Tool Use on Tool use
68.5Avg@16SDPO (on-policy)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SDPO (on-policy)Backbone=Qwen3-8B, Training Time=5h2026.01 | 68.5 | — | — | — | |
| GRPOBackbone=Qwen3-8B, Training Time=5h2026.01 | 68.1 | — | — | — | |
| GRPO (on-policy)Backbone=Qwen3-8B, Training Time=5h2026.01 | 68.1 | — | — | — | |
| GRPOBackbone=Olmo3-7B-Instruct, Training Time=5h2026.01 | 65 | — | — | — | |
| SDPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=5h2026.01 | 62.5 | — | — | — | |
| GRPOBackbone=Qwen3-8B, Training Time=1h2026.01 | 61.7 | — | — | — | |
| GRPO (on-policy)Backbone=Qwen3-8B, Training Time=1h2026.01 | 61.7 | — | — | — | |
| GRPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=5h2026.01 | 61.3 | — | — | — | |
| Qwen3-8BTraining Time=0h2026.01 | 57.5 | — | — | — | |
| SDPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=1h2026.01 | 57.3 | — | — | — | |
| SDPO (on-policy)Backbone=Qwen3-8B, Training Time=1h2026.01 | 56.4 | — | — | — | |
| GRPOBackbone=Olmo3-7B-Instruct, Training Time=1h2026.01 | 56.4 | — | — | — | |
| GRPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=1h2026.01 | 56 | — | — | — | |
| Olmo3-7B-InstructTraining Time=0h2026.01 | 39.3 | — | — | — | |
| GRPOModel Scale=8B2026.04 | — | 64.3 | 68.5 | 69 | |
| GRPOModel Scale=4B2026.04 | — | 62.9 | 62.9 | 62.9 | |
| Qwen3-4BModel Scale=4B2026.04 | — | 58.8 | — | — | |
| Qwen3-8BModel Scale=8B2026.04 | — | 57.9 | — | — | |
| SDPOModel Scale=8B2026.04 | — | 64.8 | 65.7 | 65.7 | |
| SDPOModel Scale=4B2026.04 | — | 61.1 | 61.1 | 61.1 | |
| SRPOModel Scale=8B2026.04 | — | 65.2 | 71.2 | 71.2 | |
| SRPOModel Scale=4B2026.04 | — | 61.4 | 63.1 | 67 |