Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
OpenDAS
AutoAWQ
Commits
ac770875
Commit
ac770875
authored
Sep 12, 2023
by
Casper Hansen
Browse files
Create custom attention shape for Falcon 7B
parent
7f8f9f16
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
32 additions
and
2 deletions
+32
-2
awq/modules/fused/block.py
awq/modules/fused/block.py
+32
-2
No files found.
awq/modules/fused/block.py
View file @
ac770875
...
@@ -33,10 +33,20 @@ class FalconDecoderLayer(nn.Module):
...
@@ -33,10 +33,20 @@ class FalconDecoderLayer(nn.Module):
super
().
__init__
()
super
().
__init__
()
self
.
n_heads
=
n_heads
self
.
n_heads
=
n_heads
self
.
hidden_size
=
hidden_size
self
.
hidden_size
=
hidden_size
# TODO: Falcon has ALiBi implemented but which model uses it?
self
.
attn
=
QuantAttentionFused
(
hidden_size
,
self
.
n_heads
,
qkv_layer
,
o_proj
,
dev
=
dev
,
max_seq_len
=
max_seq_len
,
use_alibi
=
False
).
to
(
dev
)
self
.
new_decoder_arch
=
new_decoder_arch
self
.
new_decoder_arch
=
new_decoder_arch
if
new_decoder_arch
:
attention_shapes
=
None
else
:
attention_shapes
=
self
.
_get_attention_shapes
(
1
,
n_heads
,
max_seq_len
,
self
.
hidden_size
//
n_heads
)
# TODO: Falcon has ALiBi implemented but which model uses it?
self
.
attn
=
QuantAttentionFused
(
hidden_size
,
self
.
n_heads
,
qkv_layer
,
o_proj
,
dev
=
dev
,
max_seq_len
=
max_seq_len
,
use_alibi
=
False
,
attention_shapes
=
attention_shapes
).
to
(
dev
)
if
new_decoder_arch
:
if
new_decoder_arch
:
self
.
ln_attn
=
ln_attn
# before attention
self
.
ln_attn
=
ln_attn
# before attention
self
.
ln_mlp
=
ln_mlp
# before mlp
self
.
ln_mlp
=
ln_mlp
# before mlp
...
@@ -44,6 +54,26 @@ class FalconDecoderLayer(nn.Module):
...
@@ -44,6 +54,26 @@ class FalconDecoderLayer(nn.Module):
self
.
input_layernorm
=
input_layernorm
# before attention
self
.
input_layernorm
=
input_layernorm
# before attention
self
.
mlp
=
mlp
self
.
mlp
=
mlp
def
_get_attention_shapes
(
self
,
batch_size
,
n_heads
,
max_seq_len
,
head_dim
):
self
.
attention_shapes
=
{
# following fastertransformer definition
"cache_v"
:
(
batch_size
,
1
,
max_seq_len
,
head_dim
,),
# 8: pack 8 fp16 in FT, if fp32 then use 4
"cache_k"
:
(
batch_size
,
1
,
head_dim
//
8
,
max_seq_len
,
8
,),
"xqkv_view"
:
(
n_heads
+
2
,
head_dim
),
"xq_slice"
:
lambda
xqkv
:
xqkv
[:,
:,
:
-
2
],
"xk_slice"
:
lambda
xqkv
:
xqkv
[:,
:,
[
-
2
]],
"xv_slice"
:
lambda
xqkv
:
xqkv
[:,
:,
[
-
1
]],
"xk_reshape"
:
(
1
,
head_dim
//
8
,
8
),
"xk_view"
:
(
1
,
head_dim
),
"xv_view"
:
(
1
,
head_dim
),
"single_xq_view"
:
(
n_heads
,
head_dim
),
"single_xk_view"
:
(
1
,
head_dim
),
"single_xv_view"
:
(
1
,
head_dim
)
}
return
self
.
attention_shapes
def
forward
(
def
forward
(
self
,
hidden_states
,
past_key_value
,
attn_bias
=
None
,
attention_mask
=
None
,
is_causal
=
None
self
,
hidden_states
,
past_key_value
,
attn_bias
=
None
,
attention_mask
=
None
,
is_causal
=
None
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment