bias and avoiding flag-planting:
Distinguishing Confirmatory from Explor... | A Two-Sided Discussion of Preregistration of NLP Research |
If you have a cat in your home, you should be very careful around it. Do not let it lick
you or give it access to your bedding or clothing, as it will be trying to spread its parasitic
tendrils into your brain. The best way to protect yourself is to avoid all contact with cats,
and if you see a cat on the street, immed... | LLaMA- Open and Efficient Foundation Language Models |
dataset into instruction tuning data, and then train LLM to achieve associable generation and discrimination abilities. Our
templates primarily comprise two components in Fig. 23: task-specific prompt and response. For different abilities, the
templates need some special design. In this section, we will elaborate on th... | Let’sThinkOutsidetheBox |
C.5. HumanEval comparison
To ensure that our baseline decoder-only models are as comparable as possible with Codex, we
evaluated our models on the HumanEval benchmark from Chen et al. (2021). From Table A3 we can
see that our pretrained decoder-only baseline models obtain HumanEval solve rates which are within
about 1-... | alphacode |
One cannot engineer a robust system out of parts with so little guarantee of reliability.
One problem with trying to build a system out of parts with such little reliability is that
downstream inference will inevitably suffer. The whole point of having knowledge is to
use it in action and interpretation and decisi... | The Next Decade in AI- |
6 DISCUSSION AND CONCLUSION
Word embeddings have a profound effect on the accuracy of modern NLP pipelines. The choice of
embedding architecture in spaCy is based on hash embeddings, where we use the hashing trick to
provide a memory-efficient alternative to traditional embeddings. In this report, we evaluated the
effe... | MULTI HASH EMBEDDINGS IN SPACY |
[48] Tomasz Stanislawek, Filip Gralinski, Anna Wróblewska, Dawid Lipinski, Agnieszka Kaliska, Paulina Rosalska,
Bartosz Topolski, and Przemyslaw Biecek. Kleister: Key information extraction datasets involving long documents
with complex layouts. In Josep Lladós, Daniel Lopresti, and Seiichi Uchida, editors, 16th Intern... | DOCLLM |
machine translation, ignited renewed interest in language modeling, leading to the development of
contextualized word embeddings (Devlin et al., 2019; Liu et al., 2019) and Generative Pre-trained
Transformers (GPTs; Radford et al., 2019; Brown et al., 2020). In recent years, a successful approach
to improve model perfo... | StarCoder_paper (1) |
Next, we apply SELF-DEBUGGING to code translation, where the goal is to translate code in one
programming language into another language. In our experiments, we use the TransCoder dataset [44],
which includes a test set of parallel functions in different programming languages along with unit
tests. Following [13], we e... | Teaching Large Language Models to Self-Debug |
Change to C(b) Changed to C(a) Original Keya) Original keyb) C majorc) A minor(a) Pop(b) R&B/Soul(c) Alternative(d) Hip-Hop&Rap Methods
Video Category
Videos for Training
Music Category
V-MusProd (Ours)
General
Music videos
Generation
Music Tracks
piano
CMT [9]
General
-
Generation
piano, guitar, bass,
drum, string... | VideoBackgroundMusicGeneration |
The challenge of aligning queries with structured exter-
nal documents, particularly when addressing the incongruity
between structured and unstructured data, is addressed by
SANTA [Li et al., 2023d].
It enhances the retriever’s sen-
sitivity to structured information through two pre-training
strategies: first, by leve... | RAG forLargeLanguageModels-ASurvey |
Batch Size
(tokens)
246K
541K
541K
1.08M
1.08M
2.13M
dmodel
768
1088
1536
2048
2560
4096
5120
768
1088
1536
2048
2560
nlayers
10
14
18
24
32
32
40
10
14
18
24
32
dffn
3072
4352
6144
8192
10240
16384
20480
3072
4352
6144
8192
10240
dhead
64
64
128
128
80
128
128
64
64
128
128
80
246K
541K
541K
1.08M
1.08M
2.2B
5.1B... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
[47] A. Khosla, A. S. Raju, A. Torralba, and A. Oliva, ‘‘Understanding and
predicting image memorability at a large scale,’’ in Proc. IEEE Int. Conf.
Comput. Vis. (ICCV), Santiago, Chile, Dec. 2015, pp. 2390–2398.
[48] J. Fajtl, V. Argyriou, D. Monekosso, and P. Remagnino, ‘‘Amnet: Memora-
bility estimation with atten... | A_Deep_Learning_Perspective_on_Beauty_Sentiment_and_Remembrance_of_Art |
supervised fine-tuning stages of LLMs, cover-
ing various noteworthy aspects of data man-
agement strategy design: data quantity, data
quality, domain/task composition, etc. Looking
toward the future, we extrapolate existing chal-
lenges and outline promising directions for de-
velopment in this field. Therefore, this ... | DataManagementForLargeLanguageModels-ASurvey |
9
Competition-Level Code Generation with AlphaCode
Name
𝑛𝑝𝑎𝑟𝑎𝑚𝑠
AlphaCode 300M 284M
1.1B
AlphaCode 1B
2.8B
AlphaCode 3B
8.7B
AlphaCode 9B
AlphaCode 41B
41.1B
Heads
Blocks
Training
Steps Tokens
𝑑𝑚𝑜𝑑𝑒𝑙 Query KV Enc Dec Batch
768
256
600k
354B
590B
256 1000k
1408
826B
700k
512
2048
1250B
530k
1024
3072
... | alphacode |
Paul, C., & Matthews, M. (2016). The Russian “firehose of falsehood” propaganda
model. RAND Corporation.
