Sherry Feng

Check out our latest blog to see how our Notification Team harnesses the power of Graph Neural Network (GNN) models to create personalized restaurant recommendations for user notifications in our latest blog.

62

Could RAFT (Retrieval-Augmented Fine-Tuning) be the next evolution of RAG architecture? I am following this with great interest. https://lnkd.in/eHDx3rHz

13

Excellent introduction to Autogen! This is the type of configurable agent framework I've been looking for

7

1 Comment

It's always funny to see when a top-tier research group like the folks at Stanford under Christopher Manning do something, and then quickly (and quietly, but not so quietly that I didn't notice) back-peddle. For those who don't know, PyReft is a library for *extremely efficiently* fine-tuning language models. You might not realize it yet, but techniques like it are going to blow regular PeFT techniques like LoRa out of the water hard very soon. The authors, in order to prove how good their technique is, have done many experiments. One of these experiments was to jailbreak the model and make it "unaligned". I think that these kind of experiments are fine/good, and certainly in line with the kind being done by the lesswrong community (who are AI doomers) with orthogonalization - but it's interesting to see that someone else decided that this kind of script was risky. For anyone who cares, you can access the scripts easily via this link (click "view file" on the "deleted" files), and see how researchers are tricking these models to become unhinged (or german speaking, or instruction tuned, or funny, etc) with surgical modifications to only ~0.0001% of the parameters. https://lnkd.in/gGtQTFNU

153

11 Comments

Great to see support for our exciting M2Lingual IFT data 🔥 🔥 🙏 🙌 : ** Contains modest size of 𝟏𝟖𝟐𝐊 𝐦𝐮𝐥𝐭𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥 𝐢𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧𝐬 📖 ** Covers 𝟕𝟏 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 (that include 37 low resource languages) 🌐 ** Contains complex multilingual instructions, evolved (following WizardLM) from seed IFT pairs from --- 19 diverse multilingual NLP tasks, --- Native speaker annotated general IFT pairs in 65 different languages 🗣️ ** Evol mutations to generate more IFT pairs are guided by 1) 𝐍𝐋𝐏 𝐭𝐚𝐬𝐤 𝐬𝐩𝐞𝐜𝐢𝐟𝐢𝐜 𝐞𝐯𝐨𝐥 𝐭𝐚𝐱𝐨𝐧𝐨𝐦𝐲, 2) reasoning, slang, dialect, transliteration, etc for general instructions, 3) 𝐦𝐮𝐥𝐭𝐢-𝐭𝐮𝐫𝐧 𝐞𝐯𝐨𝐥 𝐭𝐚𝐱𝐨𝐧𝐨𝐦𝐲 for generating different types of - {follow-up, refinement, expansion, recollection} conversational follow ups 🔄 ** 𝐌𝐮𝐥𝐭𝐢-𝐭𝐮𝐫𝐧 𝐜𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐈𝐅𝐓 pairs upto 3 turns 🗨️

16

Meta's Llama 3: A watershed moment for Open-Source LLMs? Exciting news in the tech world as Meta's launch of Llama 3 last week (https://lnkd.in/gk7VhheQ) is redefining the open source LLM landscape! 🚀   While 2023 saw remarkable strides with Llama 2, Falcon 180B, and Mistral 7B, Llama 3 is setting new standards with its unparalleled raw performance. The larger 70B version is now only second to OpenAI’s GPT-4 Turbo on the LMSYS leaderboard for English prompts (https://lnkd.in/gce2yKhy), and the smaller 8B model outperforms GPT-3.5, which powers the free version of ChatGPT. 🏆   Users are raving about Llama 3's faster inference compared to GPT, making it a game changer in the realm of super-fast image generation. Its integration with Meta products allows for real-time modifications with prompts, vastly enhancing user experience. Since it is open source, folks on X have even shown how incredible it is when paired with Groq!🌐   What's fascinating is that Llama 3 used ~75x more data (15T compared to ~200B) than the Chinchilla paper's benchmark (https://lnkd.in/gzv5H3FN), and Meta claims it still wasn't converging. This underscores the vast potential of data in enhancing models with sub-100B parameters. Yes, the training costs are higher, but the lower parameter size reduces inference costs, paving the way for running LLMs on local devices in the future! 💡   The only downside? A context length of just 8k when 128k is becoming the norm across top models. But hey, every innovation has room for improvement, right? And they have indicated that this is bound to increase in the coming months as well!   Kudos to Meta for pushing the boundaries of open-source development with Llama 3. Its impressive performance, combined with Meta's product integration, makes this a development worth watching! Do you think the 400B+ model (summer launch expected) would beat OpenAI's GPT 4 Turbo and maybe even be as good as GPT 5?🎉   #Meta #Llama3 #OpenSource #AI #LLM #TechInnovation #MachineLearning #GenAI #GPT4 #OpenAI

