fix dflash rope config parsing for updated transformers (#27477)
This commit is contained in:
@@ -31,6 +31,7 @@ from sglang.srt.speculative.dflash_utils import (
|
|||||||
parse_dflash_draft_config,
|
parse_dflash_draft_config,
|
||||||
)
|
)
|
||||||
from sglang.srt.utils import is_npu
|
from sglang.srt.utils import is_npu
|
||||||
|
from sglang.srt.utils.hf_transformers_utils import get_rope_config
|
||||||
|
|
||||||
_is_npu = is_npu()
|
_is_npu = is_npu()
|
||||||
if _is_npu:
|
if _is_npu:
|
||||||
@@ -94,8 +95,7 @@ class DFlashAttention(nn.Module):
|
|||||||
self.q_norm = RMSNorm(head_dim, eps=rms_norm_eps)
|
self.q_norm = RMSNorm(head_dim, eps=rms_norm_eps)
|
||||||
self.k_norm = RMSNorm(head_dim, eps=rms_norm_eps)
|
self.k_norm = RMSNorm(head_dim, eps=rms_norm_eps)
|
||||||
|
|
||||||
rope_theta = float(getattr(config, "rope_theta", 1000000))
|
rope_theta, rope_scaling = get_rope_config(config)
|
||||||
rope_scaling = getattr(config, "rope_scaling", None)
|
|
||||||
rope_is_neox_style = bool(
|
rope_is_neox_style = bool(
|
||||||
getattr(
|
getattr(
|
||||||
config, "rope_is_neox_style", getattr(config, "is_neox_style", True)
|
config, "rope_is_neox_style", getattr(config, "is_neox_style", True)
|
||||||
|
|||||||
Reference in New Issue
Block a user