Skip to content

ggml-cuda: Delta-Net linear attention for Qwen3-Next #18102

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

Open
hauhaut wants to merge 3 commits into
base: master
Choose a base branch
from
+2,686 −177
Open

ggml-cuda: Delta-Net linear attention for Qwen3-Next #18102

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4114537
ggml-cuda: Delta-Net linear attention for Qwen3-Next
hauhaut Dec 16, 2025
0a19293
qwen3next: trim comments
hauhaut Dec 16, 2025
128a6c2
ggml-cpu: add DELTA_NET backend + tests
hauhaut Dec 16, 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
10 changes: 10 additions & 0 deletions ggml/include/ggml.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -551,6 +551,7 @@ extern "C" {
GGML_OP_GATED_LINEAR_ATTN,
GGML_OP_RWKV_WKV7,
GGML_OP_SOLVE_TRI,
GGML_OP_DELTA_NET,

GGML_OP_UNARY,

Expand Down Expand Up @@ -2460,6 +2461,15 @@ extern "C" {
bool lower,
bool uni);

GGML_API struct ggml_tensor * ggml_delta_net(
struct ggml_context * ctx,
struct ggml_tensor * q, // [S_k, n_tokens, H_k, n_seqs] - Query (pre-permuted)
struct ggml_tensor * k, // [S_k, n_tokens, H_k, n_seqs] - Key (pre-permuted)
struct ggml_tensor * v, // [S_v, n_tokens, H_v, n_seqs] - Value (pre-permuted)
struct ggml_tensor * g, // [n_tokens, 1, H_k, n_seqs] - Gate logits (pre-permuted)
struct ggml_tensor * beta, // [1, n_tokens, H_k, n_seqs] - Beta (pre-permuted)
struct ggml_tensor * state); // [S_v, S_v*H_v, 1, n_seqs] - Recurrent state

// custom operators

typedef void (*ggml_custom1_op_t)(struct ggml_tensor * dst , const struct ggml_tensor * a, int ith, int nth, void * userdata);
Expand Down
Loading