Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
```
|
| 2 |
+
test_stage:
|
| 3 |
+
obcq_modifiers:
|
| 4 |
+
SparseGPTModifier:
|
| 5 |
+
sparsity: 0.5
|
| 6 |
+
block_size: 128
|
| 7 |
+
sequential_update: false
|
| 8 |
+
quantize:
|
| 9 |
+
QuantizationModifier:
|
| 10 |
+
ignore:
|
| 11 |
+
- LlamaRotaryEmbedding
|
| 12 |
+
- LlamaRMSNorm
|
| 13 |
+
- SiLUActivation
|
| 14 |
+
- model.layers.21.mlp.down_proj
|
| 15 |
+
- model.layers.7.mlp.down_proj
|
| 16 |
+
- model.layers.2.mlp.down_proj
|
| 17 |
+
- model.layers.20.mlp.down_proj
|
| 18 |
+
- model.layers.19.mlp.down_proj
|
| 19 |
+
post_oneshot_calibration: false
|
| 20 |
+
scheme_overrides:
|
| 21 |
+
Embedding:
|
| 22 |
+
input_activations: null
|
| 23 |
+
weights:
|
| 24 |
+
num_bits: 8
|
| 25 |
+
symmetric: false
|
| 26 |
+
percdamp: 0.01
|
| 27 |
+
prunen: 0
|
| 28 |
+
prunem: 0
|
| 29 |
+
targets:
|
| 30 |
+
- model.layers.0
|
| 31 |
+
- model.layers.1
|
| 32 |
+
- model.layers.2
|
| 33 |
+
- model.layers.3
|
| 34 |
+
- model.layers.4
|
| 35 |
+
- model.layers.5
|
| 36 |
+
- model.layers.6
|
| 37 |
+
- model.layers.7
|
| 38 |
+
- model.layers.8
|
| 39 |
+
- model.layers.9
|
| 40 |
+
- model.layers.10
|
| 41 |
+
- model.layers.11
|
| 42 |
+
- model.layers.12
|
| 43 |
+
- model.layers.13
|
| 44 |
+
- model.layers.14
|
| 45 |
+
- model.layers.15
|
| 46 |
+
- model.layers.16
|
| 47 |
+
- model.layers.17
|
| 48 |
+
- model.layers.18
|
| 49 |
+
- model.layers.19
|
| 50 |
+
- model.layers.20
|
| 51 |
+
- model.layers.21
|
| 52 |
+
target_ids:
|
| 53 |
+
- attention_mask
|
| 54 |
+
- position_ids
|
| 55 |
+
```
|