Add quantization configuration metadata
Browse files- quantization_config.json +20 -0
quantization_config.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"quant_method": "awq",
|
| 3 |
+
"bits": 4,
|
| 4 |
+
"group_size": 128,
|
| 5 |
+
"dataset": "lmms-lab/flickr30k",
|
| 6 |
+
"calibration_samples": 256,
|
| 7 |
+
"model_seqlen": 2048,
|
| 8 |
+
"block_name_to_quantize": "LlamaDecoderLayer",
|
| 9 |
+
"module_name_preceding_first_block": [
|
| 10 |
+
"model.vision_model",
|
| 11 |
+
"model.connector"
|
| 12 |
+
],
|
| 13 |
+
"quantization_tool": "llm-compressor",
|
| 14 |
+
"preserved_components": [
|
| 15 |
+
"vision_model",
|
| 16 |
+
"vision_tower",
|
| 17 |
+
"connector",
|
| 18 |
+
"lm_head"
|
| 19 |
+
]
|
| 20 |
+
}
|