models : remove redundant ggml_cont calls now supported by all unary kernels (#30198)

* models: remove redundant ggml_cont calls now supported by CUDA unary kernels

* models: restore cont_2d after sigmoid for broadcast mul, clean dflash cont

* Update src/models/qwen3next.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>

* models: restore cont_2d before sigmoid in qwen3next

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
This commit is contained in:
Muhammad SaadandSigbjørn Skjæret authored and GitHub committed 2026-10-13 09:26:52 +02:00
1 parent 4413a19241
commit 071b553fb1
3 files changed
+1 -9

No files matched your search

+1 -1
View File
@@ -337,7 +337,7 @@ static void build_dspark_markov_head(llm_graph_context & g, const llama_model &
// bonus anchor slot: pass the logits through unbiased, pad the (unread) confidence column
cat = ggml_cont(ctx0, ggml_view_2d(ctx0, base, n_vocab, n_blocks, base_stride, 0));
if (has_conf) {
cat_conf = ggml_sigmoid(ctx0, ggml_cont(ctx0, ggml_view_2d(ctx0, base, 1, n_blocks, base_stride, 0)));
cat_conf = ggml_sigmoid(ctx0, ggml_view_2d(ctx0, base, 1, n_blocks, base_stride, 0));
}
}