Skip to content

fix test case on CPU-XPU #113

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
pralay-das wants to merge 2 commits into from
+69 −330
Closed

fix test case on CPU-XPU #113

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9ab985f
fix test case on CPU-XPU
pralay-das Feb 18, 2026
77e9590
remove unnecessary triton comparison from the unit test
pralay-das Feb 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 17 additions & 17 deletions tests/test_fused_qk_norm_rope.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,14 +74,14 @@ def compute_inv_freq_yarn(
inv_freq = 1.0 / (
base
** (
torch.arange(0, rotary_dim, 2, dtype=torch.float32, device=device)
torch.arange(0, rotary_dim, 2, dtype=torch.float32, device="cpu")
/ rotary_dim
)
)

if factor != 1.0:
# YARN scaling
dim_range = torch.arange(0, rotary_dim, 2, dtype=torch.float32, device=device)
dim_range = torch.arange(0, rotary_dim, 2, dtype=torch.float32, device="cpu")

# Compute linear interpolation factor
linear_func = (dim_range - low_freq_factor) / (
Expand Down Expand Up @@ -201,10 +201,10 @@ def test_fused_qk_norm_rope_basic(
position_ids = torch.arange(num_tokens, dtype=torch.int32, device=device)

# Create a copy for reference
qkv_ref = qkv.clone().float()
q_weight_ref = q_weight.clone().float()
k_weight_ref = k_weight.clone().float()
position_ids_ref = position_ids.clone()
qkv_ref = qkv.clone().float().to("cpu")
q_weight_ref = q_weight.clone().float().to("cpu")
k_weight_ref = k_weight.clone().float().to("cpu")
position_ids_ref = position_ids.clone().to("cpu")

# Compute reference output
output_ref = fused_qk_norm_rope_reference(
Expand Down Expand Up @@ -248,7 +248,7 @@ def test_fused_qk_norm_rope_basic(

# Compare results
torch.testing.assert_close(
qkv, output_ref, rtol=precision[dtype], atol=precision[dtype]
qkv.to("cpu"), output_ref, rtol=precision[dtype], atol=precision[dtype]
)


Expand Down Expand Up @@ -278,10 +278,10 @@ def test_fused_qk_norm_rope_yarn(num_tokens, head_dim, is_neox, dtype):
position_ids = torch.arange(num_tokens, dtype=torch.int32, device=device)

# Create a copy for reference
qkv_ref = qkv.clone().float()
q_weight_ref = q_weight.clone().float()
k_weight_ref = k_weight.clone().float()
position_ids_ref = position_ids.clone()
qkv_ref = qkv.clone().float().to("cpu")
q_weight_ref = q_weight.clone().float().to("cpu")
k_weight_ref = k_weight.clone().float().to("cpu")
position_ids_ref = position_ids.clone().to("cpu")

# Compute reference output
output_ref = fused_qk_norm_rope_reference(
Expand Down Expand Up @@ -325,7 +325,7 @@ def test_fused_qk_norm_rope_yarn(num_tokens, head_dim, is_neox, dtype):

# Compare results - use slightly relaxed tolerance for YARN
torch.testing.assert_close(
qkv, output_ref, rtol=precision[dtype] * 2, atol=precision[dtype] * 2
qkv.to("cpu"), output_ref, rtol=precision[dtype] * 2, atol=precision[dtype] * 2
)


Expand Down Expand Up @@ -355,10 +355,10 @@ def test_fused_qk_norm_rope_partial_rotary(num_tokens, head_dim, rotary_dim, dty
position_ids = torch.arange(num_tokens, dtype=torch.int32, device=device)

# Create a copy for reference
qkv_ref = qkv.clone().float()
q_weight_ref = q_weight.clone().float()
k_weight_ref = k_weight.clone().float()
position_ids_ref = position_ids.clone()
qkv_ref = qkv.clone().float().to("cpu")
q_weight_ref = q_weight.clone().float().to("cpu")
k_weight_ref = k_weight.clone().float().to("cpu")
position_ids_ref = position_ids.clone().to("cpu")

# Compute reference output
output_ref = fused_qk_norm_rope_reference(
Expand Down Expand Up @@ -402,7 +402,7 @@ def test_fused_qk_norm_rope_partial_rotary(num_tokens, head_dim, rotary_dim, dty

# Compare results
torch.testing.assert_close(
qkv, output_ref, rtol=precision[dtype], atol=precision[dtype]
qkv.to("cpu"), output_ref, rtol=precision[dtype], atol=precision[dtype]
)


Expand Down
Loading