5 18 131

Sambit Mukherjee

sadhaklal

AI & ML interests

None yet

Recent Activity

liked a model 3 days ago

Soul-AILab/SoulX-Podcast-1.7B

liked a model 11 days ago

opendatalab/MinerU2.5-2509-1.2B

liked a model 3 months ago

google/gemma-3-12b-it

View all activity

Organizations

liked a model 3 days ago

Soul-AILab/SoulX-Podcast-1.7B

Text-to-Speech • 2B • Updated 1 day ago • 888 • 142

liked a model 11 days ago

opendatalab/MinerU2.5-2509-1.2B

Image-Text-to-Text • 1B • Updated Sep 29 • 100k • 263

liked a model 3 months ago

google/gemma-3-12b-it

Image-Text-to-Text • 12B • Updated Mar 21 • 1.44M • • 558

liked 3 datasets 4 months ago

liked a model 5 months ago

docling-project/SmolDocling-256M-preview

Image-Text-to-Text • 0.3B • Updated Sep 17 • 354k • 1.59k

liked a Space 5 months ago

255

SmolDocling

🦆

Convert images and text to structured documents

New activity in discord-community/LevelBot 7 months ago

Cannot verify my account in discord

➕ 1

#37 opened 7 months ago by

pdjota

commented on Tool Use, Unified 7 months ago

Excellent article. Very clearly written.

I have one question though. It seems that that model replies with either (i) a text response or (ii) a tool call. However, in the original ReAct paper, there is a "Thought" -> "Action" -> "Observation" cycle. In other words, in response to the user's query, the model first outputs a "Thought" followed by an "Action". How do I implement this (i.e., make the model "think" before performing a tool call)?

The following are the original ReAct prompts for HotpotQA (from the official ReAct GitHub repo): https://raw.githubusercontent.com/ysymyth/ReAct/refs/heads/master/prompts/prompts_naive.json

If you examine these prompts, you'll notice that the "thoughts" come before the "actions".

upvoted an article 7 months ago

Article

Tool Use, Unified

Aug 12, 2024

• 116

posted an update 7 months ago

Post

1979

What happens when you combine the Chain of Thought (CoT) reasoning capabilities of LLMs with a heuristic-guided tree search algorithm? In the Tree of Thoughts (ToT) paper, the authors (Yao et al.) have coupled GPT-4 with tree search algorithms to attack a few tasks on which left-to-right CoT struggles. And the results are impressive. For example, on the "Game of 24" task, while GPT-4 with CoT prompting only managed to solve 4% of tasks, ToT achieved a success rate of 74%.

I've written a blog post that makes the ToT paper easy to understand and implement by taking you through all the details in a step-by-step manner: https://huggingface.co/blog/sadhaklal/tree-of-thoughts

If you are interested in the topics of algorithmic AI, tree search, reasoning, planning, or "System 2" thinking, then you may find this blog post useful.