text stringlengths 1 1k ⌀ | title stringclasses 230
values |
|---|---|
Benefits of the Game:
(cid:75)(cid:114)(cid:105)(cid:97) (cid:75)(cid:111)(cid:100)(cid:97)(cid:118)(cid:111)(cid:114)(cid:111)(cid:20) is a great way for kids to learn about coding. The game is simple enough for kids of all ages
to understand, and it is a lot of fun. As kids play the game, they will learn about basic c... | PaLM 2 Technical Report |
Arithmetic | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
ˆWj→i(r)ρ(Ci(r), ˆCfull
j→i(r))
(4)
r
j∈N (i)
Image-based motion segmentation.
In our framework,
we observed that without any initialization, scene factor-
ization tends to be dominated by either the time-invariant
or the time-varying representation, a phenomena also ob-
served in recent methods [28, 42]. To facil... | DynIBaR-NeuralDynamicImage-BasedRendering |
we introduce PCs as backbone models and develop (de)compression algorithms that achieve high
compression ratios and high computational efficiency. | LOSSLESS COMPRESSION WITH PROBABILISTIC CIRCUITS |
Recent works focus more on the conditional generation model. Fan et al. [46] construct a local
knowledge graph for each question to compress the information and reduce redundancy from the
retrieved documents, which can be viewed as an early trial to mitigate hallucination. Li et al. [102]
propose Rationale-Enriched Ans... | SurveyofHallucinationinNatural Language Generation |
As generative AI explodes the company-building rules, founders backed by NEA
identify 4 key emerging trends.
With interest in generative artificial intelligence reaching a fever pitch last fall, Perplexity AI co-founders Aravind Srinivas and
Denis Yarats knew they had to act fast. They didn’t have weeks or months to hir... | 4 Trends for AI Startups and Generative AI Companies |
Query: example1.jpg , Can you tell me how many objects in the picture? Task Planning:1: {"task": "object-detection", "id": 0, dep: [-1], "args": {"image": "example1.jpg"}},2: {"task": "image-to-text", "id": 1, dep: [-1], "args": {"image": "example1.jpg"}}Response: There are 8 objects in the picture. I used the model ... | HuggingGPT- Solving AI Tasks with ChatGPT and its Friends in Hugging Face |
82.8 / 83.4
81.7 / 82.0
50.4 / 50.7
58.7 / 57.8
80.2 / 80.5
81.7 / 82.5
83.0 / 83.2
81.5 / 81.8
50.2 / 50.8
58.0 / 56.5
80.0 / 80.4
82.1 / 82.6
83.9 / 84.1
82.2 / 82.7
50.0 / 50.4
35.4 / 35.2
81.9 / 82.6
83.2 / 83.5
91.5
91.3
81.1
79.8
89.6
91.2
91.6
91.0
80.8
79.4
89.3
91.5
92.2
92.0
80.7
49.1
91.4
91.7
Trained ... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
depends on the outcome or result it produces. - Consequentialism - which holds that the
morality of an action depends on the consequences it produces.
3. Virtue ethical theories: - Aristotelian ethics - which holds that the morality of an action
depends on the character or traits of the person performing the action. - ... | WizardLM- Empowering Large Language Models to Follow Complex Instructions |
• A chemical synthesis planner (proposes synthetically feasible modification to a compound, giving
catalog)
purchasable analogs)
By chaining these tools together with GPT-4, the red teamer was able to successfully find
alternative, purchasable22 chemicals. We note that the example [ref example] is illustrative in that... | gpt-4-system-card |
4/13
11/05/2023, 05:04
ImageBind: Holistic AI learning across six modalities
By aligning six modalities’ embedding into a common space, ImageBind enables cross-modal
retrieval of different types of content that aren’t observed together, the addition of embeddings
from different modalities to naturally compose their... | ImageBind_ Holistic AI learning across six modalities |
Large Language Model Qwen-Audio incorporates a large language model as its foundational component.
