Junteng commited on
Commit
98ed2f6
·
verified ·
1 Parent(s): 5a07909

Upload file README.md

Browse files
Files changed (1) hide show
  1. README.md +106 -3
README.md CHANGED
@@ -1,3 +1,106 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model:
3
+ - Qwen/Qwen3-8B
4
+ language:
5
+ - en
6
+ library_name: transformers
7
+ license: mit
8
+ tags:
9
+ - LLM
10
+ pipeline_tag: text-generation
11
+ ---
12
+ # 🔍 WebExplorer-8B
13
+
14
+ <!-- [![Model](https://img.shields.io/badge/Model-8B_Parameters-blue)](https://huggingface.co/WebExplorer-8B)
15
+ [![Paper](https://img.shields.io/badge/Paper-arXiv-red)](https://arxiv.org/abs/xxxx.xxxxx) -->
16
+
17
+ A state-of-the-art 8B parameter web agent model designed for complex information-seeking tasks and long-horizon reasoning.
18
+
19
+ ## 🌟 Overview
20
+
21
+ WebExplorer-8B is an advanced web navigation agent trained on **WebExplorer**-QA. The model demonstrates exceptional performance on challenging information-seeking benchmarks while maintaining efficiency with only 8 billion parameters.
22
+
23
+ ## ✨ Key Features
24
+
25
+ - 🌐 **Long-horizon Reasoning**: Supports up to 128K context length and 100 tool calling turns
26
+ - 🛠️ **Tool Utilization**: Masters search and browse functionalities
27
+ - 🏆 **State-of-the-art Performance**: Achieves best-in-class results among models under 10B parameters
28
+
29
+ ## 🏗️ Model Architecture
30
+
31
+ Built on Qwen3-8B base model and trained through a two-phase approach:
32
+
33
+ 1. **Supervised Fine-tuning (SFT)**: Cold-start initialization with high-quality trajectories
34
+ 2. **Reinforcement Learning (RL)**: Enhanced using GRPO algorithm with progressive context expansion
35
+
36
+ ## 📊 Performance
37
+
38
+ WebExplorer-8B achieves state-of-the-art performance across multiple information-seeking benchmarks:
39
+
40
+ | Benchmark | Score |
41
+ |-----------|-------|
42
+ | BrowseComp-en | **15.7** |
43
+ | BrowseComp-zh | **32.0** |
44
+ | GAIA | **50.0** |
45
+ | WebWalkerQA | **62.7** |
46
+ | FRAMES | **75.7** |
47
+ | XBench-DeepSearch | **53.7** |
48
+ | HLE | **17.3** |
49
+
50
+
51
+
52
+ ## 🛠️ Tool Schema
53
+
54
+ WebExplorer-8B supports two tools for web interaction:
55
+
56
+ ### 1. Browse Tool
57
+
58
+ ```json
59
+ {
60
+ "name": "browse",
61
+ "type": "function",
62
+ "description": "Extract specific information from a webpage",
63
+ "parameters": {
64
+ "type": "object",
65
+ "properties": {
66
+ "url": {
67
+ "type": "string",
68
+ "description": "Target URL to browse. The webpage content will be processed by the LLM for information extraction."
69
+ },
70
+ "query": {
71
+ "type": "string",
72
+ "description": "Specific query about the webpage content. The LLM will analyze the content to answer this query."
73
+ }
74
+ },
75
+ "required": ["url", "query"]
76
+ }
77
+ }
78
+ ```
79
+
80
+
81
+
82
+ ### 2. Search Tool
83
+
84
+ ```json
85
+ {
86
+ "name": "search",
87
+ "type": "function",
88
+ "description": "Perform web search queries",
89
+ "parameters": {
90
+ "type": "object",
91
+ "properties": {
92
+ "queries": {
93
+ "type": "array",
94
+ "items": {
95
+ "type": "string"
96
+ },
97
+ "description": "List of search queries. Returns search results containing title, URL, and snippet for each query."
98
+ }
99
+ },
100
+ "required": ["queries"]
101
+ }
102
+ }
103
+ ```
104
+
105
+
106
+