Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
5002175e
Unverified
Commit
5002175e
authored
Jun 17, 2024
by
Joe Runde
Committed by
GitHub
Jun 18, 2024
Browse files
[Kernel] Add punica dimensions for Granite 13b (#5559)
Signed-off-by:
Joe Runde
<
Joseph.Runde@ibm.com
>
parent
daef218b
Changes
2
Show whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
12 additions
and
0 deletions
+12
-0
csrc/punica/bgmv/bgmv_config.h
csrc/punica/bgmv/bgmv_config.h
+8
-0
tests/lora/test_punica.py
tests/lora/test_punica.py
+4
-0
No files found.
csrc/punica/bgmv/bgmv_config.h
View file @
5002175e
...
...
@@ -20,6 +20,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, narrow, 1152) \
f(in_T, out_T, W_T, narrow, 1280) \
f(in_T, out_T, W_T, narrow, 1536) \
f(in_T, out_T, W_T, narrow, 1664) \
f(in_T, out_T, W_T, narrow, 1728) \
f(in_T, out_T, W_T, narrow, 1792) \
f(in_T, out_T, W_T, narrow, 2048) \
...
...
@@ -36,6 +37,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, narrow, 5120) \
f(in_T, out_T, W_T, narrow, 5504) \
f(in_T, out_T, W_T, narrow, 5632) \
f(in_T, out_T, W_T, narrow, 5888) \
f(in_T, out_T, W_T, narrow, 6144) \
f(in_T, out_T, W_T, narrow, 6400) \
f(in_T, out_T, W_T, narrow, 6848) \
...
...
@@ -45,6 +47,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, narrow, 9216) \
f(in_T, out_T, W_T, narrow, 10240) \
f(in_T, out_T, W_T, narrow, 11008) \
f(in_T, out_T, W_T, narrow, 11264) \
f(in_T, out_T, W_T, narrow, 12288) \
f(in_T, out_T, W_T, narrow, 13696) \
f(in_T, out_T, W_T, narrow, 13824) \
...
...
@@ -53,6 +56,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, narrow, 16384) \
f(in_T, out_T, W_T, narrow, 20480) \
f(in_T, out_T, W_T, narrow, 22016) \
f(in_T, out_T, W_T, narrow, 22528) \
f(in_T, out_T, W_T, narrow, 24576) \
f(in_T, out_T, W_T, narrow, 27392) \
f(in_T, out_T, W_T, narrow, 27648) \
...
...
@@ -91,6 +95,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, 1152, narrow) \
f(in_T, out_T, W_T, 1280, narrow) \
f(in_T, out_T, W_T, 1536, narrow) \
f(in_T, out_T, W_T, 1664, narrow) \
f(in_T, out_T, W_T, 1728, narrow) \
f(in_T, out_T, W_T, 1792, narrow) \
f(in_T, out_T, W_T, 2048, narrow) \
...
...
@@ -107,6 +112,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, 5120, narrow) \
f(in_T, out_T, W_T, 5504, narrow) \
f(in_T, out_T, W_T, 5632, narrow) \
f(in_T, out_T, W_T, 5888, narrow) \
f(in_T, out_T, W_T, 6144, narrow) \
f(in_T, out_T, W_T, 6400, narrow) \
f(in_T, out_T, W_T, 6848, narrow) \
...
...
@@ -116,6 +122,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, 9216, narrow) \
f(in_T, out_T, W_T, 10240, narrow) \
f(in_T, out_T, W_T, 11008, narrow) \
f(in_T, out_T, W_T, 11264, narrow) \
f(in_T, out_T, W_T, 12288, narrow) \
f(in_T, out_T, W_T, 13696, narrow) \
f(in_T, out_T, W_T, 13824, narrow) \
...
...
@@ -124,6 +131,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
f(in_T, out_T, W_T, 16384, narrow) \
f(in_T, out_T, W_T, 20480, narrow) \
f(in_T, out_T, W_T, 22016, narrow) \
f(in_T, out_T, W_T, 22528, narrow) \
f(in_T, out_T, W_T, 24576, narrow) \
f(in_T, out_T, W_T, 27392, narrow) \
f(in_T, out_T, W_T, 27648, narrow) \
...
...
tests/lora/test_punica.py
View file @
5002175e
...
...
@@ -53,6 +53,7 @@ H1 = H2 = [
1152
,
1280
,
1536
,
1664
,
2048
,
2304
,
2560
,
...
...
@@ -66,6 +67,7 @@ H1 = H2 = [
5120
,
5504
,
5632
,
5888
,
6144
,
6400
,
6848
,
...
...
@@ -75,10 +77,12 @@ H1 = H2 = [
9216
,
10240
,
11008
,
11264
,
13824
,
14336
,
15360
,
22016
,
22528
,
24576
,
27392
,
27648
,
...
...
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