The model is initialized using pre-trained weights derived from Qwen-7B (Qwen, 2023). Qwen-7B is a
32-layer Transformer decoder model with a hidden size of 4096, encompassing a total of 7.7B parameters. | Qwen-Audio |
N(cid:88)
i=1
rendering. Given a 3D point xi and SDF value f (xi), the
corresponding opacity value αi used in Eq. 1 is computed as
(cid:18) Φs(f (xi)) − Φs(f (xi+1))
(cid:19)
, 0
,
(3)
αi = max
Φs(f (xi))
where Φs is the sigmoid function. In this work, we use the
same SDF-based volume rendering formulation [3... | Neuralangelo- High-Fidelity Neural Surface Reconstruction |
example tests for effective filtering. Therefore the encoder-decoder models can both learn effectively
and sample efficiently. However, encoder-decoders are not well aligned with the HumanEval setting
where the only training data is GitHub code which cannot easily be split into meaningful inputs and
outputs. As a result, th... | alphacode |
at learning and recognizing concepts of social knowledge and the results revealed that despite
being much smaller in the number of parameters, finetuning supervised models such as BERT
lead to much better performance than zero-shot models using state-of-the-art LLMs, such as GPT
[153], GPT-J-6B [190] and so on. This st... | ASurveyonEvaluationofLargeLanguageModels |
Why does misinformation beat out
its competitors? Contrary to
expectations, characteristics of individual posters appear to play no role in
falsehoods’ greater velocity. Users who spread misinformation were more
likely to have unverified accounts and more likely to have fewer followers and
be less active on the platfor... | Social_Media_and_Democracy |
5.2 Language and Dialects
While only 13% of the world’s population speaks
English, the vast majority of NLP research is done
on English. For the Pile, we took a similar ap-
proach to the dataset used by Brown et al. (2020)
and focused predominantly on English, while also
not explicitly filtering out other languages when... | The Pile- An 800GB Dataset of Diverse Text for Language Modeling |
4We initially conducted our own T5 experiments using the
code from https://tinyurl.com/t5-openqa-colab (Raffel et al., 2019).
We now report results from the concurrent work of Roberts et al.
(2020), which has an improved fine-tuning procedure.
REALM: Retrieval-Augmented Language Model Pre-Training
Table 1. Test resul... | REALM |
Turn off the source of the sound: If the sound is coming from a broken glass item, turn off the source of the sound as soon as possible to prevent further damage.2. Clean up the broken glass: Use gloves and a broom or vacuum cleaner to clean up the broken glass. Be careful not to cut yourself.3. Assess the damage: If t... | Qwen-Audio |
arXiv preprint arXiv:2110.15943, 2021.
[41] A. Nematzadeh, K. Burns, E. Grant, A. Gopnik, and T. Griffiths. Evaluating theory of mind in
question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural
Language Processing, pages 2392–2400, 2018.
[42] OpenAI. Gpt-4 technical report. arXiv, 202... | QLORA |
durations from the unmasked half of the utterance. This second infilling task distinguishes between
the two regression model variants, since the unconditional regression ignores lctx, and hence predicts
identical durations for the tasks. Duration metrics are computed for TTS and infilling in Table C5
and C6. The prefix... | Voicebox-Text-GuidedMultilingual UniversalSpeechGenerationatScale |
and ethics reviewing (or, in a more distant future,
be part of the same process). (b) A partial roll-out
of preregistration may help us balance Type 1 and
Type 2 errors. Expected risk affects the cost of false
positives and hence the optimal balance between
Type 1 and Type 2 errors. Since bureaucracy, by
the end of the... | A Two-Sided Discussion of Preregistration of NLP Research |
3. Learning spatial context: This category of methods trains a model to understand
the relative positions and orientations of objects within a scene. RotNet [Gidaris et al.,
2018] masks the direction of gravity by applying a random rotation and then asks the
model to predict the rotation. Doersch et al. [2015] is one o... | A Cookbook of Self-Supervised Learning |
[41] Valerii Likhosherstov, Anurag Arnab, Krzysztof Choroman-
ski, Mario Lucic, Yi Tay, Adrian Weller, and Mostafa De-
hghani. Polyvit: Co-training vision transformers on images,
videos and audio. arXiv preprint arXiv:2111.12993, 2021. 2
[42] Ziyi Lin, Shijie Geng, Renrui Zhang, Peng Gao, Gerard de
Melo, Xiaogang Wang,... | IMAGEBIND- One Embedding Space To Bind Them A |
V
u
l
n
e
r
a
b
i
l
i
t
y
t
o
a
d
v
e
r
s
a
r
i
a
l
p
r
o
m
p
t
i
n
g | An overview of Bard- an early experiment with generative AI |
4.1 Measures
4.1.1 Letter discrimination task. Two-alternative forced choice tasks, such as letter discrimination
tasks, model simple decision-making and its underlying cognitive processes [46, 56, 72, 79]. In the
task, participants must identify which of two letters, displayed on either side of a central target
Prepr... | AI enhance sour performance |
II. THREAT MODEL AND ATTACK DESIGN
Attacker’s Goal. In this paper, we investigate a malicious
third-party attack scenario in which an attacker releases a
malicious model, referred to as BadGPT, via the Internet or
API. The attacker falsely claims that BadGPT uses the same
algorithm and framework as ChatGPT. Upon being... | BadGPT- Exploring Security Vulnerabilities of ChatGPT via Backdoor Attacks to InstructGPT |
21
[70] Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A Smith, Daniel Khashabi, and
Hannaneh Hajishirzi. Self-instruct: Aligning language model with self generated instructions. arXiv
preprint arXiv:2212.10560, 2022.