25

5 Comments

Prompt engineering is a rapidly-evolving topic in AI, but recent research can be grouped into four categories... (1) Reasoning: Simple prompting techniques are effective for many problems, but more sophisticated strategies are required to solve multi-step reasoning problems. - [1] uses zero-shot CoT prompting to automatically generate problem-solving rationales to use for standard CoT prompting. - [2] selects CoT exemplars based on their complexity (exemplars that have the maximum number of reasoning steps are selected first). - [3] improves CoT prompting by asking the LLM to progressively refine the generated rationale. - [4] decomposes complex tasks into several sub-tasks that can be solved via independent prompts and later aggregated into a final answer. (2) Tool Usage: LLMs are powerful, but they have notable limitations. We can solve many of these limitations by teaching the LLM how to leverage external, specialized tools. - [5, 6] finetune a language model to teach it how to leverage a fixed, simple set of text-based APIs when answering questions. - [7] uses a central LLM-based controller to generate a program—written in natural language—that composes several tools to solve a complex reasoning task. - [8] uses a retrieval-based finetuning technique to teach an LLM to adaptively make calls to APIs based on their documentation when solving a problem. - [9] uses an LLM as a central controller for leveraging a variety of tools in the form of deep learning model APIs. - [10, 11] integrates code-capable LLMs with a sandboxed Python environment to execute programs when solving problems. (3) Context Window: Given the emphasis of recent LLMs on long contexts for RAG / few-shot learning, the properties of context windows and in-context learning have been studied in depth. - [12] shows that including irrelevant context in the LLM’s prompt can drastically deteriorate performance. - [13] finds that LLMs pay the most attention to information at the beginning/end of the prompt, while information placed in the middle of a long context is forgotten. - [14] proposes a theoretically-grounded strategy for optimally selecting few-shot exemplars. (4) Better Writing: One of the most popular use-cases of LLMs is for improving human writing, and prompt engineering can be used to make more effective writing tools with LLMs. - [15] improves the writing abilities of an LLM by first generating an outline and then filling in each component of the outline one-by-one. - [16] uses a smaller LLM to generate a “directional stimulus” (i.e., a textual hint) that can be used as extra context to improve an LLM’s writing ability on a given task. - [17] improves the quality of LLM-generated summaries by iteratively prompting the LLM to increase the information density of the summary.

221

8 Comments

4 models made available today: - Meta Chameleon: 7B & 34B language models - Meta Multi-Token Prediction LLM - Meta JASCO: text-to-music models - Meta AudioSeal: audio watermarking model

31

At Uniphore, we’ve developed a highly efficient fine-tuning methodology that turns smaller LLMs (and others) into truly formidable tools—and makes LLM-based AI more accessible to enterprises everywhere without sacrificing their needs for data privacy, cost and latency with high accuracy. We benchmarked fine-tuned Llama-3 8B, Llama-3 8B Vanilla and GPT-4 on eight different datasets, covering tasks such as question answering (Q&A), Q&A with structured data/tables, named entity recognition (NER), summarization and agent-based functions that are most relevant for the enterprise AI space. As a result, we achieved a 15% median increase in accuracy as compared to GPT-4 and a 26% median increase in performance as compared to Vanilla Llama-3 8B and an 11x reduction in inference cost. We also developed our in-house framework with optimized code for parameter-efficient instruction fine-tuning of LLMs using the best-in-class available toolsets along with our modifications, which led to an eight-fold reduction in the training time. This allows us and our customers to instruction fine-tune models using our APIs in just a few hours based on their datasets and tasks, ensuring they work effectively for their applications. Details here: https://lnkd.in/gST5tQNZ

137

4 Comments

Meta is still evil. Their embrace of open source LLMs fits nicely into a grand scheme for dominating our endorphins. Evil Plan: 1. Build a VR platform that would be fun if it were busy. [Horizon Worlds] 2. Normalize interacting with bots on existing platforms. [Facebook] 3. Open source a high-quality LLM to incubate an ecosystem capable of producing fun NPCs to make their VR platform lively. [Llama]

