Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -25,3 +25,40 @@ ToolMesh is a large open-source tool-use dataset with reasoning traces, designed
|
|
| 25 |
* Hybrid Training with Augmented Open-Source Data
|
| 26 |
* In addition to the synthesized trajectories, we also incorporated a large amount of processed open-source data, including XLAM, When2Call, Glaive, ToolACE, Button, and API_Gen. The processing steps involved quality filtering and response reconstruction. Experimental results demonstrate that both our synthesized data and the post-processed open-source data significantly contribute to performance improvements.
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
* Hybrid Training with Augmented Open-Source Data
|
| 26 |
* In addition to the synthesized trajectories, we also incorporated a large amount of processed open-source data, including XLAM, When2Call, Glaive, ToolACE, Button, and API_Gen. The processing steps involved quality filtering and response reconstruction. Experimental results demonstrate that both our synthesized data and the post-processed open-source data significantly contribute to performance improvements.
|
| 27 |
|
| 28 |
+
# Performance
|
| 29 |
+
|
| 30 |
+
* Overall
|
| 31 |
+
|
| 32 |
+
\begin{table}[]
|
| 33 |
+
\begin{tabular}{llllllll}
|
| 34 |
+
& Tau1-airline & Tau1-retail & Tau2-airline & Tau2-retail & Tau2-telecom & BFCL-v4 & BFCL-v4-agentic \\
|
| 35 |
+
& Tau1-airline & Tau1-retail & Tau2-airline & Tau2-retail & Tau2-telecom & BFCL-v4 & BFCL-v4-agentic \\
|
| 36 |
+
qwen3-8b(FC) & 36 & 35.65 & 32 & 43.9 & 28.1 & 42.21 & 14.35 \\
|
| 37 |
+
qwen3-8b(FC) w ToolMind & & & 48 & 59.6 & 31.6 & {\ul \textbf{46.92}} & {\ul 20.97} \\
|
| 38 |
+
& & & & & & & \\
|
| 39 |
+
qwen3-14b(FC) & & & 36 & 52.6 & 33.3 & 45.14 & 16.90 \\
|
| 40 |
+
qwen3-14b(FC) w ToolMind & & & 56 & 59.6 & 31.6 & 50.54 & 26.67 \\
|
| 41 |
+
& & & & & & & \\
|
| 42 |
+
qwen3-32b(FC) & 38 & 59.13 & 32.0 & 59.6 & 31.6 & 46.55 & 23.36 \\
|
| 43 |
+
qwen3-32b(FC) w ToolMind & & & & & & &
|
| 44 |
+
\end{tabular}
|
| 45 |
+
\end{table}
|
| 46 |
+
|
| 47 |
+
* Ablation
|
| 48 |
+
|
| 49 |
+
\begin{table}[]
|
| 50 |
+
\begin{tabular}{llllllll}
|
| 51 |
+
& Tau1-airline & Tau1-retail & Tau2-airline & Tau2-retail & Tau2-telecom & BFCL-v4 & BFCL-v4-agentic \\
|
| 52 |
+
qwen3-8b(FC) & 36 & 35.65 & 32 & 43.9 & 28.1 & 42.21 & 14.35 \\
|
| 53 |
+
qwen3-8b(FC) w Augmented Open-Source Data (26w) & & & 44 & 57.9 & 24.6 & 45.88 & 20.22 \\
|
| 54 |
+
qwen3-8b(FC) w Synthesized Data(16w) & 42 & 42.61 & 42 & 43.0 & 31.6 & {\ul 46.87} & {\ul \textbf{24.37}} \\
|
| 55 |
+
qwen3-8b(FC) w ToolMind (42w) & & & 48 & 59.6 & 31.6 & {\ul \textbf{46.92}} & {\ul 20.97}
|
| 56 |
+
\end{tabular}
|
| 57 |
+
\end{table}
|
| 58 |
+
|
| 59 |
+
# Dataset Schema
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
# Other Information
|
| 64 |
+
If you find the data or code useful, please cite:
|