diff --git a/composte-hermes.yml b/composte-hermes.yml
index 6f23054..c68ca98 100644
--- a/composte-hermes.yml
+++ b/composte-hermes.yml
@@ -11,6 +11,7 @@ services:
       - NVIDIA_VISIBLE_DEVICES=0,1
       - HUGGING_FACE_HUB_TOKEN=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
       - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
+      - VLLM_ALLOW_LONG_MAX_MODEL_LEN=1
     volumes:
       - /opt/vllm/huggingface:/root/.cache/huggingface
     deploy:
@@ -25,6 +26,7 @@ services:
        - --tensor-parallel-size=2
        - --gpu-memory-utilization=0.90
        - --max-model-len=65536
+       - --hf-overrides={"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":32768}}
        - --enable-prefix-caching
        - --enable-auto-tool-choice
        - --tool-call-parser=hermes