Function Calling on BFCL Memory
28.22Task AccuracySGLang FP4
Evaluation Results
| Method | Links | |
|---|---|---|
| SGLang FP4Model=Falcon3-10B-Instruct2026.05 | 28.22 | |
| SGLang BF16Model=Qwen3-32B2026.05 | 25.78 | |
| SGLang BF16Model=Qwen3-14B2026.05 | 25.11 | |
| TriAxialKV MixedModel=Qwen3-32B2026.05 | 25.11 | |
| TriAxialKV INT4Model=Qwen3-14B2026.05 | 24.89 | |
| TriAxialKV INT4Model=Qwen3-32B2026.05 | 24.67 | |
| TriAxialKV MixedModel=Qwen3-14B2026.05 | 24.22 | |
| SGLang BF16Model=Falcon3-10B-Instruct2026.05 | 24 | |
| TriAxialKV MixedModel=Falcon3-10B-Instruct2026.05 | 24 | |
| SGLang BF16Model=Qwen3-235B-A22B-Instruct-25072026.05 | 23.78 | |
| SGLang FP4Model=Qwen3-235B-A22B-Instruct-25072026.05 | 23.56 | |
| KIVIModel=Qwen3-235B-A22B-Instruct-25072026.05 | 23.33 | |
| TriAxialKV INT4Model=Qwen3-235B-A22B-Instruct-25072026.05 | 23.11 | |
| KIVIModel=Falcon3-10B-Instruct2026.05 | 22.89 | |
| TriAxialKV MixedModel=Qwen3-235B-A22B-Instruct-25072026.05 | 22.67 | |
| TriAxialKV INT4Model=Falcon3-10B-Instruct2026.05 | 22 | |
| KIVIModel=Qwen3-32B2026.05 | 21.56 | |
| KIVIModel=Qwen3-14B2026.05 | 20.67 | |
| SGLang FP4Model=Qwen3-32B2026.05 | 20.22 | |
| SGLang FP4Model=Qwen3-14B2026.05 | 18 |