diff --git a/benchmarks/yaml/GLM45-air-32k-bf16.yaml b/benchmarks/yaml/GLM45-air-32k-bf16.yaml
new file mode 100644
index 000000000..b14dce761
--- /dev/null
+++ b/benchmarks/yaml/GLM45-air-32k-bf16.yaml
@@ -0,0 +1,5 @@
+max_model_len: 32768
+max_num_seqs: 128
+tensor_parallel_size: 4
+use_cudagraph: True
+load_choices: "default_v1"
diff --git a/benchmarks/yaml/GLM45-air-32k-wfp8afp8.yaml b/benchmarks/yaml/GLM45-air-32k-wfp8afp8.yaml
new file mode 100644
index 000000000..5e4afe79e
--- /dev/null
+++ b/benchmarks/yaml/GLM45-air-32k-wfp8afp8.yaml
@@ -0,0 +1,6 @@
+max_model_len: 32768
+max_num_seqs: 128
+tensor_parallel_size: 4
+use_cudagraph: True
+load_choices: "default_v1"
+quantization: wfp8afp8
diff --git a/benchmarks/yaml/request_yaml/GLM-32k.yaml b/benchmarks/yaml/request_yaml/GLM-32k.yaml
new file mode 100644
index 000000000..c70bb5af6
--- /dev/null
+++ b/benchmarks/yaml/request_yaml/GLM-32k.yaml
@@ -0,0 +1,8 @@
+top_p: 0.95
+temperature: 0.6
+metadata:
+  min_tokens: 1
+max_tokens: 12288
+repetition_penalty: 1.0
+frequency_penalty: 0
+presence_penalty: 0