server : allow RANK pooling batch splitting for causal LLM rerankers (ie. Qwen3 and Qwen3-VL) (#28876)

* server : allow splitting RANK pooling for causal LLM rerankers

Rerank models fall into two categories: bidirectional cross-encoders
(BERT, etc.) that require all tokens in a single physical batch, and
causal LLMs repurposed as rerankers (Qwen3, Qwen3-VL) that can use
chunked prefill like any other decoder.

Previously the server rejected all RANK-pooling inputs larger than
n_ubatch, and the graph builder hardcoded QWEN3/QWEN3VL arch checks to
determine last-token pooling. This broke long-document and multimodal
reranking for causal models.

Fix: expose llama_get_causal_attn(ctx) so the server can check the
effective runtime attention type (reflecting any --attention override
or set_causal_attn call). Also expose llama_model_is_causal(model)
for querying the static architectural property from GGUF metadata.

can_split() now permits chunked prefill for RANK pooling when the
context is causal. The graph builder's inline arch check is replaced
with the same cparams.causal_attn predicate, removing the duplication.

Assisted-by: Opencode/Qwen3.8-27B

* remove unused llama_model_is_causal, fix whitespace

Assisted-by: opencode

---------

Co-authored-by: timothywang21 <timothywang21@users.noreply.github.com>
This commit is contained in:
Tim Wangandtimothywang21 authored and GitHub committed 2026-09-27 23:28:10 +02:00
1 parent a97cce86a8
commit 4da6337767
5 files changed
+32 -8

No files matched your search

+8
View File
@@ -1259,6 +1259,10 @@ void llama_context::set_causal_attn(bool value) {
sched_need_reserve = true;
}
bool llama_context::get_causal_attn() const {
return cparams.causal_attn;
}
void llama_context::set_warmup(bool value) {
LLAMA_LOG_DEBUG("%s: value = %d\n", __func__, value);
@@ -3933,6 +3937,10 @@ void llama_set_causal_attn(llama_context * ctx, bool causal_attn) {
ctx->set_causal_attn(causal_attn);
}
bool llama_get_causal_attn(const llama_context * ctx) {
return ctx->get_causal_attn();
}
void llama_set_warmup(llama_context * ctx, bool warmup) {
ctx->set_warmup(warmup);
}