model: support vllm decision 2 model (#30378)

* models: support pplx-decider

* vllm d2

* update

* clarify naming rule

* stricter validation

* fix ci

* rm redundant filter_tensors
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-10-12 13:39:55 +02:00
1 parent ccdbe29053
commit 097cf33d94
23 files changed
+549 -19

No files matched your search

+5
View File
@@ -3604,6 +3604,11 @@ private:
n_past = std::min(n_past, slot.task->n_tokens_shared);
}
// the outputs of a decision are read from the batch, so its first position must be evaluated
if (n_decision_first >= 0) {
n_past = std::min(n_past, n_decision_first);
}
// if there is an alora invoked, don't cache after the invocation start
if (slot.alora_invocation_start > 0) {
SLT_DBG(slot, "only caching to alora invocation start (n_past = %d, alora_invocation_start = %d)\n", n_past, slot.alora_invocation_start);