Our new work on continuous chain of thought.
10.12.2024 16:51
👍 4
🔁 0
💬 0
📌 0
Analysis: AD picks high temp for creative & low for fact-seeking prompts, automatically via training.
Our methods AD & Latent Pref Optimization are general & can be applied to train other hyperparams or latent features.
Excited how people could *adapt* this research!
🧵4/4
22.11.2024 13:06
👍 2
🔁 0
💬 0
📌 0
We train on a mix of tasks:
GSM8K - requires factuality (low temp)
Stories - requires creativity (high temp)
UltraFeedback - general instruction following, requires mix
Results: Adaptive Decoding outperforms any fixed temperature, automatically choosing via the AD layer.
🧵3/4
22.11.2024 13:06
👍 2
🔁 0
💬 2
📌 0
Recipe 👩🍳:
Adaptive Decoder (AD) Layer:
- Assigns probability to each hyperparam choice (decoding temp) given hidden state. Given temp, sample a token.
Training (Latent PO):
- Train AD by sampling params+tokens & use reward model on rejected hyperparam preference pairs
🧵2/4
22.11.2024 13:06
👍 1
🔁 0
💬 1
📌 0
🚨 Adaptive Decoding via Latent Preference Optimization 🚨
- New layer for Transformer, selects decoding params automatically *per token*
- Learnt via new method Latent Preference Optimization
- Outperforms any fixed temperature decoding, choosing creativity or factuality
arxiv.org/abs/2411.09661
🧵1/4
22.11.2024 13:06
👍 43
🔁 6
💬 2
📌 0
