Add base-model few-shot/generative results: SVD-LLM, SVD-LLM V2, ASVD on 7B and 13B base (24 cells)
Browse files
README.md
CHANGED
|
@@ -134,12 +134,13 @@ Perplexity is **token-level** over fixed non-overlapping 2048-token windows, fol
|
|
| 134 |
| | ASVD | 41.43 | 68.01 | 29.71 | 7.51 |
|
| 135 |
| **LLaMA-2-7B (base)** | *dense FP16* | 45.97 | 78.95 | 64.39 | 13.12 |
|
| 136 |
| | TR-3 (ours) | 39.49 | 71.11 | 39.75 | 6.52 |
|
| 137 |
-
| | SVD-LLM |
|
| 138 |
-
| | SVD-LLM V2 ‡ |
|
| 139 |
-
| | ASVD |
|
| 140 |
| **LLaMA-2-13B (base)** | *dense FP16* | 55.54 | 82.14 | 70.69 | 23.28 |
|
| 141 |
-
| | SVD-LLM |
|
| 142 |
-
| | SVD-LLM V2 ‡ |
|
|
|
|
| 143 |
| **Qwen2.5-3B** | *dense FP16* | 66.42 | 75.26 | 45.78 | 10.31 |
|
| 144 |
| | TR-3 (ours) | 56.02 | 66.39 | 27.61 | 41.32 |
|
| 145 |
| | TT-3 (open chain) | 41.97 | 51.65 | 11.86 | 9.78 |
|
|
|
|
| 134 |
| | ASVD | 41.43 | 68.01 | 29.71 | 7.51 |
|
| 135 |
| **LLaMA-2-7B (base)** | *dense FP16* | 45.97 | 78.95 | 64.39 | 13.12 |
|
| 136 |
| | TR-3 (ours) | 39.49 | 71.11 | 39.75 | 6.52 |
|
| 137 |
+
| | SVD-LLM | 32.74 | 58.36 | 16.43 | 1.59 |
|
| 138 |
+
| | SVD-LLM V2 ‡ | 34.33 | 59.36 | 19.34 | 1.97 |
|
| 139 |
+
| | ASVD | 26.58 | 61.21 | 20.69 | 1.29 |
|
| 140 |
| **LLaMA-2-13B (base)** | *dense FP16* | 55.54 | 82.14 | 70.69 | 23.28 |
|
| 141 |
+
| | SVD-LLM | 42.70 | 63.80 | 21.87 | 4.25 |
|
| 142 |
+
| | SVD-LLM V2 ‡ | 44.67 | 65.00 | 24.49 | 4.93 |
|
| 143 |
+
| | ASVD | 42.44 | 70.15 | 35.50 | 5.23 |
|
| 144 |
| **Qwen2.5-3B** | *dense FP16* | 66.42 | 75.26 | 45.78 | 10.31 |
|
| 145 |
| | TR-3 (ours) | 56.02 | 66.39 | 27.61 | 41.32 |
|
| 146 |
| | TT-3 (open chain) | 41.97 | 51.65 | 11.86 | 9.78 |
|