Article
GGML and llama.cpp join HF to ensure the long-term progress of Local AI
- +4
•
487
Moon Lockwood
moonlock
·
AI & ML interests
art programming music psychology
Recent Activity
upvoted an article 25 days ago
GGML and llama.cpp join HF to ensure the long-term progress of Local AI upvoted a collection 9 months ago
DeepSeek-R1 liked a model over 1 year ago
nisten/Biggie-SmoLlm-0.15B-BaseOrganizations
None yet
upvoted an article 25 days ago
upvoted a collection 9 months ago
liked a
model over 1 year ago
reactedto akhaliq's post with 👍 about 2 years ago
Post
Chain-of-Thought Reasoning Without Prompting
paper page: Chain-of-Thought Reasoning Without Prompting (2402.10200)
In enhancing the reasoning capabilities of large language models (LLMs), prior research primarily focuses on specific prompting techniques such as few-shot or zero-shot chain-of-thought (CoT) prompting. These methods, while effective, often involve manually intensive prompt engineering. Our study takes a novel approach by asking: Can LLMs reason effectively without prompting? Our findings reveal that, intriguingly, CoT reasoning paths can be elicited from pre-trained LLMs by simply altering the decoding process. Rather than conventional greedy decoding, we investigate the top-k alternative tokens, uncovering that CoT paths are frequently inherent in these sequences. This approach not only bypasses the confounders of prompting but also allows us to assess the LLMs' intrinsic reasoning abilities. Moreover, we observe that the presence of a CoT in the decoding path correlates with a higher confidence in the model's decoded answer. This confidence metric effectively differentiates between CoT and non-CoT paths. Extensive empirical studies on various reasoning benchmarks show that the proposed CoT-decoding substantially outperforms the standard greedy decoding.
paper page: Chain-of-Thought Reasoning Without Prompting (2402.10200)
In enhancing the reasoning capabilities of large language models (LLMs), prior research primarily focuses on specific prompting techniques such as few-shot or zero-shot chain-of-thought (CoT) prompting. These methods, while effective, often involve manually intensive prompt engineering. Our study takes a novel approach by asking: Can LLMs reason effectively without prompting? Our findings reveal that, intriguingly, CoT reasoning paths can be elicited from pre-trained LLMs by simply altering the decoding process. Rather than conventional greedy decoding, we investigate the top-k alternative tokens, uncovering that CoT paths are frequently inherent in these sequences. This approach not only bypasses the confounders of prompting but also allows us to assess the LLMs' intrinsic reasoning abilities. Moreover, we observe that the presence of a CoT in the decoding path correlates with a higher confidence in the model's decoded answer. This confidence metric effectively differentiates between CoT and non-CoT paths. Extensive empirical studies on various reasoning benchmarks show that the proposed CoT-decoding substantially outperforms the standard greedy decoding.
liked 5
datasets about 2 years ago
Locutusque/UltraTextbooks
Viewer • Updated • 5.52M • 662 • 198
wikimedia/wikipedia
Viewer • Updated • 61.6M • 87.6k • 1.15k
meta-math/MetaMathQA
Viewer • Updated • 395k • 24.6k • 449
CollectiveCognition/chats-data-2023-09-22
Viewer • Updated • 156 • 31 • 19
teknium/OpenHermes-2.5
Viewer • Updated • 1M • 10.4k • 803
So good! This kind of architecture has been on my mind, thank you very much for sharing. The approach and method used for generating 'pixel tokens' looks great.
Curious about the use of OpenAI API in the demo?
3
#1 opened over 2 years ago
by 
1littlecoder
liked 2
models almost 3 years ago
liked 4
datasets almost 3 years ago
