Skip to content

GitLab

Unverified Commit ab1091d5 authored by wangxiyuan's avatar wangxiyuan Committed by GitHub
Browse files

[Misc][Attention][Quantization] init property earlier (#13733) 


Signed-off-by: default avatarwangxiyuan <wangxiyuan1007@gmail.com>
parent 1e15aaef
Show whitespace changes
Inline Side-by-side
Showing with 5 additions and 4 deletions
vllm/attention/layer.py
View file @ ab1091d5
...@@ -85,6 +85,11 @@ class Attention(nn.Module): ...@@ -85,6 +85,11 @@ class Attention(nn.Module):
self._k_scale_float = 1.0 self._k_scale_float = 1.0
self._v_scale_float = 1.0 self._v_scale_float = 1.0
self.num_heads = num_heads
self.head_size = head_size
self.num_kv_heads = num_kv_heads
self.sliding_window = sliding_window
quant_method = quant_config.get_quant_method( quant_method = quant_config.get_quant_method(
self, prefix=prefix) if quant_config else None self, prefix=prefix) if quant_config else None
if quant_method is not None: if quant_method is not None:
...@@ -116,10 +121,6 @@ class Attention(nn.Module): ...@@ -116,10 +121,6 @@ class Attention(nn.Module):
alibi_slopes, sliding_window, kv_cache_dtype, alibi_slopes, sliding_window, kv_cache_dtype,
blocksparse_params, logits_soft_cap, attn_type, blocksparse_params, logits_soft_cap, attn_type,
**extra_impl_args) **extra_impl_args)
self.num_heads = num_heads
self.head_size = head_size
self.num_kv_heads = num_kv_heads
self.sliding_window = sliding_window
self.backend = backend_name_to_enum(attn_backend.get_name()) self.backend = backend_name_to_enum(attn_backend.get_name())
self.dtype = dtype self.dtype = dtype
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment