EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Update python API of activation, topk, norm and rope and remove vllm dependency (#6614)

Browse Source 
Co-authored-by: Wu, Chunyuan <chunyuan.wu@intel.com>
Co-authored-by: jianan-gu <jianan.gu@intel.com>
Co-authored-by: sdp <sdp@gnr799219.jf.intel.com>
This commit is contained in:
YanbingJiang
2025-06-18 13:11:50 +08:00
committed by GitHub
parent e56685ac1b
commit 094c116f7d
23 changed files with 270 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
test/srt/cpu/test_gemm.py
View File

@@ -60,8 +60,8 @@ class TestGemm(CustomTestCase):
)
atol = rtol = precision[ref.dtype]
self.assertTrue(torch.allclose(ref, out, atol=atol, rtol=rtol))
self.assertTrue(torch.allclose(ref, out2, atol=atol, rtol=rtol))
torch.testing.assert_close(ref, out, atol=atol, rtol=rtol)
torch.testing.assert_close(ref, out2, atol=atol, rtol=rtol)
def test_bf16_gemm(self):
for params in itertools.product(
@@ -100,13 +100,13 @@ class TestGemm(CustomTestCase):
out = torch.ops.sgl_kernel.int8_scaled_mm_cpu(
Aq2, Bq, As2, Bs, bias if has_bias else None, torch.bfloat16, False
)
self.assertTrue(torch.allclose(ref_out, out, atol=atol, rtol=rtol))
torch.testing.assert_close(ref_out, out, atol=atol, rtol=rtol)
# test the fused version
fused_out = torch.ops.sgl_kernel.int8_scaled_mm_with_quant(
A, Bq, Bs, bias if has_bias else None, torch.bfloat16, False
)
self.assertTrue(torch.allclose(ref_out, fused_out, atol=atol, rtol=rtol))
torch.testing.assert_close(ref_out, fused_out, atol=atol, rtol=rtol)
def test_int8_gemm(self):
for params in itertools.product(
@@ -165,7 +165,7 @@ class TestGemm(CustomTestCase):
prepack,
)
atol = rtol = precision[ref.dtype]
self.assertTrue(torch.allclose(ref, opt, atol=atol, rtol=rtol))
torch.testing.assert_close(ref, opt, atol=atol, rtol=rtol)
def test_fp8_gemm(self):
for params in itertools.product(