diff --git a/benchmarks/yaml/deepseek-32k-tp8-wint4.yaml b/benchmarks/yaml/deepseek-32k-tp8-wint4.yaml
new file mode 100644
index 000000000..421c8e34d
--- /dev/null
+++ b/benchmarks/yaml/deepseek-32k-tp8-wint4.yaml
@@ -0,0 +1,9 @@
+quantization: wint4
+load_choices: "default_v1"
+graph_optimization_config:
+  use_cudagraph: True
+  use_unique_memory_pool: True
+no_enable_prefix_caching: True
+max_num_seqs: 256
+max_model_len: 32768
+tensor_parallel_size: 8
diff --git a/benchmarks/yaml/request_yaml/deepseek-32k.yaml b/benchmarks/yaml/request_yaml/deepseek-32k.yaml
new file mode 100644
index 000000000..12d1198a6
--- /dev/null
+++ b/benchmarks/yaml/request_yaml/deepseek-32k.yaml
@@ -0,0 +1,10 @@
+temperature: 0.8
+top_p: 0.8
+presence_penalty: 0
+repetition_penalty: 1.0
+frequency_penalty: 0
+max_tokens: 12288
+metadata:
+  min_tokens: 1
+chat_template_kwargs:
+  enable_thinking: false