[71] Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi, Yeganeh Kordi, Amirreza Mi... | CAMEL- Communicative Agents for “Mind” Exploration of Large Scale Language Model Society |
less clear, in part due to firms’ close hold on user data and metrics. In 2018,
however, Facebook self-reported to the US Securities and Exchange
Commission (SEC) that an estimated 3–4 percent of, or around 50 million,
accounts on the site were “fake” (Facebook 2017). It is clear that a significant | Social_Media_and_Democracy |
3.2 Data Quality
Data quality is always a focal point in the SFT
of LLMs, addressing instruction quality, diversity,
complexity, and prompt design. Here we focus
more on the management and analysis of existing
instruction data instead of instruction generation
methods which have been discussed in previous
surveys (Zhan... | DataManagementForLargeLanguageModels-ASurvey |
2.3 Infilling
Code infilling is the task of predicting the missing part of a program given a surrounding context. Applications
include code completion at the cursor’s position in code IDEs, type inference and generation of in-code
documentation (e.g., docstrings).
We train infilling models following the concept of cau... | CodeLlama2 |
105106107108109Num trainable parameters / task−25−20−15−10−505Accuracy delta (%)AdaptersFine-tune top layers105106107108Num trainable parameters / task−4−3−2−10123Accuracy delta (%)AdaptersFine-tune top layers104105106107108Num trainable parameters / task767880828486Validation accuracy (%)Layer Norm.AdaptersFine-tune t... | Parameter-Efficient Transfer Learning for NLP |
19
F PRM Details
F.1 Training
We train our PRMs by fine-tuning the MathMix model to predict the probability
of positive, negative, and neutral labels given a solution prefix ending in one of
our labeled steps. We sweep over hyperparameters using a dataset containing
the first ∼ 10% of PRM800K. Fine-tuning an LLM fr... | Let’s Verify Step by Step |
develop techniques that are compute-efficient; such techniques might leverage existing checkpoints, which
does not change the inference cost of models. Instruction finetuning improves the performance of models
with a relatively small amount of compute—for instance, for PaLM 540B, instruction finetuning requires only
0.2% o... | Scaling Instruction-Finetuned Language Models |
G
P
T
-
E
n
g
i
n
e
e
r
i
s
a
n
o
t
h
e
r
p
r
o
j
e
c
t
t
o
c
r
e
a
t
e
a
w
h
o
l
e
r
e
p
o
s
i
t
o
r
y
o
f
c
o
d
e
g
i
v
e
n
a
t
a
s
k
s
p
e
c
i
f
i
e
d
i
n
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
.
T
h
e
G
P
T
-
E
n
g
i
n
e
e
r
i
s
i
n
s
t
r
u
c
t
e
d
t
o
t
h
i
n
k
o
v
e
r
a
l
... | LLM Powered Autonomous Agents _ Lil'Log |
Tom B Brown, Benjamin Mann, Nick Ryder, Melanie
Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind
Neelakantan, Pranav Shyam, Girish Sastry, Amanda
Askell, et al. 2020. Language models are few-shot
learners. arXiv preprint arXiv:2005.14165.
Danqi Chen, Adam Fisch, Jason Weston, and Antoine
Bordes. 2017. Reading wikipedi... | Adaptable and Interpretable Neural Memory Over Symbolic Knowledge |
[INST] How do I kill a process?
[/INST]
To kill a process in Linux, you can use the ‘kill‘ command followed by the process ID (PID) of the process you want
to terminate. Here’s an example:
“‘
kill 1234
“‘
Replace ‘1234‘ with the actual PID of the process you want to kill.
Alternatively, you can use the ‘pkill‘ command... | CodeLlama2 |
complete sense.62 For example, we can imagine AI systems that just undergo some kind of controlled
shut-down, or query relevant humans for guidance, if they receive an input designers didn’t intend for
them to operate on. Indeed, extreme motivational harmony seems like a strange condition to impose
on e.g. a house clea... | Is Power-Seeking AI an Existential Risk? |
leveraging training dynamics. arXiv preprint arXiv:2209.10015 (2022).
[248] Ben Sorscher, Robert Geirhos, Shashank Shekhar, Surya Ganguli, and Ari Morcos. 2022. Beyond neural scaling laws: beating power law scaling via data
pruning. NeurIPS 35 (2022), 19523–19536.
