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
chenpangpang
transformers
Commits
3b7675b2
Unverified
Commit
3b7675b2
authored
Dec 26, 2023
by
Sourab Mangrulkar
Committed by
GitHub
Dec 26, 2023
Browse files
fix FA2 when using quantization (#28203)
parent
fa21ead7
Changes
5
Hide whitespace changes
Inline
Side-by-side
Showing
5 changed files
with
15 additions
and
15 deletions
+15
-15
src/transformers/models/falcon/modeling_falcon.py
src/transformers/models/falcon/modeling_falcon.py
+3
-3
src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py
src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py
+3
-3
src/transformers/models/llama/modeling_llama.py
src/transformers/models/llama/modeling_llama.py
+3
-3
src/transformers/models/mistral/modeling_mistral.py
src/transformers/models/mistral/modeling_mistral.py
+3
-3
src/transformers/models/mixtral/modeling_mixtral.py
src/transformers/models/mixtral/modeling_mixtral.py
+3
-3
No files found.
src/transformers/models/falcon/modeling_falcon.py
View file @
3b7675b2
...
...
@@ -617,11 +617,11 @@ class FalconFlashAttention2(FalconAttention):
# cast them back in float16 just to be sure everything works as expected.
input_dtype
=
query_layer
.
dtype
if
input_dtype
==
torch
.
float32
:
if
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
# Handle the case where the model is quantized
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
el
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
target_dtype
=
self
.
config
.
_pre_quantization_dtype
elif
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
else
:
target_dtype
=
self
.
query_key_value
.
weight
.
dtype
...
...
src/transformers/models/gpt_bigcode/modeling_gpt_bigcode.py
View file @
3b7675b2
...
...
@@ -375,11 +375,11 @@ class GPTBigCodeFlashAttention2(GPTBigCodeAttention):
# cast them back in float16 just to be sure everything works as expected.
input_dtype
=
query
.
dtype
if
input_dtype
==
torch
.
float32
:
if
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
# Handle the case where the model is quantized
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
el
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
target_dtype
=
self
.
config
.
_pre_quantization_dtype
elif
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
else
:
target_dtype
=
self
.
c_attn
.
weight
.
dtype
...
...
src/transformers/models/llama/modeling_llama.py
View file @
3b7675b2
...
...
@@ -528,11 +528,11 @@ class LlamaFlashAttention2(LlamaAttention):
input_dtype
=
query_states
.
dtype
if
input_dtype
==
torch
.
float32
:
if
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
# Handle the case where the model is quantized
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
el
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
target_dtype
=
self
.
config
.
_pre_quantization_dtype
elif
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
else
:
target_dtype
=
self
.
q_proj
.
weight
.
dtype
...
...
src/transformers/models/mistral/modeling_mistral.py
View file @
3b7675b2
...
...
@@ -428,11 +428,11 @@ class MistralFlashAttention2(MistralAttention):
# cast them back in float16 just to be sure everything works as expected.
input_dtype
=
query_states
.
dtype
if
input_dtype
==
torch
.
float32
:
if
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
# Handle the case where the model is quantized
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
el
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
target_dtype
=
self
.
config
.
_pre_quantization_dtype
elif
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
else
:
target_dtype
=
self
.
q_proj
.
weight
.
dtype
...
...
src/transformers/models/mixtral/modeling_mixtral.py
View file @
3b7675b2
...
...
@@ -477,11 +477,11 @@ class MixtralFlashAttention2(MixtralAttention):
# cast them back in float16 just to be sure everything works as expected.
input_dtype
=
query_states
.
dtype
if
input_dtype
==
torch
.
float32
:
if
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
# Handle the case where the model is quantized
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
el
if
hasattr
(
self
.
config
,
"_pre_quantization_dtype"
):
target_dtype
=
self
.
config
.
_pre_quantization_dtype
elif
torch
.
is_autocast_enabled
():
target_dtype
=
torch
.
get_autocast_gpu_dtype
()
else
:
target_dtype
=
self
.
q_proj
.
weight
.
dtype
...
...
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