Skip to content

Mamba2 SSD #16982

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

Draft
gabe-l-hart wants to merge 9 commits into
base: master
Choose a base branch
from
+575 −28
Draft

Mamba2 SSD #16982

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
9d715f4
feat: First-pass at porting SSD impl from previous work
gabe-l-hart Dec 3, 2025
960bb52
fix: Increase max nodes for models known to use mamba2
gabe-l-hart Dec 3, 2025
0dee5b1
TEMP: Disable everything after ssm_conv in mamba2
gabe-l-hart Dec 6, 2025
4b8f1d5
feat: Add a batched version of ssm_conv
gabe-l-hart Dec 6, 2025
953bb62
un-TEMP: Re-enable the SSM portion
gabe-l-hart Dec 6, 2025
65feb1c
feat: Optimized SSM_SCAN kernel for metal
gabe-l-hart Dec 6, 2025
51adb32
feat: Optimization pass over SSD implementation
gabe-l-hart Dec 6, 2025
6340ab1
TEMP (wip): Partial work towards a unified kernel implementation of SSD
gabe-l-hart Dec 8, 2025
11067a7
fix: Provide macos-specific backtrace printing to avoid terminal death
gabe-l-hart Dec 8, 2025
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
Prev Previous commit
Next Next commit
TEMP: Disable everything after ssm_conv in mamba2 
Branch: Mamba2SSD

Signed-off-by: Gabe Goodhart <[email protected]>
  • Loading branch information
@gabe-l-hart
gabe-l-hart committed Dec 8, 2025
commit 0dee5b15000858760adcf825ed31fde5b78bf0bc
6 changes: 6 additions & 0 deletions src/llama-graph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -244,6 +244,12 @@ void llm_graph_input_rs::set_input(const llama_ubatch * ubatch) {
const int64_t n_rs = mctx->get_n_rs();

if (s_copy) {
//DEBUG
if (!s_copy->buffer) {
return;
}
//DEBUG

GGML_ASSERT(ggml_backend_buffer_is_host(s_copy->buffer));
int32_t * data = (int32_t *) s_copy->data;

Expand Down
Loading