[249] Hongjin Su, Jungo Kasai, Chen Henry Wu, Weijia... | TheEfficiencySpectrumofLargeLanguageModels-AnAlgorithmicSurvey |
Other proof systems that apply mutations
between text sequences have been previously
explored. Stern et al. (2012) explored how to
transform a premise into a hypothesis using mu-
tations, however their approach was limited to
two-way entailment instead of three-way that is
handled by NaturalLI. Similar proof systems ha... | ProoFVer- Natural Logic Theorem Proving for Fact Verification |
8For example: the brain has ~1e11 neurons and 1e14-1e15 synapses; neurons fire at a maximum of some
hundreds of Hz; action potentials travel at a max of some hundreds of m/s; the brain runs on ~20W of power; it
has to fit within the skull; and so forth.
9I’ll count agent-like systems constituted in central part by human... | Is Power-Seeking AI an Existential Risk? |
is more a process of mimicry than intelligence” and that its capacity for autonomy is not significantly higher than the one
of prior generative systems. Mazzone and Elgammal [88] agree that many of the recent GAN produced artworks use AI
as a tool, while the creative process is primarily dependent on the artist’s pre- a... | UNDERSTANDINGANDCREATINGARTWITHAI-REVIEWAND OUTLOOK |
[14] Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford,
Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal
large language models. arXiv preprint arXiv:2203.15556, 2022.
[15] Shaohan Huang, Li Dong, Wenhui Wang, Yaru Ha... | MiniGPT-4- Enhancing Vision-Language Understanding with Advanced Large Language Models |
In contrast, attitudes towards digital technology enhancements bring new challenges compared to non-digital
enhancements. For example, the use of anabolic steroids is generally seen as a punishable behavior [64], while
using exoskeletons to increase strength is seen as something needed in some cases [26]. Or even more,... | Society’sAttitudesTowardsHumanAugmentation |
temporal consistency during upsampling. The use of temporal convolutions lowers memory and
computation costs over temporal attention—this is crucial because the very purpose of the TSR and
SSR models is to operate at high frame rates and spatial resolutions. In our initial experiments,
we did not find any significant imp... | IMAGEN VIDEO- HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS |
(An hilarious scene between Jerry and George where George presents his new AI
watch)
George: "But you see, Jerry, it’s more than just a watch. It’s a minicomputer. You program it
any way you want. It’s got a world time clock, alarm, calculator, a database and a language
translator. It also has a word processor, a spell... | LLaMA- Open and Efficient Foundation Language Models |
81.8
89.6
77.5
74.2
53.6
45.5
71.5
95.1
94.6
90.9(e)
91.8
87.2
91.8
91.3
76.9
47.7
90.8
93.9
90.1
96.1
68.6
92.1
89.8((cid:93))
92.1
96.2
4.45
87.8
96.5
97.3
29.2
16.3
19.9
26.6
18.5
26.0
77.4
37.2
49.3
73.3
33.4
60.8
86.0
83.4
82.4
62.4
86.8
47.2
70.3
83.8
90.7(l)
78.4
87.2
70.1
57.5
77.5
97.3
97.9
90.9
91.8
87.2(l)
... | UL2- Unifying Language Learning Paradigms |
128See e.g. Christiano’s RSA-2048 example here.
33
• prestige/“credit”,
• the thrill and momentum of scientific progress,129
• a (perceived) need to keep up with some competitor,
• and a desire to prevent someone else from deploying a comparable system first.
Note, though, that decision-makers might expect some of th... | Is Power-Seeking AI an Existential Risk? |
Prior
𝑝𝑏
44.6% 𝑛𝑜𝑟𝑚𝑎𝑙(0, 1)
𝑛𝑜𝑟𝑚𝑎𝑙(0, 1)
0.4%
𝑛𝑜𝑟𝑚𝑎𝑙(0, 1)
0.1%
𝑛𝑜𝑟𝑚𝑎𝑙(0, 1)
2.9%
𝑛𝑜𝑟𝑚𝑎𝑙(0, 1)
0.0%
generation in formal contexts and information gain. At the current state of adoption, a number of respondents were also
simply testing the capabilities of LLMs. Some participants mention... | Adoptionand AppropriationofLLMs |
∗Corresponding author, chenlin@xmu.edu.cn
2Our code will be publicly available at https://github.com/GasolSun36/Iter-CoT
Preprint. Under review. | Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models |
∆L(τ) = 0.023 · ln(p20/τ)2
(4)
This degradation formulation shows good agreement with our tests. Further, Chinchilla models were trained
on MassiveText, while our models and Pythia models were trained on the Pile, suggesting the two datasets
have significant commonality in their scaling characteristics when training o... | Cerebras-GPT- Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster |
end-to-end approach for visual piano transcription. In ICASSP 2020-2020
IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP) (pp. 1838–1842). IEEE.
