From 48f65f8a4912952fe0cdd0e6c5b6d93742556321 Mon Sep 17 00:00:00 2001 From: Drejfus Date: Thu, 6 Aug 2026 13:06:46 +0200 Subject: [PATCH] docs: remove obsolete Eagle3 prefix caching limitation --- demos/continuous_batching/speculative_decoding/README.md | 1 - 1 file changed, 1 deletion(-) diff --git a/demos/continuous_batching/speculative_decoding/README.md b/demos/continuous_batching/speculative_decoding/README.md index c5633bdd08..56d5e8f89b 100644 --- a/demos/continuous_batching/speculative_decoding/README.md +++ b/demos/continuous_batching/speculative_decoding/README.md @@ -172,7 +172,6 @@ The built-in fallback is `5`. The same applies to `assistant_confidence_threshol Eagle3 deployments currently have following known limitations: - stateful mode (pipeline_type: LM) not supported, - concurrency not supported - max 1 request can be processed at a time (**ALWAYS** use rest_workers=2 when deploying Eagle3 pipeline), -- prefix caching not supported, - only greedy sampling is supported (enforced by OVMS if pipeline configured properly), - MoE models not supported