Multimodal Language Understanding on Qwen3 1.7B
58.65Average PerformanceBF16
Evaluation Results
| Method | Links | |
|---|---|---|
| BF16Bit-width (W-A-KV)=W16-A16-KV162026.04 | 58.65 | |
| EDGERAZORBit-width (W-A-KV)=W4-A8-KV82026.04 | 58.57 | |
| FlatQuantBit-width (W-A-KV)=W4-A8-KV82026.04 | 57.9 | |
| SpinQuantBit-width (W-A-KV)=W4-A8-KV82026.04 | 56.22 | |
| LQERBit-width (W-A-KV)=W4-A8-KV82026.04 | 55.28 | |
| EDGERAZORBit-width (W-A-KV)=W3-A8-KV82026.04 | 53 | |
| FlatQuantBit-width (W-A-KV)=W3-A8-KV82026.04 | 49.16 | |
| SpinQuantBit-width (W-A-KV)=W3-A8-KV82026.04 | 47.51 | |
| EDGERAZORBit-width (W-A-KV)=W2-A8-KV82026.04 | 47.03 | |
| LQERBit-width (W-A-KV)=W3-A8-KV82026.04 | 46.78 | |
| EDGERAZORBit-width (W-A-KV)=W1.58-A8-KV82026.04 | 43.91 | |
| OmniQuantBit-width (W-A-KV)=W4-A8-KV82026.04 | 43.7 | |
| ABQ-LLMBit-width (W-A-KV)=W4-A8-KV82026.04 | 43.29 | |
| OmniQuantBit-width (W-A-KV)=W3-A8-KV82026.04 | 40.71 | |
| ABQ-LLMBit-width (W-A-KV)=W3-A8-KV82026.04 | 37.82 | |
| OmniQuantBit-width (W-A-KV)=W2-A8-KV82026.04 | 32.56 | |
| ABQ-LLMBit-width (W-A-KV)=W2-A8-KV82026.04 | 31.07 | |
| LQERBit-width (W-A-KV)=W2-A8-KV82026.04 | 30.63 | |
| QuaRotBit-width (W-A-KV)=W2-A8-KV82026.04 | 30.39 | |
| QuaRotBit-width (W-A-KV)=W3-A8-KV82026.04 | 30.17 | |
| QuaRotBit-width (W-A-KV)=W4-A8-KV82026.04 | 30.16 | |
| QoQBit-width (W-A-KV)=W4-A8-KV82026.04 | 29.91 | |
| FlatQuantBit-width (W-A-KV)=W2-A8-KV82026.04 | 29.87 | |
| SpinQuantBit-width (W-A-KV)=W2-A8-KV82026.04 | 29.62 |