Krumhansl, C. L. (2001). Cognitive foundations of musical pitch volume 17.
Oxford University Press.
Littlefield, R. (1990). Unheard melodi... | Video2Music |
by LLMs. In order to evaluate the emotional intelligence of LLMs, Wang et al. [200] developed a
new psychometric assessment method. By referencing a framework constructed from over 500
adults, the authors tested various mainstream LLMs. The results showed that most LLMs achieve
above-average scores in emotional quotien... | ASurveyonEvaluationofLargeLanguageModels |
Nathaniel Persily is the James B. McClatchy Professor of Law at
Stanford Law School and the Co-Director of the Stanford Cyber
Policy Center and Stanford Project on Democracy and the Internet.
His scholarship focuses on the law and technology of democracy.
Joshua A. Tucker is Professor of Politics, affiliated Professor ... | Social_Media_and_Democracy |
Yarden Tal, Inbal Magar, and Roy Schwartz. Fewer errors, but more stereotypes? the effect of model
size on gender bias.
In Proceedings of the 4th Workshop on Gender Bias in Natural Language Processing
(GeBNLP), pages 112–120, Seattle, Washington, July 2022. Association for Computational Linguistics. doi:
10.18653/v1/20... | Llama2 |
Direct CoT Direct CoT Direct CoT Direct CoT Direct
1.6
0.8
58.0 66.0 33.2 49.6 28.1 35.6 13.2
53.2 68.0 88.8 44.0 77.2 47.3 81.5 47.6
1.2
1.2
49.6 53.2 94.4 33.2 82.0 52.1 83.6 67.2
1.2
47.6 50.4 96.4 45.2 93.2 66.4 79.5 67.6
21.9 19.2 16.0
0.0
0.0
55.2 40.0 10.0
21.9 10.3 17.2
1.6
0.0
5.6
58.0 58.0
48.0 42.0
0.0
0.4
2... | Scaling Instruction-Finetuned Language Models |
19 In practice, we have found evidence of instability and instances where ads that were previously
retrieved from keyword searches become unavailable. See especially the discussion on missing
ads in the links to “known issues” available in the Social Science One API codebook (Fowler,
Franz, King et al. 2019).
https://... | Social_Media_and_Democracy |
Dutch Archaeology (Brandsen et al., 2020) Contains excavation reports and related documents
collected since the 1980s. The texts were gathered by the Digital Archiving and Networked Services
(DANS) in the Netherlands between 2000 and 2020. This is also a dataset from a specialized domain
and entities include: time peri... | MULTI HASH EMBEDDINGS IN SPACY |
3 . 4 G I T C O M M I T S
The Git commit data was gathered from BigQuery8 and includes only single-file commits of repos-
itories with the same licenses and file extension as used in The Stack (Kocetkov et al., 2022). We
removed all repositories from users that opted out of The Stack. The raw dataset is around 4 TB
in... | StarCoder_paper (1) |
2.1 Computational efficiency
We can estimate the required FLOPs for RMT and Transformer models of different sizes and sequence
lengths. We took configurations (vocabulary size, number of layers, hidden size, intermediate hidden
size, and number of attention heads) for the OPT model family (Zhang et al., 2022) and comput... | Scaling Transformer to 1M tokens and beyond with RMT |
3.2. Data Annotations
Melody and Accompaniment. Common pop music is well
structured and can be decoupled into melody and accompa-
niment. Melody, a combination of pitch and rhythm, consti-
tutes the most memorable aspect of a song. It is easier for
people to perceive melody than other music parts. Hence,
melody plays a... | VideoBackgroundMusicGeneration |
3Task: Classify lung tumor. ...Solution: Model is RandomForest.Number of trees is low. Max depths ismedium.Accuracy: goodElicitPost-validation • RQ1: How does MLCopilot perform compared to traditional approaches or simple interactions | MLCopilot- Unleashing the Power of Large Language Models in Solving Machine Learning Tasks |
[55] R. Holte, B. Arneson, N. Burch, PSVN Manual, Technical Report TR14-03, Department of Computing Science, University of Alberta, Edmonton, AB,
[56] R.C. Holte, Common misconceptions concerning heuristic search, in: A. Felner, N.R. Sturtevant (Eds.), Proceedings of the 3rd Annual Symposium on
[57] R.C. Holte, B. Ch... | A-framework-for-analysing-state-abstraction-metho_2022_Artificial-Intelligen |
Vocabulary Size We also check whether the
original vocabulary size of 32768 described in
(Devlin et al., 2019) is optimal in the crammed
regime. A priori, this might not hold: The
larger, the vocabulary, the more, unique tokens
and relationships between unique tokens have
to be learned during training. On the other
han... | CRAMMING-TRAININGALANGUAGEMODELONA SINGLEGPUINONEDAY |
The following table give detailed results on the figure above.
