Commit
•
e8a7b31
1
Parent(s):
dfc9eed
Upload config.json with huggingface_hub
Browse files- config.json +19 -10
config.json
CHANGED
|
@@ -9,7 +9,12 @@
|
|
| 9 |
"compression_config": {
|
| 10 |
"config_groups": {
|
| 11 |
"group_0": {
|
| 12 |
-
"input_activations":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
"block_structure": null,
|
| 14 |
"dynamic": false,
|
| 15 |
"group_size": null,
|
|
@@ -19,22 +24,26 @@
|
|
| 19 |
"strategy": "channel",
|
| 20 |
"symmetric": true,
|
| 21 |
"type": "int"
|
| 22 |
-
}
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
"weights": {
|
| 28 |
"block_structure": null,
|
| 29 |
"dynamic": false,
|
| 30 |
"group_size": null,
|
| 31 |
"num_bits": 8,
|
| 32 |
"observer": "minmax",
|
| 33 |
"observer_kwargs": {},
|
| 34 |
-
"strategy": "
|
| 35 |
"symmetric": true,
|
| 36 |
-
"type": "
|
| 37 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
}
|
| 39 |
},
|
| 40 |
"format": "pack-quantized",
|
|
|
|
| 9 |
"compression_config": {
|
| 10 |
"config_groups": {
|
| 11 |
"group_0": {
|
| 12 |
+
"input_activations": null,
|
| 13 |
+
"output_activations": null,
|
| 14 |
+
"targets": [
|
| 15 |
+
"Linear"
|
| 16 |
+
],
|
| 17 |
+
"weights": {
|
| 18 |
"block_structure": null,
|
| 19 |
"dynamic": false,
|
| 20 |
"group_size": null,
|
|
|
|
| 24 |
"strategy": "channel",
|
| 25 |
"symmetric": true,
|
| 26 |
"type": "int"
|
| 27 |
+
}
|
| 28 |
+
},
|
| 29 |
+
"kv_cache": {
|
| 30 |
+
"input_activations": null,
|
| 31 |
+
"output_activations": {
|
|
|
|
| 32 |
"block_structure": null,
|
| 33 |
"dynamic": false,
|
| 34 |
"group_size": null,
|
| 35 |
"num_bits": 8,
|
| 36 |
"observer": "minmax",
|
| 37 |
"observer_kwargs": {},
|
| 38 |
+
"strategy": "tensor",
|
| 39 |
"symmetric": true,
|
| 40 |
+
"type": "float"
|
| 41 |
+
},
|
| 42 |
+
"targets": [
|
| 43 |
+
"re:.*k_proj",
|
| 44 |
+
"re:.*v_proj"
|
| 45 |
+
],
|
| 46 |
+
"weights": null
|
| 47 |
}
|
| 48 |
},
|
| 49 |
"format": "pack-quantized",
|