upload auto_round format
Browse filesSigned-off-by: n1ck-guo <heng.guo@intel.com>
- config.json +4 -5
config.json
CHANGED
@@ -26,11 +26,11 @@
|
|
26 |
"quantization_config": {
|
27 |
"amp": true,
|
28 |
"autoround_version": "0.4.2.dev",
|
|
|
29 |
"batch_size": 8,
|
30 |
"bits": 4,
|
31 |
-
"damp_percent": 0.01,
|
32 |
"data_type": "int",
|
33 |
-
"
|
34 |
"enable_minmax_tuning": true,
|
35 |
"enable_norm_bias_tuning": false,
|
36 |
"enable_quanted_input": true,
|
@@ -41,12 +41,11 @@
|
|
41 |
"lr": 0.001,
|
42 |
"minmax_lr": 0.001,
|
43 |
"nsamples": 512,
|
44 |
-
"quant_method": "
|
45 |
"scale_dtype": "torch.float16",
|
46 |
"seqlen": 2048,
|
47 |
"sym": true,
|
48 |
-
"to_quant_block_names": null
|
49 |
-
"true_sequential": false
|
50 |
},
|
51 |
"rms_norm_eps": 1e-05,
|
52 |
"rope_scaling": {
|
|
|
26 |
"quantization_config": {
|
27 |
"amp": true,
|
28 |
"autoround_version": "0.4.2.dev",
|
29 |
+
"backend": "auto_round:gptq:exllamav2",
|
30 |
"batch_size": 8,
|
31 |
"bits": 4,
|
|
|
32 |
"data_type": "int",
|
33 |
+
"dataset": "NeelNanda/pile-10k",
|
34 |
"enable_minmax_tuning": true,
|
35 |
"enable_norm_bias_tuning": false,
|
36 |
"enable_quanted_input": true,
|
|
|
41 |
"lr": 0.001,
|
42 |
"minmax_lr": 0.001,
|
43 |
"nsamples": 512,
|
44 |
+
"quant_method": "intel/auto-round",
|
45 |
"scale_dtype": "torch.float16",
|
46 |
"seqlen": 2048,
|
47 |
"sym": true,
|
48 |
+
"to_quant_block_names": null
|
|
|
49 |
},
|
50 |
"rms_norm_eps": 1e-05,
|
51 |
"rope_scaling": {
|