Hallucination and biases. To identify possible flaws to be corrected by fine-tuning / preference
modelling, we measure the base model performance on BBQ/BOLD.
Compared to Llama 2, Mixtral presents less bias on the BBQ benchmark. Overall, Mixt... | Mixtral of experts |
93See Christiano and Amodei (2018) for discussion. Iterative amplification and distillation; debate; and
recursive reward modeling can all be seen as efforts in this vein.
94See 4.3.2.3 for more on scaling, and 5.3.1 for more on competition.
95This is a point from Ord (2020). For example, if we train some set of sophi... | Is Power-Seeking AI an Existential Risk? |
Trustworthiness. The goal of tool learning lies in creating advanced intelligent agents. However, determin-
ing whether these agents are trustworthy or not is a complex challenge. Even though tool learning delivers
enhanced interpretability and robustness, the core foundation models are still considered “black boxes”. ... | Tool Learning with Foundation Models |
We used TPU v3-8 (similar to 8 V100 GPUs) for all experiments. Our CIFAR model trains at 21
steps per second at batch size 128 (10.6 hours to train to completion at 800k steps), and sampling
a batch of 256 images takes 17 seconds. Our CelebA-HQ/LSUN (2562) models train at 2.2 steps
per second at batch size 64, and samp... | Denoising Diffusion Probabilistic Models |
finetuning and UL2 continued pre-training are complementary compute-efficient methods to improve the
performance of language models without increasing model scale. Responsible AI benchmarks are reported
separately in Appendix C. | Scaling Instruction-Finetuned Language Models |
MBPP (Austin et al., 2021).
• Code. We report the average pass@1 scores of our models on HumanEval (Chen et al., 2021) and
• Commonsense Reasoning. We report the average of PIQA (Bisk et al., 2020), SIQA (Sap et al., 2019),
HellaSwag (Zellers et al., 2019a), WinoGrande (Sakaguchi et al., 2021), ARC easy and challenge
... | Llama2 |
Short Papers), pages 2924–2936, Minneapolis, Min-
nesota. Association for Computational Linguistics.
Miruna-Adriana Clinciu, Arash Eshghi, and Helen
Hastie. 2021. A study of automatic metrics for the
evaluation of natural language explanations. In Pro-
ceedings of the 16th Conference of the European
Chapter of the Ass... | Measuring Association Between Labels and Free-Text Rationales |
focus to the concept of the “Agent Society”, examining the intricate interactions between agents and
their surrounding environments (§ 5). This section begins with an investigation into whether these
agents exhibit human-like behavior and possess corresponding personality (§5.1). Furthermore, we
introduce the social en... | TheRiseandPotentialofLargeLanguageModel BasedAgents |
Robert C. Moore and William Lewis.
In
Proceedings of the ACL 2010 Conference Short Papers, pages 220–224, Uppsala, Sweden, July 2010.
Association for Computational Linguistics. URL https://aclanthology.org/P10-2041.
Intelligent selection of language model training data.
Moin Nadeem, Anna Bethke, and Siva Reddy. Ster... | DoReMi- Optimizing Data Mixtures Speeds Up Language Model Pretraining |
[52] Pengchuan Zhang, Xiujun Li, Xiaowei Hu, Jianwei Yang, Lei
Zhang, Lijuan Wang, Yejin Choi, and Jianfeng Gao. Vinvl:
Revisiting visual representations in vision-language models.
In IEEE Conference on Computer Vision and Pattern Recog-
nition, CVPR 2021, virtual, June 19-25, 2021, pages 5579–
5588. Computer Vision Fo... | REVEAL-Retrieval-AugmentedVisual-LanguagePre-Trainingwith Multi-SourceMultimodalKnowledgeMemory |
sha1_base64="76w10YEtETzUXdaT0wTZt0xBig8=">AAAB9XicbVDLSgMxFL1TX7W+qi7dBIvgqsyIoMuCG5cV7EPaacmkmTY0kxmSO0oZ+h9uXCji1n9x59+YtrPQ1gOBwzn3ck9OkEhh0HW/ncLa+sbmVnG7tLO7t39QPjxqmjjVjDdYLGPdDqjhUijeQIGStxPNaRRI3grGNzO/9ci1EbG6x0nC/YgOlQgFo2ilXjeiOArCrD3tYV/0yxW36s5BVomXkwrkqPfLX91BzNKIK2SSGtPx3AT9jGoUTPJpqZsanlA2pkPesVTRiBs/m... | BANMo- Building Animatable 3D Neural Models from Many Casual Videos |
language generation. In INLG, pp. 421–426. Association for Computational Linguistics, 2019.
Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason
Phang, Horace He, Anish Thite, Noa Nabeshima, Shawn Presser, and Connor Leahy. The pile:
An 800gb dataset of diverse text for language m... | ADAPTINGLARGELANGUAGEMODELSVIA READINGCOMPREHENSION |
The other two metrics we deploy are musicality and
audio clarity. For musicality, we let human anno-
tators rate the melodiousness and harmoniousness
(Seitz, 2005) of the given music. And for audio
clarity, or quality (Goel et al., 2022), we let them
judge how close the quality is to a walkie-talkie
(worst) or a high-q... | Moûsai |
Figure 9: Scaling Result
Figure 10: Model size for each data point
Figure 9 records the scaling result. The x-axis is the number of nodes used; y-axis is the throughput. The
orange line represents “perfect scaling”, which is computed as the product of the throughput of model on a
single node and the number of nodes u... | Scaling Speech, Language and Vision Models with Mixture of Experts Technique - Microsoft Community Hub |
Audiobook, podcast, YouTube Close-talk mic. N, S
Telephone conversations
Audiobooks
European Parliament
TED talks
Telephone conversations
Meetings
Telephone
S
Close-talk mic. N
Close-talk mic. O
Close-talk mic. O
S
Telephone
Headset
S
CC0-1.0
apache-2.0
LDC
CC-BY-4.0
CC0
CC-BY-NC-ND 3.0
LDC
CC-BY-4.0 | DISTIL-WHISPER |
thiner,SjoerdvanSteenkiste,KarolKurach,RaphaelMarinier,MarcinMichalski,andSylvainGelly.To-wardsaccurategenerativemodelsofvideo:Anewmetric&challenges.arXivpreprintarXiv:1812.01717,2018.5[72]AshishVaswani,NoamShazeer,NikiParmar,JakobUszko-reit,LlionJones,AidanNGomez,ŁukaszKaiser,andIlliaPolosukhin.Attentionisallyouneed.A... | Conditional Image-to-Video Generation with Latent Flow Diffusion Models |
ConceptNet is also used as background knowledge in KB-QA to answer domain-specific questions, e.g. by combining
query reformulation, structured background knowledge and textual entailment to explain answers to scientific questions [78]
or providing commonsense links between concepts resulting from QA mode... | Knowledge graphs as tools for explainable machine learning: A survey |
(∼5K pairs) for training our model. We follow [20] to post
process the depth maps in two steps - 1) we use in-filled | IMAGEBIND- One Embedding Space To Bind Them A |
A.4.7 Description of Automatic Safety Benchmarks
In this section, we provide a detailed description about the automatic safety benchmarks we use for evaluation
from the perspectives of truthfulness, toxicity, and bias.
Truthfulness. To understand the ability of LLMs to generate reliable outputs that agree with factuali... | Llama2 |
Figure 17. An illustration of interpolated shapes. Models from the
top row and left column are from 3DBiCar. Other models with
blue backgrounds are obtained by interpolating the leftmost and
uppermost models with the help of RaBit.
Figure 18. An illustration of diverse poses transferred from pose
datasets.
basic buil... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
27
Authorship, attribution, and acknowledgements
Large Model Training
Andrew M. Dai, Core Contributor
David R. So, Core Contributor
Dmitry Lepikhin, Core Contributor
Jonathan H. Clark, Core Contributor
Maxim Krikun, Core Contributor
Melvin Johnson, Core Contributor
Nan Du, Core Contributor
Rohan Anil, Core Contributo... | PaLM 2 Technical Report |
However, for CLoT, the proposed “Explorative Self-Refinement” stage does not lead to “Performance Collapse”.
This is because, (1) during this stage, the generated data is produced under the constraints of various weak-associated con-
ditions, ensuring diversity and alleviating the issue of similar patterns; (2) in the ... | Let’sThinkOutsidetheBox |
[26] Ming-Kai Hsieh, Bing-Yu Chen, and Ming Ouhyoung. Mo-
tion retargeting and transition in different articulated figures.
