auto_round format

Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>

config.json

@@ -20,11 +20,11 @@
   "quantization_config": {
     "amp": true,
     "autoround_version": "0.4.2.dev",
+    "backend": "auto_round:gptq:exllamav2",
     "batch_size": 8,
     "bits": 4,
-    "damp_percent": 0.01,
     "data_type": "int",
-    "desc_act": false,
+    "dataset": "NeelNanda/pile-10k",
     "enable_minmax_tuning": true,
     "enable_norm_bias_tuning": false,
     "enable_quanted_input": true,
@@ -35,12 +35,11 @@
     "lr": 0.001,
     "minmax_lr": 0.001,
     "nsamples": 512,
-    "quant_method": "gptq",
+    "quant_method": "intel/auto-round",
     "scale_dtype": "torch.float16",
     "seqlen": 2048,
     "sym": true,
-    "to_quant_block_names": "model.layers",
-    "true_sequential": false
+    "to_quant_block_names": "model.layers"
   },
   "rms_norm_eps": 1e-06,
   "rope_scaling": {

quantize_config.json → quantization_config.json

@@ -17,9 +17,8 @@
   "low_gpu_mem_usage": true,
   "to_quant_block_names": "model.layers",
   "enable_norm_bias_tuning": false,
+  "dataset": "NeelNanda/pile-10k",
   "autoround_version": "0.4.2.dev",
-  "quant_method": "gptq",
-  "desc_act": false,
-  "true_sequential": false,
-  "damp_percent": 0.01
+  "quant_method": "intel/auto-round",
+  "backend": "auto_round:gptq:exllamav2"
 }