From 5156d5a413517d9ecb17c877aa36764ad2673375 Mon Sep 17 00:00:00 2001
From: Baizhou Zhang <sobereddiezhang@gmail.com>
Date: Sun, 20 Apr 2025 17:28:52 -0700
Subject: [PATCH] Add test config yamls for Deepseek v3 (#5433)

---
 test/srt/configs/deepseek_v3.yaml             | 28 +++++++++++++++++++
 .../srt/configs/deepseek_v3_long_context.yaml | 28 +++++++++++++++++++
 test/srt/parse_results.py                     |  1 +
 3 files changed, 57 insertions(+)
 create mode 100644 test/srt/configs/deepseek_v3.yaml
 create mode 100644 test/srt/configs/deepseek_v3_long_context.yaml

diff --git a/test/srt/configs/deepseek_v3.yaml b/test/srt/configs/deepseek_v3.yaml
new file mode 100644
index 000000000..82d059cba
--- /dev/null
+++ b/test/srt/configs/deepseek_v3.yaml
@@ -0,0 +1,28 @@
+tasks:
+  - name: sglang-8192-1024-concurrency1
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 1 --num-prompts 5 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency2
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 2 --num-prompts 10 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency4
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 4 --num-prompts 20 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency8
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 8 --num-prompts 32 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency16
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 16 --num-prompts 48 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency24
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 24 --num-prompts 72 --output-file deepseek_v3_results.jsonl
+
+  - name: sglang-8192-1024-concurrency32
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 8192 --random-output-len 1024 --max-concurrency 32 --num-prompts 96 --output-file deepseek_v3_results.jsonl
diff --git a/test/srt/configs/deepseek_v3_long_context.yaml b/test/srt/configs/deepseek_v3_long_context.yaml
new file mode 100644
index 000000000..df416a429
--- /dev/null
+++ b/test/srt/configs/deepseek_v3_long_context.yaml
@@ -0,0 +1,28 @@
+tasks:
+  - name: sglang-32000-100-concurrency1
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 1 --num-prompts 5 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency2
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 2 --num-prompts 10 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency4
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 4 --num-prompts 20 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency8
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 8 --num-prompts 32 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency16
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 16 --num-prompts 48 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency24
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 24 --num-prompts 72 --output-file deepseek_v3_long_context_results.jsonl
+
+  - name: sglang-32000-100-concurrency32
+    server_cmd: python3 -m sglang.launch_server --model deepseek-ai/DeepSeek-V3-0324 --tp 8 --trust-remote-code --disable-radix-cache --max-prefill-tokens 32768
+    client_cmd: python3 -m sglang.bench_serving --dataset-name random --random-range-ratio 1 --random-input-len 32000 --random-output-len 100 --max-concurrency 32 --num-prompts 96 --output-file deepseek_v3_long_context_results.jsonl
diff --git a/test/srt/parse_results.py b/test/srt/parse_results.py
index de1d5cf27..e6ff16a51 100644
--- a/test/srt/parse_results.py
+++ b/test/srt/parse_results.py
@@ -16,6 +16,7 @@ output_file = f"{base_name}_summary.csv"
 
 fields = [
     "max_concurrency",
+    "input_throughput",
     "output_throughput",
     "mean_ttft_ms",
     "median_ttft_ms",