Inference UTs check for trition support from accelerator (#6782)

      Instead of checking if installed or not check for support. Skip if not
supported.

Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com>

tests/unit/ops/transformer/inference/test_attention.py

@@ -27,8 +27,8 @@ def ref_torch_attention(q, k, v, mask, sm_scale):
 @pytest.mark.parametrize("causal", [True, False])
 @pytest.mark.parametrize("use_flash", [True, False])
 def test_attention(BATCH, H, N_CTX, D_HEAD, causal, use_flash, dtype=torch.float16):
-    if not deepspeed.HAS_TRITON:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
 
     minus_inf = -65504.0
     dev = deepspeed.accelerator.get_accelerator().device_name()

tests/unit/ops/transformer/inference/test_gelu.py

@@ -61,8 +61,8 @@ def test_gelu(batch, sequence, channels, dtype, use_triton_ops):
     activations_ds = torch.randn((batch, sequence, channels), dtype=dtype, device=device)
     activations_ref = activations_ds.clone().detach()
 
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
     ds_out = run_gelu_ds(activations_ds, use_triton_ops)
     ref_out = run_gelu_reference(activations_ref)
     assert (allclose(ds_out, ref_out))

tests/unit/ops/transformer/inference/test_layer_norm.py

@@ -45,8 +45,8 @@ def ds_triton_implementation(vals, gamma, beta, epsilon):
 @pytest.mark.parametrize("dtype", get_dtypes())
 @pytest.mark.parametrize("use_triton_ops", [False, True])
 def test_layer_norm(batch, seq_len, channels, dtype, use_triton_ops):
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
 
     vals = torch.randn((batch, seq_len, channels), dtype=dtype, device=get_accelerator().current_device_name())
     gamma = torch.randn((channels), dtype=dtype, device=get_accelerator().current_device_name())
@@ -93,8 +93,8 @@ def residual_ds_triton_implementation(vals, bias, res, gamma, beta, epsilon):
 @pytest.mark.parametrize("dtype", get_dtypes())
 @pytest.mark.parametrize("use_triton_ops", [False, True])
 def test_layer_norm_residual(batch, seq_len, channels, dtype, use_triton_ops):
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
 
     vals = torch.randn((batch, seq_len, channels), dtype=dtype, device=get_accelerator().current_device_name())
     residual = torch.randn((batch, seq_len, channels), dtype=dtype, device=get_accelerator().current_device_name())
@@ -163,8 +163,8 @@ def test_layer_norm_residual_store_pre_ln_res(batch, seq_len, channels, dtype):
 @pytest.mark.parametrize("residual", [True, False])
 @pytest.mark.parametrize("input_bias", [True, False])
 def test_triton_layer_norm(M, N, dtype, residual, input_bias, eps=1e-5, device='cuda'):
-    if not deepspeed.HAS_TRITON:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
     dev = get_accelerator().device_name()
     torch.manual_seed(0)
     # create data

tests/unit/ops/transformer/inference/test_matmul.py

@@ -42,8 +42,8 @@ def run_matmul_ds(a, b, use_triton_ops=False):
 @pytest.mark.parametrize("dtype", [torch.float16])
 @pytest.mark.parametrize("use_triton_ops", [True])
 def test_matmul_4d(B, H, M, K, N, dtype, use_triton_ops):
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
 
     # skip autotune in testing
     from deepspeed.ops.transformer.inference.triton.matmul_ext import fp16_matmul

tests/unit/ops/transformer/inference/test_residual_add.py

@@ -74,8 +74,8 @@ def run_residual_add_reference(hidden_state, residual, attn_output, attn_bias, f
 @pytest.mark.parametrize("use_triton_ops", [True, False])
 def test_residual_add(batch, sequence, hidden_dim, dtype, mlp_after_attn, add_bias, mp_size, pre_attn_norm,
                       use_triton_ops):
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
     ds_out = torch.randn((batch, sequence, hidden_dim), dtype=dtype, device=get_accelerator().device_name())
     residual = torch.randn((batch, sequence, hidden_dim), dtype=dtype, device=get_accelerator().device_name())
     attn_output = torch.randn((batch, sequence, hidden_dim), dtype=dtype, device=get_accelerator().device_name())

tests/unit/ops/transformer/inference/test_softmax.py

@@ -40,8 +40,8 @@ def run_softmax_ds(input, use_triton_ops=False):
 @pytest.mark.parametrize("dtype", [torch.float16, torch.float32])
 @pytest.mark.parametrize("use_triton_ops", [True])
 def test_softmax(batch, sequence, channels, dtype, use_triton_ops):
-    if not deepspeed.HAS_TRITON and use_triton_ops:
-        pytest.skip("triton has to be installed for the test")
+    if not deepspeed.get_accelerator().is_triton_supported():
+        pytest.skip("triton is not supported on this system")
 
     device = deepspeed.accelerator.get_accelerator().device_name()
     input_ds = torch.randn((batch, sequence, channels), dtype=dtype, device=device)