Skip to content

fix accuracy #4142

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

Draft
wants to merge 3 commits into
base: main
Choose a base branch
from
+6 −7
Draft

fix accuracy #4142

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d727d63
calculate inv_freq on device
yao-fengchen Nov 20, 2025
6139945
adapt for dlinfer attn
yao-fengchen Nov 20, 2025
17c386e
update code
yao-fengchen Nov 20, 2025
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
8 changes: 4 additions & 4 deletions lmdeploy/pytorch/backends/dlinfer/ascend/op_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,7 +212,7 @@ def get_total_slots():
elif is_unpaged_prefill:
# prepare some params of unpaged_prefill attention stage.
q_start_loc_cpu, kv_seqlens_cpu = None, None
q_seqlens_cpu = step_context.q_seqlens.cpu()
q_seqlens_cpu = step_context.q_seqlens.cpu().to(torch.int32)
if SocVersion.is_Ascend910():
single_attention_mask = torch.logical_not(
torch.tril(
Expand Down Expand Up @@ -251,7 +251,7 @@ def get_total_slots():
step_context.block_offsets = step_context.block_offsets\
.repeat_interleave(step_context.q_seqlens, 0)
dynamo.mark_dynamic(step_context.block_offsets, [0, 1])
kv_seqlens = step_context.kv_seqlens.to(torch.int32)
kv_seqlens = step_context.kv_seqlens.cpu().to(torch.int32)
if not step_context.is_decoding:
if is_unpaged_prefill:
if SocVersion.is_Ascend910():
Expand All @@ -270,10 +270,10 @@ def get_total_slots():
raise ValueError(f"dlinfer doesn't support {SocVersion.device_name()} device currently.")
kv_seqlens = kv_seqlens.repeat_interleave(step_context.q_seqlens, 0)
if not is_unpaged_prefill and AscendOpsBackend.enable_aclgraph():
kv_seqlens = kv_seqlens.cpu().tolist()
kv_seqlens = kv_seqlens.cpu().to(torch.int32)
else:
if step_context.is_decoding:
kv_seqlens_cpu = step_context.kv_seqlens.cpu()
kv_seqlens_cpu = step_context.kv_seqlens.cpu().to(torch.int32)
elif is_unpaged_prefill:
pass
else:
Expand Down
5 changes: 2 additions & 3 deletions lmdeploy/pytorch/backends/dlinfer/rotary_embedding.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,14 +40,13 @@ def _rotary_embedding_fwd(position_ids: torch.Tensor,
class DlinferRotaryEmbeddingImpl(RotaryEmbeddingImpl, nn.Module):
"""Base rotary embedding."""

def __init__(self, dim: int, base: int = 10000, scaling_factor: float = 1.0):
def __init__(self, dim: int, base: float = 10000.0, scaling_factor: float = 1.0):
super().__init__()
self.scaling_factor = scaling_factor
self.dim = dim
self.base = base
# yapf: disable
inv_freq = 1.0 / (self.base
** (torch.arange(0, self.dim, 2, dtype=torch.int64).float() / self.dim)).float().cuda()
inv_freq = 1.0 / (self.base**(torch.arange(0, self.dim, 2, dtype=torch.float, device='cuda') / self.dim))
# yapf: enable
self.register_buffer('inv_freq', inv_freq, persistent=False)

Expand Down