vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-07-09 17:08:33 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

cuda: enable topk-moe fusion for 288 experts (#25267)

Browse source 
* cuda: enable topk-moe fusion for 288 experts

The topk-moe fusion only accepted power-of-2 expert counts (or the
special-cased 576), so models with 288 experts (e.g. Step-3.7-Flash)
fell back to the unfused per-layer routing chain: softmax/sigmoid,
argsort, get_rows, sum_rows, div, clamp, scale. At batch size 1 that
is ~330 extra tiny graph nodes per token.

288 is a multiple of the warp size, so the existing kernel already
handles it; this adds the missing template instantiation and accepts
288 in the eligibility check.

Measured on gfx1151 with Step-3.7-Flash IQ4_XS (llama-bench,
-b 4096 -ub 4096 -fa 1 -dio 1 -ctk q8_0 -ctv q8_0; machine idle,
before/after paired so pp4096 stays matched as a load control):

  test            | before         | after
  ----------------+----------------+----------------
  pp4096          | 460.99 ± 0.45  | 462.47 ± 0.34   (unchanged)
  tg128           |  19.10 ± 0.04  |  19.56 ± 0.03   (+2.4%)
  tg128 @ d30000  |  12.68 ± 0.04  |  12.69 ± 0.03   (unchanged)

Prompt processing is unaffected (the fusion only touches decode
routing). The decode gain is ~+2.4% at shallow context and fades with
depth: by 30k tokens each step is attention-bound over the KV cache,
so removing the fixed routing overhead is no longer visible.

Assisted-By: Claude Fable 5 <noreply@anthropic.com>

* Update tests/test-backend-ops.cpp

Co-authored-by: Oliver Simons <osimons@nvidia.com>

* Add comment for case 288 in topk-moe.cu

---------

Co-authored-by: Oliver Simons <osimons@nvidia.com>
This commit is contained in:
Piotr Wilkin (ilintar) 2026-07-03 15:36:55 +02:00 • committed by GitHub
parent 067de93718
commit 75a48a9055
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 8 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml/src/ggml-cuda/topk-moe.cu
View file

@ -312,6 +312,10 @@ static void launch_topk_moe_cuda(ggml_backend_cuda_context & ctx,
ggml_cuda_kernel_launch(topk_moe_cuda<256, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 288: // StepFun 3.7
ggml_cuda_kernel_launch(topk_moe_cuda<288, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
break;
case 512:
ggml_cuda_kernel_launch(topk_moe_cuda<512, has_bias>, launch_params,
logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
@ -377,8 +381,10 @@ bool ggml_cuda_should_use_topk_moe(const ggml_tensor * gating_op,
const ggml_tensor * weights,
const ggml_tensor * logits,
const ggml_tensor * ids) {
// must match an instantiation of launch_topk_moe_cuda: a power of 2 up to 512,
// or one of the non-power-of-2 expert counts of supported models
const int n_expert = ids->nb[1] / ids->nb[0];
if (((n_expert & (n_expert - 1)) != 0 || n_expert > 512) && n_expert != 576) {
if (((n_expert & (n_expert - 1)) != 0 || n_expert > 512) && n_expert != 288 && n_expert != 576) {
return false;
}

1
tests/test-backend-ops.cpp
View file

@ -9219,6 +9219,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_topk_moe({128, 1, 1, 1}, 128, with_norm, bias_probs, gate, scale_w));
test_cases.emplace_back(new test_topk_moe({129, 1, 1, 1}, 128, with_norm, bias_probs, gate, scale_w));
test_cases.emplace_back(new test_topk_moe({160, 4, 1, 1}, 160, with_norm, bias_probs, gate, scale_w));
test_cases.emplace_back(new test_topk_moe({288, 22, 1, 1}, 8, with_norm, bias_probs, gate, scale_w)); // Used by StepFun 3.7
}
}
}