zq/add label for llm in device_configs (DeepLink-org#809)

* add label for llm in device_configs


---------

Co-authored-by: zhousike <[email protected]>

Author: NeosZhang

impl/ascend/device_configs.py

@@ -3,6 +3,8 @@
 from skip import Skip
 
 device_configs = {
+    # topk llm used
+    # normal llm used
     # temp for 910B
     'join': dict(
         name=['stack'],
@@ -481,7 +483,7 @@
         ),
     ),
 
-    'pow_tensor': dict(
+    'pow_tensor': dict( # llm used
         name=['pow'],
         tensor_para=dict(
             args=[
@@ -493,7 +495,7 @@
         ),
     ),
 
-    'pow_tensor_only_0_1': dict(
+    'pow_tensor_only_0_1': dict( # llm used
         name=['pow'],
         tensor_para=dict(
             args=[
@@ -505,7 +507,7 @@
         ),
     ),
 
-    'pow_diff_dtype': dict(
+    'pow_diff_dtype': dict( # llm used
         name=['pow'],
         tensor_para=dict(
             args=[
@@ -517,7 +519,7 @@
         ),
     ),
 
-    'bmm': dict(
+    'bmm': dict( # llm used
         name=['bmm'],
         atol=3e-2,
         rtol=3e-2,
@@ -535,13 +537,13 @@
         ),
     ),
 
-    'reduce_op': dict(
+    'reduce_op': dict( # llm used
         name=['sum'],
         atol=1e-3,
         rtol=1e-3,
     ),
 
-    'reduce_partial_op': dict(
+    'reduce_partial_op': dict( # llm used
         atol=1e-3,
         rtol=1e-3,
         name=['sum'],
@@ -571,7 +573,7 @@
         ),
     ),
 
-    'nll_loss': dict(
+    'nll_loss': dict( # llm used
         name=['nll_loss'],
         tensor_para=dict(
             args=[
@@ -584,7 +586,7 @@
         ),
     ),
 
-    'nll_loss_empty_tensor': dict(
+    'nll_loss_empty_tensor': dict( # llm used
         name=['nll_loss'],
         tensor_para=dict(
             args=[
@@ -675,7 +677,7 @@
         rtol = 1e-1,
     ),
 
-    'embedding': dict(
+    'embedding': dict( # llm used
         name=["embedding"],
         tensor_para=dict(
             args=[
@@ -736,7 +738,7 @@
         ),
     ),
 
-    'split': dict(
+    'split': dict( # llm used
         name=['split'],
         tensor_para=dict(
             args=[
@@ -1024,13 +1026,13 @@
         ),
     ),
 
-    'mm': dict(
+    'mm': dict( # llm used
         name=['mm'],
         atol=2e-2,
         rtol=2e-2,
     ),
 
-    'mm_diff_dtype': dict(
+    'mm_diff_dtype': dict( # llm used
         name=['mm'],
         atol=2e-2,
         rtol=2e-2,
@@ -1156,7 +1158,7 @@
         ),
     ),
 
-    'norm': dict(
+    'norm': dict( # llm used
         name=['norm'],
         tensor_para=dict(
             args=[
@@ -1232,7 +1234,7 @@
         ),
     ),
 
-    'gather': dict(
+    'gather': dict( # llm used
         name=['gather'],
         tensor_para=dict(
             args=[
@@ -1244,7 +1246,7 @@
         ),
     ),
 
-    'gather_0dim': dict(
+    'gather_0dim': dict( # llm used
         name=['gather'],
         tensor_para=dict(
             args=[
@@ -1256,7 +1258,7 @@
         ),
     ),
 
-    'gather_not_float': dict(
+    'gather_not_float': dict( # llm used
         name=['gather'],
         tensor_para=dict(
             args=[
@@ -1268,7 +1270,7 @@
         ),
     ),
 
-    'scatter': dict(
+    'scatter': dict( # llm used
         name=['scatter'],
         tensor_para=dict(
             args=[
@@ -1280,7 +1282,7 @@
         ),
     ),
 
-    'scatter_scalar': dict(
+    'scatter_scalar': dict( # llm used
         name=['scatter'],
         para=dict(
             # In this case, for float32 (but not float64), no matter what the value parameter is,
@@ -1291,7 +1293,7 @@
         ),
     ),
 
-    'index_put_acc_three_indices': dict(
+    'index_put_acc_three_indices': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1303,7 +1305,7 @@
         ),
     ),
 
-    'index_put_acc_two_indices': dict(
+    'index_put_acc_two_indices': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1315,7 +1317,7 @@
         ),
     ),
 
-    'index_put_acc_one_indices': dict(
+    'index_put_acc_one_indices': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1327,7 +1329,7 @@
         ),
     ),
 
-    'index_put_acc_bool_indices_zeros': dict(
+    'index_put_acc_bool_indices_zeros': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1339,7 +1341,7 @@
         ),
     ),
 
-    'index_put_one_indices': dict(
+    'index_put_one_indices': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1351,7 +1353,7 @@
         ),
     ),
 
-    'index_put_bool_indices_value': dict(
+    'index_put_bool_indices_value': dict( # llm used
         name=['index_put'],
         tensor_para=dict(
             args=[
@@ -1436,7 +1438,7 @@
         ),
     ),
 
-    'copy': dict(
+    'copy': dict( # llm used
         name=["copy_"],
         tensor_para=dict(
             # FIXME data type DT_COMPLEX128 of input [dst] is not supported
@@ -1454,7 +1456,7 @@
         )
     ),
 
-    'copy_input_no_contiguous': dict(
+    'copy_input_no_contiguous': dict( # llm used
         name=["copy_"],
         tensor_para=dict(
             # FIXME not supported complex
@@ -1472,7 +1474,7 @@
         )
     ),
 
-    'copy_other_no_contiguous': dict(
+    'copy_other_no_contiguous': dict( # llm used
         name=["copy_"],
         tensor_para=dict(
             # FIXME data type DT_COMPLEX64 of input [dst] is not supported
@@ -1491,7 +1493,7 @@
         )
     ),
 
-    'copy_all_no_contiguous': dict(
+    'copy_all_no_contiguous': dict( # llm used
         name=["copy_"],
         tensor_para=dict(
             # FIXME data type DT_COMPLEX64 of input [dst] is not supported
@@ -1509,7 +1511,7 @@
         )
     ),
 
-    'fill_not_float': dict(
+    'fill_not_float': dict( # llm used
         name=["fill_"],
         tensor_para=dict(
             args=[
@@ -1581,7 +1583,7 @@
         ),
     ),
 
-    'repeat': dict(
+    'repeat': dict( # llm used
         name=['repeat'],
         tensor_para=dict(
             args=[
@@ -1641,7 +1643,7 @@
         ),
     ),
 
-    'triu': dict(
+    'triu': dict( # llm used
         name=['triu'],
         tensor_para=dict(
             args=[
@@ -1677,7 +1679,7 @@
         ),
     ),
 
-    'reduce_partial_op_4': dict(
+    'reduce_partial_op_4': dict( # llm used
         name=['sum'],
         interface=['torch'],
         atol=1e-4,