new build

Dockerfile

@@ -45,5 +45,8 @@ RUN apt-get update && apt-get install -y git \
 # Make sure we have the latest up to date chat template
 COPY glm_5.1_chat_template.jinja /opt/chat_template.jinja
 
-# Working LMCache config
-COPY lmcache-config.yaml /opt/lmcache-config.yaml
+# GLM 5.1 LMCache config
+COPY lmcache-config-glm-51.yaml /opt/lmcache-config-glm-51.yaml
+
+# DEEPSEEK v4 LMCache config
+COPY lmcache-config-dsv4.yaml /opt/lmcache-config-dsv4.yaml

lmcache-config-dsv4.yaml

@@ -0,0 +1,10 @@
+chunk_size: 256
+local_cpu: true
+max_local_cpu_size: 20.0
+remote_url: "redis://10.66.0.100:6379"
+remote_serde: naive
+save_decode_cache: true
+use_layerwise: true
+save_unfull_chunk: true
+blocking_timeout_secs: 60
+cache_policy: LRU

lmcache-config-glm-51.yaml

@@ -5,3 +5,5 @@ save_decode_cache: false
 enable_lazy_memory_allocator: true
 lazy_memory_initial_ratio: 1.0
 use_gpu_connector_v3: true
+remote_url: "redis://10.66.0.100:6379"
+remote_serde: naive