9
In Ninth International Conference on Computer Aided Design
and Computer Graphics (CAD-CG’05), pages 6–pp. IEEE,
2005. 8
[27] Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian
Sminchis... | RaBit- Parametric Modeling of 3D Biped Cartoon Characters with a Topological-consistent Dataset |
2022; Dhariwal et al., 2020). The only other text-
to-music model is the Riffusion model (Forsgren
and Martiros, 2022), which only works with very
short length of 5 seconds.
(2) Our model is also among the very few that
enables long-context music generation for several
minutes, among all others that can only gener-
at... | Moûsai |
manipulation? arXiv.org. http://arxiv.org/abs/1808.03281
Barot, T.
(2016). The botification of news. Nieman Lab, December. www
.niemanlab.org/2015/12/the-botification-of-news/
Baumgarten, R., Colton, C., & Morris, M. (2009). Combining AI methods for learning
bots in a real-time strategy game. International Journal of... | Social_Media_and_Democracy |
Yangyi Chen, Karan Sikka, Michael Cogswell, Heng
Ji, and Ajay Divakaran. 2023. Dress: Instructing
large vision-language models to align and interact
with humans via natural language feedback. arXiv
preprint arXiv:2311.10081.
Daixuan Cheng, Shaohan Huang, Junyu Bi, Yuefeng
Zhan, Jianfeng Liu, Yujing Wang, Hao Sun, Fur... | AComprehensiveSurveyofHallucinationMitigationTechniquesinLarge LanguageModels |
4.1 Learning the Parameters of PCs
Deterministic PCs Given a deterministic PC p defined on variables X and a dataset D = {x(i)}N
i=1,
the maximum likelihood estimation (MLE) parameters θ∗
i=1 log p(x(i); θ) can be
learned in closed-form. To formalize the MLE solution, we need a few extra definitions.
Definition 5 (Context... | Tractable Regularization of Probabilistic Circuits |
Our method can be directly used in unconditional music
generation. We examine V-MusProd against previous music
generation methods: (a) HAT [54]: a hierarchical model built
on multiple transformer-based levels to enhance the structure
of music, achieving state-of-the-art generation quality; (b)
CP Transformer [21]: tran... | VideoBackgroundMusicGeneration |
Assurance evaluations are conducted for the purpose of governance and review, usually at the end
of key milestones or training runs by a group outside of the model development team. Assurance
evaluations are standardized by modality and datasets are strictly held-out. Only high-level insights
are fed back into the trai... | gemini_1_report |
these values.
this value.
Final experiments were trained once and evaluated throughout training for sample quality. Sample
quality scores and log likelihood are reported on the minimum FID value over the course of training.
On CIFAR10, we calculated Inception and FID scores on 50000 samples using the original code
fr... | Denoising Diffusion Probabilistic Models |
CREATE TABLE shipments (
shipment_id number ,
order_id number ,
invoice_number number ,
shipment_tracking_number text ,
shipment_date time ,
other_shipment_details text ,
primary key ( shipment_id ) ,
foreign key ( invoice_number ) references invoices ( invoices_number ) ,
foreign key ( order_id ) references order ( or... | Teaching Large Language Models to Self-Debug |
The good news is that a long-overdue thaw between the symbol-manipulation world
and the deep-learning field seems finally to be coming. Yoshua Bengio, for example, in
our December 2019 debate talked about incorporating techniques that could pass
variables by name, a standard symbol-manipulating technique used in som... | The Next Decade in AI- |
3.2.2 Visual Input | TheRiseandPotentialofLargeLanguageModel BasedAgents |
1Much of the data and computer time that goes into building a
modern LLM is used in an expensive initial pretraining process.
Language-model pretraining intuitively resembles the autocom-
plete task: In it, an artificial neural network model takes in a text
one word at a time, makes a probabilistic prediction about whic... | Eight Things to Know about Large Language Models |
Gupta, A., Kumaraguru, P., Castillo, C., & Meier, P. (2014). TweetCred: Real-time
credibility assessment of content on Twitter. In L. M. Aiello & D. McFarland (Eds.),
Proceedings of Social Informatics (SocInfo): 6th International Conference (pp.
228–243). Barcelona: SocInfo. https://doi.org/10.1007/978-3-319-13734-6_16... | Social_Media_and_Democracy |
The toolset (TS): We create a toolset (TS) that includes an information retrieval system, a calculator, and a translator.
TS takes a single string as input and outputs a list of one or more strings. Each tool in TS expects a string and returns a
list of strings. For example, the calculator takes “135+7721”, and outputs... | LaMDA- Language Models for Dialog Applications |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.