support for autoround type
Browse filesSigned-off-by: n1ck-guo <heng.guo@intel.com>
src/leaderboard/read_evals.py
CHANGED
@@ -2,6 +2,7 @@ import glob
|
|
2 |
import json
|
3 |
import math
|
4 |
import os
|
|
|
5 |
from dataclasses import dataclass
|
6 |
|
7 |
import dateutil
|
@@ -145,6 +146,7 @@ class EvalResult:
|
|
145 |
except Exception as e:
|
146 |
self.status = "Failed"
|
147 |
print(f"Could not find request file for {self.org}/{self.model}")
|
|
|
148 |
|
149 |
def update_with_dynamic_file_dict(self, file_dict):
|
150 |
self.license = file_dict.get("license", "?")
|
|
|
2 |
import json
|
3 |
import math
|
4 |
import os
|
5 |
+
import traceback
|
6 |
from dataclasses import dataclass
|
7 |
|
8 |
import dateutil
|
|
|
146 |
except Exception as e:
|
147 |
self.status = "Failed"
|
148 |
print(f"Could not find request file for {self.org}/{self.model}")
|
149 |
+
print(traceback.format_exc())
|
150 |
|
151 |
def update_with_dynamic_file_dict(self, file_dict):
|
152 |
self.license = file_dict.get("license", "?")
|
src/submission/submit.py
CHANGED
@@ -148,6 +148,10 @@ def add_new_eval(
|
|
148 |
in_group_size = quantization_config.get('in_group_size')
|
149 |
bits = int(nbits_per_codebook * num_codebooks / in_group_size)
|
150 |
precision = f"{bits}bit"
|
|
|
|
|
|
|
|
|
151 |
|
152 |
if precision == "4bit":
|
153 |
weight_dtype = "int4"
|
|
|
148 |
in_group_size = quantization_config.get('in_group_size')
|
149 |
bits = int(nbits_per_codebook * num_codebooks / in_group_size)
|
150 |
precision = f"{bits}bit"
|
151 |
+
if "auto-round" in quant_method:
|
152 |
+
hardware = "gpu"
|
153 |
+
quant_type = "AutoRound"
|
154 |
+
precision = f"{quantization_config.get('bits', '4bit')}bit"
|
155 |
|
156 |
if precision == "4bit":
|
157 |
weight_dtype = "int4"
|