Voozh

Shibo Hao

795 posts

Shibo Hao

@Ber18791531

@thinkymachines | Prev. @UCSanDiego @PKU1898 @AIatMeta @Bloomberg Ph.D. Fellow 🥥Coconut, 🎶Reasoning-via-planning, 🛠️ToolkenGPT

San Diego, CA

Joined March 2020

Pinned
👁 user avatar
Shibo Hao
@Ber18791531
Dec 10, 2024
Is human language the most effective space for LLMs to reason? 🤔 Introducing Coconut 🥥 (Chain of Continuous Thought): our exploration toward LLM 𝐥𝐚𝐭𝐞𝐧𝐭 reasoning. - Encodes multiple reasoning paths simultaneously🌐 - Shows advantages in planning-intense tasks 🚀
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Jul 29, 2023
📣 Introducing LLM Reasoners (github.com/Ber666/llm-rea…), a library for advanced reasoning with LLMs. Simply define a reward function and an optional world model, and let LLM Reasoners take care of the rest, including Reasoning Algorithms, Visualization, LLM calling, etc.
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Dec 23, 2024
Introducing OREO: an offline RL method to improve LLM multi-step reasoning 📈 52.5% on MATH with 1.5B LLM (w/o using augmented problem set) ✅ No paired data needed & Better credit assignment than DPO 🚀 Free test-time boosts with value-guided tree search
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
May 24, 2023
🤖LLMs (autoregressive transformers) are not designed to reason in non-linear structures like humans... 🤔Instead of using it as a straight-minded reasoner, what about regarding it as a world model🌎and planning with it? Check out our new work: Reasoning via Planning (RAP) 🧵
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
May 22, 2023
🤔ChatGPT plug-in is impressive, but is "learning tools in the context" the ultimate solution? Check out our ToolkenGPT (arxiv.org/abs/2305.11554), which can handle massive tools and understand them better with new "toolken" embeddings.
👁 user avatar
Shibo Hao
@Ber18791531
Jan 22, 2025
It's exciting to see Kimi-k1.5 uses a similar RL training objective to our (response-level) OREO! The difference is they estimate Z with some rollouts, while we learn a model for that
👁 Image
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Dec 23, 2024
Introducing OREO: an offline RL method to improve LLM multi-step reasoning 📈 52.5% on MATH with 1.5B LLM (w/o using augmented problem set) ✅ No paired data needed & Better credit assignment than DPO 🚀 Free test-time boosts with value-guided tree search
👁 user avatar
Shibo Hao
@Ber18791531
Dec 9, 2024
Super honored to receive the Bloomberg fellowship🎉 Huge thanks to my amazing advisor @ZhitingHu and all collaborators!
👁 user avatar
Tech At Bloomberg
@TechAtBloomberg
Dec 9, 2024
Congratulations to the seventh cohort of @Bloomberg's #DataScience Ph.D. Fellowship! Learn more about each of our three 2024-2025 Fellows and what their research is focusing on: bloom.bg/4iGPF8X #AI #ML #NLProc #LLMs #NeurIPS2024
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Jul 10, 2024
Excited to share that our paper “LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Models” is accepted to @COLM_conf 2024! arxiv.org/abs/2404.05221 #LLMReasoners #COLM #COLM24
👁 user avatar
Maitrix.org
@MaitrixOrg
Apr 23, 2024
Releasing 🔥LLM Reasoners v1.0🔥 🥇Popular library for advanced LLM reasoning - Reasoning-via-Planning (RAP)🎶 - Chain-of-Thoughts (CoT)⛓️ - Tree-of-Thoughts (ToT)🌴 - Grace decoding💄 - Beam search🔎 🥇Enhances #Llama3, GPT4, LLMs on @huggingface llm-reasoners.net
👁 Image
👁 arXiv logo
arxiv.org
LLM Reasoners: New Evaluation, Library, and Analysis of...
Generating accurate step-by-step reasoning is essential for Large Language Models (LLMs) to address complex problems and enhance robustness and interpretability. Despite the flux of research on...
👁 user avatar
Shibo Hao
@Ber18791531
Jun 25, 2024
With the context lengths of LLMs going beyond millions of tokens...📈 How to fully exploit their abilities with a super detailed prompt?🤔 LLM Reasoner now supports PromptAgent🤖: Let it find the best prompt for your task with advanced search methods🤩 shorturl.at/zU0Xk
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Jun 18, 2024
Spent a morning reading it and then found out the numbers they reported are pass@k 😅 github.com/trotsky1997/Ma…
👁 user avatar
Aran Komatsuzaki
@arankomatsuzaki
Jun 12, 2024
Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B Significantly improves success rates across MATH and Olympiad-level benchmarks arxiv.org/abs/2406.07394
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
Nov 18, 2023
Sooo honored to receive the best paper award🥳!! Huge thanks to my amazing advisor @ZhitingHu and collaborators @LtyLeoii22 @zhenwang9102 🙌
👁 user avatar
SoCal NLP Symposium
@socalnlp
Nov 18, 2023
Next, we have @Ber18791531 et. al’s amazing work, that also won an award at #SoCalNLP2023! 👏🏼
👁 Image
👁 user avatar
Shibo Hao
@Ber18791531
May 23, 2024
Check out 𝒫𝒶𝓃𝒹𝑜𝓇𝒶, our new work towards a general 𝘄𝗼𝗿𝗹𝗱 𝗺𝗼𝗱𝗲𝗹 🌎 It predicts future states (videos🎥) given any actions (free text💬) Proud to be part of this amazing team!
👁 user avatar
Maitrix.org
@MaitrixOrg
May 23, 2024
🔥Introducing Pandora 🌏 🪐 a World Model that generates videos of world states with real-time language control 🎥🕹️ Simulate the world across domains in an _interactive_ way! check out more world-model.ai
👁 Image
00:00
👁 user avatar
Shibo Hao
@Ber18791531
Oct 24, 2023
Reasoning-via-Planning (RAP🎶) is accepted to EMNLP 2023! With Llama-2 70B, RAP can even work on Blocksworld problems requiring > 10 steps to solve😯 More results & discussion are supplemented in our updated paper: arxiv.org/abs/2305.14992…
👁 Image
👁 Image
👁 user avatar
Zhiting Hu
👁 xAI
@ZhitingHu
May 24, 2023
🧠Human brain has internal world model for conscious reasoning 🤖Language model reasoning also need world model! Reasoning via Planning (RAP) is a new principled framework for this. It prompts LM as WM & reasons w/ MC TreeSearch Better v ChainOfThought on plan-gen, math, logic
👁 user avatar
Shibo Hao
@Ber18791531
Aug 30, 2023
🥳 Update: LLM Reasoners has enhanced its support for Llama2 by integrating HF🤗, bnb, exllama, etc. Wanna try RAP🎶 or ToT🌴, but don’t want to give exponential money💸 to OpenAI? Now in LLM Reasoners, you can explore the reasoning ability of Llama2-70b with just 40 GB VRAM😺
👁 user avatar
Shibo Hao
@Ber18791531
Jul 29, 2023
📣 Introducing LLM Reasoners (github.com/Ber666/llm-rea…), a library for advanced reasoning with LLMs. Simply define a reward function and an optional world model, and let LLM Reasoners take care of the rest, including Reasoning Algorithms, Visualization, LLM calling, etc.
👁 Image

URL: https://x.com/Ber18791531

⇱ Shibo Hao (@Ber18791531) / X