Add model card and metadata

#1
by nielsr HF Staff - opened
Files changed (1) hide show
  1. README.md +47 -0
README.md ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pipeline_tag: text-generation
3
+ library_name: transformers
4
+ license: apache-2.0
5
+ ---
6
+
7
+ <h1 align="center"> Learning Adaptive Parallel Reasoning <br> with Language Models </h1>
8
+
9
+ <p align="center">
10
+ <a href="https://www.jiayipan.com/" style="text-decoration: none;">Jiayi Pan</a><sup>*</sup>,
11
+ <a href="https://xiuyuli.com/" style="text-decoration: none;">Xiuyu Li</a><sup>*</sup>,
12
+ <a href="https://tonylian.com/" style="text-decoration: none;">Long Lian</a><sup>*</sup>,
13
+ <a href="https://sea-snell.github.io/" style="text-decoration: none;">Charlie Victor Snell</a>,
14
+ <a href="https://yifeizhou02.github.io/" style="text-decoration: none;">Yifei Zhou</a>,<br>
15
+ <a href="https://www.adamyala.org/" style="text-decoration: none;">Adam Yala</a>,
16
+ <a href="https://people.eecs.berkeley.edu/~trevor/" style="text-decoration: none;">Trevor Darrell</a>,
17
+ <a href="https://people.eecs.berkeley.edu/~keutzer/" style="text-decoration: none;">Kurt Keutzer</a>,
18
+ <a href="https://www.alanesuhr.com/" style="text-decoration: none;">Alane Suhr</a>
19
+ </p>
20
+
21
+ <p align="center">
22
+ UC Berkeley and UCSF &nbsp;&nbsp;&nbsp;<sup>*</sup> Equal Contribution
23
+ </p>
24
+
25
+ <p align="center">
26
+ <a href="https://arxiv.org/abs/2504.15466">📃 Paper</a>
27
+
28
+ <a href="https://github.com/Parallel-Reasoning/APR" >💻 Code</a>
29
+ </p>
30
+
31
+ ![APR](./assets/apr.png)
32
+
33
+ **TL;DR**:
34
+ We present Adaptive Parallel Reasoning (APR), a novel framework that enables language models to learn to orchestrate both serialized and parallel computations. APR trains language models to use `spawn()` and `join()` operations through end-to-end supervised training and reinforcement learning, allowing models to dynamically orchestrate their own computational workflows.
35
+ APR efficiently distributes compute, reduces latency, overcomes context window limits, and achieves state‑of‑the‑art performance on complex reasoning tasks (e.g., 83.4% vs. 60.0% accuracy at 4K context on Countdown).
36
+
37
+ ## Citation
38
+ If you find this work useful in your research, please consider citing:
39
+
40
+ ```bibtex
41
+ @article{pan2025learning,
42
+ title = {Learning Adaptive Parallel Reasoning with Language Models},
43
+ author = {Jiayi Pan and Xiuyu Li and Long Lian and Charlie Snell and Yifei Zhou and Adam Yala and Trevor Darrell and Kurt Keutzer and Alane Suhr},
44
+ year = {2025},
45
+ journal = {arXiv preprint arXiv: 2504.15466}
46
+ }
47
+ ```