llama.cpp
GGML WebGPU: Support for ADD, MUL, RMS_NORM, GET_ROWS operators
#16018
Merged
Merged
GGML WebGPU: Support for ADD, MUL, RMS_NORM, GET_ROWS operators #16018
reeselevine
merged
39 commits into
ggml-org:master
from
reeselevine:master
Add paramater buffer pool, batching of submissions, refactor command …
30ba139e
Add header for linux builds
04d7b272
Free staged parameter buffers at once
01c8ced2
Format with clang-format
bfff27f1
Fix thread-safe implementation
b8012ecc
Use device implicit synchronization
cddda7e7
Merge remote-tracking branch 'upstream/master' into fixes
1d5726a2
Update workflow to use custom release
6a20e396
Remove testing branch workflow
ea39068e
some f32 tests passing
96d107e5
Merge branch 'ggml-org:master' into master
4c587428
Disable set_rows until it's implemented
ae8edbfd
f32 add all tests passing
39aa11d9
Merge branch 'master' of https://github.com/reeselevine/llama.cpp int…
2c577269
Begin work on set_rows
6a6135cc
Work on set rows
b2dbfcdc
Add error buffers for reporting unsupported SET_ROWS indices
248f7a51
Remove extra comments
4ad09861
most recent merge
ac522434
Merge remote-tracking branch 'origin/master' into addition
1b16a911
Add templated addition, clean up code
7f9ee10e
Get addition and multiplication working
c1021970
Merge pull request #1 from reeselevine/addition
efc0cb0c
Implement rms_norm
7fbe84cd
Add get_rows implementation
dc7bc4a2
Add new get_rows files
b7635c40
Refactor use of wg size entry
42935317
Fix compilation
ff412050
Merge remote-tracking branch 'upstream/master'
a5da4370
Try manually unrolled q4_0 quant
77f8b965
Revert "Try manually unrolled q4_0 quant"
102f2254
Move to constant max wg size
b0bd49ff
Check for tensor size in supports_op
fc91520f
Vectorize f32 and change default workgroup size
45617844
Merge remote-tracking branch 'upstream/master'
26742e28
Move f32 get_rows from < 4 to % 4 != 0
cfa4fc15
fix linter errors
94228793
Add in-place tests
b877e07f
Merge remote-tracking branch 'upstream/master'
be35439f
ggerganov
approved these changes
on 2025-09-17
Login to write a write a comment.
Login via GitHub