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:
1 parent
a97cce86a8
commit
4da6337767
5 files changed
+32
-8
No files matched your search
+1
-1
@@ -297,7 +297,7 @@ void llm_graph_input_cls::set_input(const llama_ubatch * ubatch) {
|
||||
|
||||
const bool last = (
|
||||
cparams.pooling_type == LLAMA_POOLING_TYPE_LAST ||
|
||||
(cparams.pooling_type == LLAMA_POOLING_TYPE_RANK && (arch == LLM_ARCH_QWEN3 || arch == LLM_ARCH_QWEN3VL)) // qwen3 reranking & embedding models use last token
|
||||
(cparams.pooling_type == LLAMA_POOLING_TYPE_RANK && cparams.causal_attn)
|
||||
);
|
||||
|
||||
for (int i = 0; i < n_tokens; ++i) {
|
||||
|
||||
Reference in new issue
Block a user