DaYin commited on
Commit
8c3e2cd
1 Parent(s): 73d6b4f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +55 -0
README.md CHANGED
@@ -1,3 +1,58 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
+ datasets:
4
+ - ai2lumos/lumos_multimodal_ground_iterative
5
+ language:
6
+ - en
7
+ tags:
8
+ - language-agent
9
+ - question-answering
10
+ - reasoning
11
+ - grounding
12
  ---
13
+
14
+ # 🪄 Agent Lumos: Unified and Modular Training for Open-Source Language Agents
15
+ <p align="center">
16
+ 🌐<a href="https://allenai.github.io/lumos">[Website]</a> &nbsp;
17
+ 📝<a href="https://arxiv.org/abs/2311.05657">[Paper]</a> &nbsp;
18
+ 🤗<a href="https://huggingface.co/datasets?sort=trending&search=ai2lumos">[Data]</a> &nbsp;
19
+ 🤗<a href="https://huggingface.co/models?sort=trending&search=ai2lumos">[Model]</a> &nbsp;
20
+ 🤗<a href="https://huggingface.co/spaces/ai2lumos/lumos_data_demo">[Demo]</a> &nbsp;
21
+ </p>
22
+
23
+ We introduce 🪄**Lumos**, Language Agents with **Unified** Formats, **Modular** Design, and **Open-Source** LLMs. **Lumos** unifies a suite of complex interactive tasks and achieves competitive performance with GPT-4/3.5-based and larger open-source agents.
24
+
25
+ **Lumos** has following features:
26
+ * 🧩 **Modular Architecture**:
27
+ - 🧩 **Lumos** consists of planning, grounding, and execution modules built based on LLAMA-2-7B/13B and off-the-shelf APIs.
28
+ - 🤗 **Lumos** utilizes a unified data format that encompasses multiple task types, thereby enabling the developed agent framework to conveniently support a range of interactive tasks.
29
+ * 🌍 **Diverse Training Data**:
30
+ - 🌍 **Lumos** is trained with ~56K diverse high-quality subgoal/action annotations from ground-truth reasoning steps in existing benchmarks with GPT-4.
31
+ - ⚒️ **Lumos** data can be instrumental for future research in developing open-source agents for complex interactive tasks.
32
+ * 🚀 **Competitive Performance**:
33
+ - 🚀 **Lumos** is comparable or even beats **GPT-series** agents on web/complex QA tasks Mind2Web and HotpotQA, and **larger open agents** on math and multimodal tasks.
34
+ - 🚀 **Lumos** exceeds contemporaneous agents that have been **fine-tuned** with in-domain HotpotQA, Mind2Web and ScienceQA annotations, such as **FiReAct**, **AgentLM**, and **AutoAct**.
35
+ - 🚀 **Lumos** performs better than open agent baseline formulations including **chain-of-thoughts** and **integrated** training.
36
+ - 🚀 **Lumos** surpasses larger open LLM agents and domain-specific agents on unseen tasks, WebShop and InterCode_SQL.
37
+
38
+ ## Model Overview
39
+ `lumos_multimodal_ground_iterative` is a **grounding** module checkpoint finetuned on **multimodal** task in **Lumos-Iterative (Lumos-I)** formulation.
40
+
41
+ The training annotation is shown below:
42
+
43
+ | Training Data | Number |
44
+ |---|---|
45
+ |[`lumos_multimodal_ground_iterative`](https://huggingface.co/datasets/ai2lumos/lumos_multimodal_ground_iterative)|19541|
46
+
47
+
48
+ ## Citation
49
+
50
+ If you find this work is relevant with your research, please feel free to cite our work!
51
+ ```
52
+ @article{yin2023lumos,
53
+ title={Agent Lumos: Unified and Modular Training for Open-Source Language Agents},
54
+ author={Yin, Da and Brahman, Faeze and Ravichander, Abhilasha and Chandu, Khyathi and Chang, Kai-Wei and Choi, Yejin and Lin, Bill Yuchen},
55
+ journal={arXiv preprint arXiv:2311.05657},
56
+ year={2023}
57
+ }
58
+ ```