Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

llama : enable chunked fused GDN path (#20340)

Browse Source 
* llama : enable chunked fused GDN path

* models : avoid Q and K repeats when using fused GDA

* cont : fix comment

Co-authored-by: Aman Gupta <amangupta052@gmail.com>

* cont : fix the fix

Co-authored-by: Aman Gupta <amangupta052@gmail.com>

* cont : fix

* metal : add GDN kernel (#20361)

* metal : add Metal backend for GGML_OP_GATED_DELTA_NET

Add a fused Metal kernel for the gated delta net recurrence op
(#19504), enabling GPU-accelerated inference for DeltaNet-based
models (Qwen3.5, etc.) on Apple Silicon.

Supports both GDA (scalar gate) and KDA (per-row gate) modes
with head_size 64 and 128. Unsupported configurations (head_size
32, non-contiguous tensors) gracefully fall back to CPU.

Performance: Qwen3.5-0.8B Q4_K_M on M4 Max
  tg128: 170 -> 213 t/s (+25%)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* metal : validate contiguity of all input tensors in supports_op

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* metal : add algorithm equivalence comment for GDA decay path

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* cont : unslop + optimize

* cont : clean-up

---------

Co-authored-by: Paul Flynn <paul@arkavo.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

* CUDA: AR gated delta net improvements (#20391)

* Add FastDiv to gated_delta_net_cuda

* Shard columns across warps

This reduces register pressure (avoids spill for S_v = 128) and gives
the warp-scheduler more CTAs to schedule (thus hiding data-access
latencies).

* Remove unneded include in gated_delta_net.cu

* Improve comments

* Apply code-formating

* Make sharding HIP-compatible

1. Use ggml_cuda_get_physical_warp_size() to determine warp size flexibly
2. Add test with partial warp to test sum reduction on CUDA

* Remove fastdiv_s64, as we can treat neqk1 and rq3 as uint32_t

* Rename variables

* Enable GDN also for prefill, move TODO for chunked_GDN

* Actually remove the TODO from 206890897546bd16602c3b79394fd5ea09ef199f

* Get warp size at runtime

warp_size is not known at compile time in hip host code.

* Don't expose ggml_cuda_get_physical_warp_size on host

---------

Co-authored-by: uvos <devnull@uvos.xyz>

* llama : refactor llm_build_delta_net_base API

---------

Co-authored-by: Aman Gupta <amangupta052@gmail.com>
Co-authored-by: Paul Flynn <paul@arkavo.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Oliver Simons <osimons@nvidia.com>
Co-authored-by: uvos <devnull@uvos.xyz>
This commit is contained in:
Georgi Gerganov
2026-03-11 22:46:40 +02:00
committed by GitHub
co-authored by uvos Aman Gupta Paul Flynn Claude Opus 4.6 Oliver Simons
parent f90bd1dd84
commit d28961d81e
20 changed files with 674 additions and 165 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/ops.cpp
+5 -6
View File
@@ -10443,8 +10443,8 @@ static void ggml_compute_forward_gated_delta_net_one_chunk(
const float * state_in_base = (const float *)src_state->data;
const int64_t rq1 = nev1 / neq1;
const int64_t rk1 = nev1 / nek1;
//const int64_t rq1 = nev1 / neq1;
//const int64_t rk1 = nev1 / nek1;
const int64_t rq3 = nev3 / neq3;
const int64_t rk3 = nev3 / nek3;
@@ -10454,8 +10454,8 @@ static void ggml_compute_forward_gated_delta_net_one_chunk(
const int64_t iv1 = ir % H; // head_index
const int64_t iv3 = ir / H; // sequence
const int64_t iq1 = iv1 / rq1;
const int64_t ik1 = iv1 / rk1;
const int64_t iq1 = iv1 % neq1;
const int64_t ik1 = iv1 % nek1;
const int64_t iq3 = iv3 / rq3;
const int64_t ik3 = iv3 / rk3;
@@ -10475,7 +10475,7 @@ static void ggml_compute_forward_gated_delta_net_one_chunk(
const float * v_d = (const float *)((const char *)src_v->data + iv3 * nbv3 + t * nbv2 + iv1 * nbv1);
const float beta_val = *(const float *)((const char *)src_beta->data + iv3 * nbb3 + t * nbb2 + iv1 * nbb1);
const float * g_d = (const float *)((const char *)src_g->data + iv3 * nbg3 + t * nbg2 + iv1 * nbg1);
const float * g_d = (const float *)((const char *)src_g->data + iv3 * nbg3 + t * nbg2 + iv1 * nbg1);
if (kda) {
for (int64_t i = 0; i < S_v; ++i) {
@@ -10508,7 +10508,6 @@ static void ggml_compute_forward_gated_delta_net_one_chunk(
attn_data += S_v * H; // advance to next token
}
}
}