[Build] Migrate vendored DeepGEMM from pybind to TORCH_LIBRARY (abi3) (#48962)
Signed-off-by: Chris Leonard <chleonar@redhat.com> Co-authored-by: Shengqi Chen <harry-chen@outlook.com>
This commit is contained in:
co-authored by
Shengqi Chen
parent
cbe1f9740d
commit
cc73cca9c8
@@ -179,6 +179,7 @@ def _deepgemm_fp8_gemm_nt_warmup(
|
||||
|
||||
device = w.device
|
||||
a1q = torch.empty((max_tokens, k), device=device, dtype=torch.float8_e4m3fn)
|
||||
# Must be initialized (UE8M0 packing asserts zero sign/mantissa bits).
|
||||
a1q_scales = torch.zeros(
|
||||
(max_tokens, k // block_m), device=device, dtype=torch.float32
|
||||
)
|
||||
@@ -284,6 +285,7 @@ def _deepgemm_grouped_fp8_gemm_nt_contiguous_warmup(
|
||||
((MAX_M, k // block_m), torch.float32),
|
||||
((MAX_M, n), torch.bfloat16),
|
||||
)
|
||||
# Must be initialized (UE8M0 packing asserts zero sign/mantissa bits).
|
||||
a1q_scales.zero_()
|
||||
|
||||
for num_tokens, align_used, expert_ids in warmup_cases:
|
||||
|
||||
Reference in New Issue
Block a user