Skip to content
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

Support bloom prefix #2245

Merged
merged 7 commits into from
Oct 17, 2023
Merged

Support bloom prefix #2245

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
247112f
support bloom prefix
rainyfly Oct 12, 2023
ee193ee
support_bloom_prefix
rainyfly Oct 13, 2023
b930913
Merge branch 'llm' of https://github.com/PaddlePaddle/FastDeploy into…
rainyfly Oct 13, 2023
eecd3fd
support bloom prefix
rainyfly Oct 13, 2023
13f69b1
Update code for bloom prefix
rainyfly Oct 17, 2023
0d3dc7a
Merge branch 'llm' of https://github.com/PaddlePaddle/FastDeploy into…
rainyfly Oct 17, 2023
a04373c
update code
rainyfly Oct 17, 2023
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
11 changes: 7 additions & 4 deletions llm/fastdeploy_llm/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,6 +325,7 @@ def generate_position_ids_for_chatglm(seq_len_list):
def update_mask_for_bloom(inputs):
import math
global attention_mask
global tgt_generation_mask

def get_alibi_slopes(num_heads):
closest_power_of_2 = 2**math.floor(math.log2(num_heads))
Expand Down Expand Up @@ -368,6 +369,7 @@ def get_alibi_slopes(num_heads):
alibi_encoder + (1 - inputs["attention_mask"]
) * paddle.finfo(inputs["attention_mask"].dtype).min)
attention_mask = inputs["attention_mask"]
tgt_generation_mask = inputs["tgt_generation_mask"]
inputs["tgt_generation_mask"] = (
alibi_decoder + (1 - inputs["tgt_generation_mask"]) *
paddle.finfo(inputs["tgt_generation_mask"].dtype).min)
Expand Down Expand Up @@ -411,15 +413,15 @@ def dy_input_preprocess(inputs):
paddle.ones(
shape=[length, length], dtype=args.dtype))
if not model_id:
attention_mask[i, 0, :length, :
attention_mask[i, :, :length, :
max_prefix_len] = paddle.zeros(
[1, length, max_prefix_len],
dtype=args.dtype)
tgt_generation_mask[i, 0, 0, max_prefix_len : length + max_prefix_len] = paddle.ones(
shape=[1, length], dtype=args.dtype
)
else:
attention_mask[i, 0, :length, :
attention_mask[i, :, :length, :
max_prefix_len] = paddle.ones(
[1, length, max_prefix_len],
dtype=args.dtype)
Expand All @@ -428,7 +430,7 @@ def dy_input_preprocess(inputs):
shape=[1, max_prefix_len + length],
dtype=args.dtype)

attention_mask[i, 0, :length, max_prefix_len:max_prefix_len +
attention_mask[i, :, :length, max_prefix_len:max_prefix_len +
length] = paddle.tril(
paddle.ones(
shape=[length, length],
Expand All @@ -444,7 +446,6 @@ def dy_input_preprocess(inputs):
dtype=args.dtype)
arange_tensor_encoder[i, :, :length + max_prefix_len] = paddle.arange(
length + max_prefix_len).astype(args.dtype)
inputs["tgt_pos"] = inputs["tgt_pos"] + max_prefix_len


else:
Expand Down Expand Up @@ -497,6 +498,8 @@ def dy_input_preprocess(inputs):
inputs["attention_mask"] = attention_mask

if "bloom" in args.architecture:
if args.is_ptuning:
inputs["tgt_pos"] = inputs["tgt_pos"] + max_prefix_len
inputs["arange_tensor_encoder"] = arange_tensor_encoder
update_mask_for_bloom(inputs)

Expand Down