4

2 Comments

More opportunities

13

Learn how to build an autonomous research assistant using LangGraph with a team of specialized AI agents #llms #chatgpt4 #gemini #aiml #llmops https://lnkd.in/d6TUbZNj

2

LLMs can take gigabytes of memory to store, which limits what can be run on consumer hardware. But quantization can dramatically compress models, making a wider selection of models available to developers. You can often reduce model size by 4x or more while maintaining reasonable performance. In our new short course Quantization Fundamentals taught by Hugging Face's Younes Belkada and Marc Sun, you'll: - Learn how to quantize nearly any open source model - Use int8 and bfloat16 (Brain float 16) data types to load and run LLMs using PyTorch and the Hugging Face Transformers library - Dive into the technical details of linear quantization to map 32-bit floats to 8-bit integers As models get bigger and bigger, quantization becomes more important for making models practical and accessible. Please check out the course here: https://lnkd.in/g66yNW8W

10,606

144 Comments

Happy to share that we’ve had two papers accepted at ACL 2024: 1. FOFO: A Benchmark to Evaluate LLMs’ Format-Following Capability. This paper introduces FoFo, an innovative benchmark designed to assess large language models’ (LLMs) ability to follow complex, domain-specific formats. This crucial yet underexplored capability is vital for their deployment as AI agents. Link: https://lnkd.in/gsjy6_nx 2. LLM-REDIAL: A Large-Scale Dataset for Conversational Recommender Systems Created from User Behaviors with LLMs. This paper introduces the LLM-REDIAL dataset to facilitate research in CRS. It is constructed by leveraging large language models (LLMs) to generate high-quality dialogues. Link: https://lnkd.in/g_Zgqhcu Congratulations to all the authors: Chen Xing, Jiangshu Du, Xinyi Yang, Yihao Feng, Ran Xu, Wenpeng Yin, Caiming Xiong

74

In this episode, we discuss SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales by Tianyang Xu, Shujin Wu, Shizhe Diao, Xiaoze Liu, Xingyao Wang, Yangyi Chen, Jing Gao. The paper introduces SaySelf, a framework for training large language models (LLMs) to produce accurate, fine-grained confidence estimates and self-reflective rationales explaining their uncertainties. This is achieved by analyzing inconsistencies in multiple reasoning chains, summarizing uncertainties in natural language, and applying supervised fine-tuning alongside reinforcement learning to calibrate confidence levels. Experimental results show that SaySelf effectively reduces confidence calibration errors and maintains task performance, enhancing LLMs' reliability by mitigating overconfidence in erroneous outputs.

1

Just Arrived in Vienna for #ICML2024, I will be presenting our work on improving GNNs' generalization ability with ensemble learning (https://lnkd.in/gimXe5fS) at Oral 5A (Thu 25 Jul 10:45 a.m. — 11 a.m.) and Poster 2407 (Thu 25 Jul 11:30 a.m. — 1 p.m.). Paper title: From Coarse to Fine: Enable Comprehensive Graph Self-supervised Learning with Multi-granular Semantic Ensemble. A heartfelt thank you to all my collaborators for their incredible support; this work would not have been possible without them. I am also looking forward to reconnecting with old friends and making new ones. If you are attending ICML or are interested in my research, please join me during my presentation session. Additionally, please feel free to reach out to exchange ideas if you are interested in topics such as (1) improving graph model generalization across different datasets, and (2) foundational models for time series forecasting. See you in Vienna! 🇦🇹 #ICML2024 #MachineLearning #ArtificialIntelligence

14

Sparse Autoencoders (SAEs) are one of the most promising and popular methods for LLM interpretability and explainability. I wrote a simple, intuitive introduction to SAEs, complete with diagrams and reference PyTorch code. Among other things, I cover why we use SAEs, how they are used for interventions, and challenges with using SAEs. https://lnkd.in/gzrvi6Kg

14

Super glad to have contributed a small part to the updated Gemini 1.5 report in measuring in context planning capabilities https://t.co/Pltp92WcNo with Huaixiu Steven Zheng Swaroop Mishra Hugh Zhang Xinyun Chen and Azade Nova

122

3 Comments