[CI/Build] Fix unused fake implementation testing (#55797)
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Jee Jee Li committed
Sep 8, 2026 at 11:36 UTC
607a6e48c5b9780ae8405497f77bee65698c61c7
6 files changed
-66
tests/kernels/helion/test_dynamic_per_token_scaled_fp8_quant.py
-11
@@ -152,14 +152,3 @@ class TestDynamicPerTokenScaledFp8QuantIntegration:
152
assert kernel_wrapper.op_name == "dynamic_per_token_scaled_fp8_quant"
153
assert kernel_wrapper._config_picker is not None
154
assert kernel_wrapper._mutates_args == ["result", "scale"]
155
-
156
- def test_fake_impl_functionality(self):
157
- skip_if_platform_unsupported("dynamic_per_token_scaled_fp8_quant")
158
- from vllm.kernels.helion.register import get_registered_kernels
159
-
160
- registered_kernels = get_registered_kernels()
161
- kernel_wrapper = registered_kernels["dynamic_per_token_scaled_fp8_quant"]
162
- fake_impl = kernel_wrapper._fake_impl
163
-
164
- args = _generate_fake_input(16, 4096)
165
- assert fake_impl(*args) is None
tests/kernels/helion/test_fused_qk_norm_rope.py
-11
@@ -248,14 +248,3 @@ class TestFusedQkNormRopeIntegration:
248
assert kernel_wrapper.op_name == "fused_qk_norm_rope"
249
assert kernel_wrapper._config_picker is not None
250
assert kernel_wrapper._mutates_args == ["qkv"]
251
-
252
- def test_fake_impl_functionality(self):
253
- skip_if_platform_unsupported("fused_qk_norm_rope")
254
- from vllm.kernels.helion.register import get_registered_kernels
255
-
256
- registered_kernels = get_registered_kernels()
257
- kernel_wrapper = registered_kernels["fused_qk_norm_rope"]
258
- fake_impl = kernel_wrapper._fake_impl
259
-
260
- args = _generate_fake_input(16, 4096, 128)
261
- assert fake_impl(*args) is None
tests/kernels/helion/test_per_token_group_fp8_quant.py
-11
@@ -230,14 +230,3 @@ class TestPerTokenGroupFp8QuantIntegration:
230
assert kernel_wrapper.op_name == "per_token_group_fp8_quant"
231
assert kernel_wrapper._config_picker is not None
232
assert kernel_wrapper._mutates_args == ["output_q", "output_s"]
233
-
234
- def test_fake_impl_functionality(self):
235
- skip_if_platform_unsupported("per_token_group_fp8_quant")
236
- from vllm.kernels.helion.register import get_registered_kernels
237
-
238
- registered_kernels = get_registered_kernels()
239
- kernel_wrapper = registered_kernels["per_token_group_fp8_quant"]
240
- fake_impl = kernel_wrapper._fake_impl
241
-
242
- args = _generate_fake_input(16, 4096, 128)
243
- assert fake_impl(*args) is None
tests/kernels/helion/test_rms_norm_dynamic_per_token_quant.py
-11
@@ -194,14 +194,3 @@ class TestRmsNormDynamicPerTokenQuantIntegration:
194
assert kernel_wrapper.op_name == "rms_norm_dynamic_per_token_quant"
195
assert kernel_wrapper._config_picker is not None
196
assert kernel_wrapper._mutates_args == ["result", "scale", "residual"]
197
-
198
- def test_fake_impl_functionality(self):
199
- skip_if_platform_unsupported("rms_norm_dynamic_per_token_quant")
200
- from vllm.kernels.helion.register import get_registered_kernels
201
-
202
- registered_kernels = get_registered_kernels()
203
- kernel_wrapper = registered_kernels["rms_norm_dynamic_per_token_quant"]
204
- fake_impl = kernel_wrapper._fake_impl
205
-
206
- args = _generate_fake_input(16, 4096)
207
- assert fake_impl(*args) is None
tests/kernels/helion/test_rms_norm_per_block_quant.py
-11
@@ -298,14 +298,3 @@ class TestRmsNormPerBlockQuantIntegration:
298
assert kernel_wrapper.op_name == "rms_norm_per_block_quant"
299
assert kernel_wrapper._config_picker is not None
300
assert kernel_wrapper._mutates_args == ["result", "scale", "residual"]
301
-
302
- def test_fake_impl_functionality(self):
303
- skip_if_platform_unsupported("rms_norm_per_block_quant")
304
- from vllm.kernels.helion.register import get_registered_kernels
305
-
306
- registered_kernels = get_registered_kernels()
307
- kernel_wrapper = registered_kernels["rms_norm_per_block_quant"]
308
- fake_impl = kernel_wrapper._fake_impl
309
-
310
- args = _generate_fake_input(16, 4096, 128)
311
- assert fake_impl(*args) is None
tests/kernels/helion/test_silu_and_mul_per_block_quant.py
-11
@@ -211,14 +211,3 @@ class TestSiluAndMulPerBlockQuantIntegration:
211
assert kernel_wrapper.op_name == "silu_and_mul_per_block_quant"
212
assert kernel_wrapper._config_picker is not None
213
assert kernel_wrapper._mutates_args == ["out", "scales"]
214
-
215
- def test_fake_impl_functionality(self):
216
- skip_if_platform_unsupported("silu_and_mul_per_block_quant")
217
- from vllm.kernels.helion.register import get_registered_kernels
218
-
219
- registered_kernels = get_registered_kernels()
220
- kernel_wrapper = registered_kernels["silu_and_mul_per_block_quant"]
221
- fake_impl = kernel_wrapper._fake_impl
222
-
223
- args = _generate_fake_input(16, 4096, 128)
224
- assert fake_impl(*args) is None