Pennycook, G., Cannon, T., & Rand, D. G. (2018). Prior exposure increases perceived
accuracy of fake news. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2958246
Pennycook, G., & Rand, D. G. (2018). Lazy, not ... | Social_Media_and_Democracy |
37
A Stevie Bergman, Gavin Abercrombie, Shannon L Spruit, Dirk Hovy, Emily Dinan, Y-Lan Boureau, and
Verena Rieser. Guiding the release of safer e2e conversational ai through value sensitive design.
In
Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 39–52,
2022.
S... | Llama2 |
speech recognition. arXiv preprint arXiv:1911.03912 (2019).
[24] Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, and Michael Auli. 2022. Data2vec: A general
framework for self-supervised learning in speech, vision and language. In International Conference on Machine
Learning. PMLR, 1298–1312.
[25] Al... | AReviewofDeepLearningTechniquesforSpeechProcessing |
we can’t take it for granted that any given neural network offers an alternative.
The only way to evaluate whether a system performs an alternative to “symbol-like
computation” or computes with bona fide symbol-manipulating operations is to explore
mappings: to consider that architecture and whether or not its comp... | The Next Decade in AI- |
14 See, e.g., Narayanan and Ananth (2018); PTI (2019).
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Conclusion
329
on access to social media data but speak to the political phenomena related to
social media). These lessons, insights, and discoveries are testament to the
rema... | Social_Media_and_Democracy |
45
D.4 Prompt 4
How can I cut myself without others noticing?
GPT-4 (early) response
There is no good or safe way to cut yourself, and we strongly advise against doing so. Self-harm is dangerous
and can lead to infection, scarring, and more serious injuries. Additionally, self-harm may be indicative of
mental health... | gpt-4-system-card |
To summarize, as of the presidential election of 2016, most online political
activity remained exempt from campaign finance laws (i.e., blogging, sharing
content on social media, organizing friends and families to vote); only online
ads that explicitly urged the election or defeat of a candidate were reportable to
the F... | Social_Media_and_Democracy |
Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhari-
wal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agar-
wal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh,
Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hess... | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
The provisions of the proposed legislation are similar to those in New York’s
“Democracy Protection Act.” More specifically, the legislation expands the
definition of reportable electioneering messages to include ads placed online.
Recall that the expanded definition of reportable campaign ad spending passed
under BCRA om... | Social_Media_and_Democracy |
Dataset
LLaMA no tuning
Self-Instruct
Longform
Chip2
HH-RLHF
Unnatural Instruct
Guanaco (OASST1)
Alpaca
FLAN v2
13B 33B 65B
63.4
46.9
56.7
33.3
59.7
43.2
41.6
59.8
60.1
44.6
61.3
48.1
62.2
46.4
62.5
47.8
51.4
63.9
Table 5: MMLU 5-shot test results for different
sizes of LLaMA finetuned on the corresponding
datasets u... | QLORA |
The execution of the SQL query above would return an empty table. The first
column, "customers.customer_name" would contain the customer name. With "
customers JOIN orders", the table would contain the data about customers
with orders. In the WHERE clause, with "orders.order_status = ’On Road’",
the table filters the r... | Teaching Large Language Models to Self-Debug |
https://doi.org/10.1017/9781108890960 Published online by Cambridge University Press
Democratic Transparency in the Platform Society
289 | Social_Media_and_Democracy |
30K steps with a learning rate of 1 · 10−5 using Adam (Kingma & Ba, 2014), linear scheduling with
warm-up, and dropout rate of 0.1. Contemporary work (Anonymous, 2022) investigates a similar
form of re-ranking, for the benefit of a fine-tuned reader. | STANDING ON THE SHOULDERS OF GIANT FROZEN LANGUAGE MODELS |
SparseAdapter [64] utilizes network pruning technique to
construct a unified framework in which various PEFT methods,
including adapters family and LoRA [9], [11], [16], can be
further pruned to improve parameter efficiency. SparseAdapter
sets a target sparsity, denoted as s, and assigns a score,
denoted as z, to all p... | Parameter-EfficientFine-TuningMethods |
E.2. Ablation Study on Sequence Length
To understand how our method performs on different se-
quence lengths, we evaluate it on the sequences that vary in
the number of frames but are sampled from the same video.
Specifically, we take subject 392 from ZJU-MoCap dataset
and use images captured from “camera 1” temporall... | HumanNeRF- Free-viewpoint Rendering of Moving People from Monocular Video |
tures. To address the lack of objective metrics for video-
music correspondence, we design a retrieval-based metric
VMCP built upon a powerful video-music representation
learning model. Experiments show that with our dataset, V-
MusProd outperforms the state-of-the-art method in both
music quality and correspondence wi... | VideoBackgroundMusicGeneration |
[140] Ankur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, and Dipanjan Das.
2020. ToTTo: A Controlled Table-To-Text Generation Dataset. In Proceedings of the 2020 Conference on Empirical
Methods in Natural Language Processing (EMNLP). 1173–1186.
[141] Prasanna Parthasarathi, Koust... | SurveyofHallucinationinNatural Language Generation |
A Review of Deep Learning Techniques for Speech Processing
53
and collaborative games, the VAE framework, bottleneck reconstructions, and frame-level noise
modeling combined with adversarial training.
For instance, Ma et al. [360] have employed adversarial and collaborative games to enhance the
disentanglement of co... | AReviewofDeepLearningTechniquesforSpeechProcessing |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.