File tree Expand file tree Collapse file tree 3 files changed +14
-7
lines changed Expand file tree Collapse file tree 3 files changed +14
-7
lines changed Original file line number Diff line number Diff line change @@ -251,6 +251,8 @@ Qwen/Qwen2-0.5B-Instruct:
251
251
accuracy : 30.930
252
252
- quant_algo : FP8
253
253
accuracy : 31.140
254
+ Qwen/Qwen2-1.5B :
255
+ - accuracy : 32.58
254
256
Qwen/Qwen2-7B-Instruct :
255
257
- accuracy : 36.148
256
258
- quant_algo : W8A16
Original file line number Diff line number Diff line change @@ -1137,12 +1137,6 @@ class TestQwen2_0_5BInstruct(CliFlowAccuracyTestHarness):
1137
1137
def test_auto_dtype (self ):
1138
1138
self .run (dtype = 'auto' )
1139
1139
1140
- @pytest .mark .skip (reason = "https://nvbugs/5280195" )
1141
- @pytest .mark .skip_less_device (4 )
1142
- def test_auto_dtype_cp4 (self ):
1143
- "RCCA: https://nvbugs/5170106"
1144
- self .run (dtype = 'auto' , cp_size = 4 )
1145
-
1146
1140
@skip_post_blackwell
1147
1141
def test_weight_only (self ):
1148
1142
self .run (quant_algo = QuantAlgo .W8A16 )
@@ -1154,6 +1148,17 @@ def test_fp8(self):
1154
1148
quant_algo = QuantAlgo .FP8 )
1155
1149
1156
1150
1151
+ class TestQwen2_1_5B (CliFlowAccuracyTestHarness ):
1152
+ MODEL_NAME = "Qwen/Qwen2-1.5B"
1153
+ MODEL_PATH = f"{ llm_models_root ()} /Qwen2-1.5B"
1154
+ EXAMPLE_FOLDER = "models/core/qwen"
1155
+
1156
+ @pytest .mark .skip_less_device (4 )
1157
+ def test_auto_dtype_cp4 (self ):
1158
+ "RCCA: https://nvbugs/5170106"
1159
+ self .run (dtype = 'auto' , cp_size = 4 )
1160
+
1161
+
1157
1162
class TestQwen2_7BInstruct (CliFlowAccuracyTestHarness ):
1158
1163
MODEL_NAME = "Qwen/Qwen2-7B-Instruct"
1159
1164
MODEL_PATH = f"{ llm_models_root ()} /Qwen2-7B-Instruct"
Original file line number Diff line number Diff line change @@ -400,9 +400,9 @@ accuracy/test_cli_flow.py::TestGemma2_9BIt::test_weight_only[int4]
400
400
accuracy/test_cli_flow.py::TestQwen1_5MoeA2_7BChat::test_auto_dtype
401
401
accuracy/test_cli_flow.py::TestQwen1_5MoeA2_7BChat::test_weight_only
402
402
accuracy/test_cli_flow.py::TestQwen2_0_5BInstruct::test_auto_dtype
403
- accuracy/test_cli_flow.py::TestQwen2_0_5BInstruct::test_auto_dtype_cp4
404
403
accuracy/test_cli_flow.py::TestQwen2_0_5BInstruct::test_weight_only
405
404
accuracy/test_cli_flow.py::TestQwen2_0_5BInstruct::test_fp8
405
+ accuracy/test_cli_flow.py::TestQwen2_1_5B::test_auto_dtype_cp4
406
406
accuracy/test_llm_api.py::TestQwen2_7BInstruct::test_auto_dtype
407
407
accuracy/test_llm_api.py::TestQwen2_7BInstruct::test_weight_only
408
408
accuracy/test_cli_flow.py::TestQwen2_7BInstruct::test_int4_awq_prequantized
You can’t perform that action at this time.
0 commit comments