qqtang/sskj
1
2
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(paths): derive repository files from root

Browse Source
This commit is contained in:
Quantong Qiu 2026-07-20 17:31:16 +08:00
parent 580b6c1c56
commit 359df6a7b6
10 changed files with 20 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
README.md
View File

@ -1,10 +1,13 @@
# sskj — 多平台大模型推理性能基准测试项目
> **更新（2026-07-20 17:25:40 +0800）**
> - RTX 6000D 的 SGLang DSV4-Flash 默认传入 `--cuda-graph-max-bs-decode 16`，避免 SM120 sparse MLA 在大 batch CUDA Graph 预捕获时 OOM。该参数仅限制预捕获图的 decode batch 上限，不限制服务的 `max-running-requests` 或 benchmark 并发；其他平台默认不受影响。
> - H20、RTX 6000D、P800 的配置不再硬编码 sskj 仓库绝对路径：仓库内的环境、数据集、缓存和 P800 patch 均通过共享的 `${ROOT_DIR}` 推导，仓库迁移后无需逐个修改路径。
> **更新（2026-07-20，PR 自动化测试）**
> - 仅用于验证 Gitee 受保护分支的自动 Pull Request 创建与评审流程；不修改实验代码、配置或结果口径。
> **更新（2026-07-20，`903084c`）**
> - RTX 6000D 的 SGLang DSV4-Flash 默认传入 `--cuda-graph-max-bs-decode 16`，避免 SM120 sparse MLA 在大 batch CUDA Graph 预捕获时 OOM。该参数仅限制预捕获图的 decode batch 上限，不限制服务的 `max-running-requests` 或 benchmark 并发；其他平台默认不受影响。
> - H20 / RTX 6000D 的 TPxDP matrix baseline 不再显式限制模型上下文或服务内并发：vLLM 使用默认 `max-model-len` / `max-num-seqs`，SGLang 使用默认 `context-length` / `max-running-requests`。
> - SGLang 固定使用 `--moe-runner-backend marlin`；vLLM 保持框架默认 MoE backend。
> - `run_adaptive_concurrency_add16.sh` 在首点 OOM 时会重启服务并按 `C=16 -> 8 -> 1` 回退；仅 H20 / RTX 6000D matrix 默认开启，H200 / P800 保持原行为。

2
experiments/h20/dsv4_dspark_h20_sglang_tp_dp_matrix/config.env
View File

@ -45,7 +45,7 @@ DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
# Dataset used by sglang.bench_serving --dataset-name random.
# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
# downloading from HuggingFace, which usually fails on offline H20 nodes.
DATASET_PATH="${DATASET_PATH:-/data1/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

2
experiments/h20/dsv4_dspark_h20_vllm_tp_dp_matrix/config.env
View File

@ -68,7 +68,7 @@ DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-lmsysorg/sglang:latest}"
# Dataset used by sglang.bench_serving --dataset-name random.
# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
# downloading from HuggingFace, which usually fails on offline H20 nodes.
DATASET_PATH="${DATASET_PATH:-/data1/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

4
experiments/h20/dsv4_h20_sglang_tp_dp_matrix/config.env
View File

@ -14,7 +14,7 @@ SGLANG_PORT="${SGLANG_PORT:-30031}"
# Python interpreter for orchestration scripts (parse_backend.py, compare.py, etc.)
# and the benchmark client. Defaults to the system python3 if the sglang venv
# does not exist on the host.
VENV_CLIENT="${VENV_CLIENT:-/data1/yy/sskj/envs/sglang}"
VENV_CLIENT="${VENV_CLIENT:-${ROOT_DIR}/envs/sglang}"
# Run the benchmark client natively (0) or inside Docker (1).
USE_DOCKER_CLIENT="${USE_DOCKER_CLIENT:-1}"
@ -44,7 +44,7 @@ DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
# Dataset used by sglang.bench_serving --dataset-name random.
# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
# downloading from HuggingFace, which usually fails on offline H20 nodes.
DATASET_PATH="${DATASET_PATH:-/data1/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

4
experiments/h20/dsv4_h20_vllm_tp_dp_matrix/config.env
View File

@ -14,7 +14,7 @@ VLLM_PORT="${VLLM_PORT:-30030}"
# Python interpreter for orchestration scripts (parse_backend.py, compare.py, etc.)
# and the benchmark client. Defaults to the system python3 if the sglang venv
# does not exist on the host.
VENV_CLIENT="${VENV_CLIENT:-/data1/yy/sskj/envs/sglang}"
VENV_CLIENT="${VENV_CLIENT:-${ROOT_DIR}/envs/sglang}"
# Run the benchmark client natively (0) or inside Docker (1).
USE_DOCKER_CLIENT="${USE_DOCKER_CLIENT:-1}"
@ -50,7 +50,7 @@ DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-lmsysorg/sglang:latest}"
# Dataset used by sglang.bench_serving --dataset-name random.
# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
# downloading from HuggingFace, which usually fails on offline H20 nodes.
DATASET_PATH="${DATASET_PATH:-/data1/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

4
experiments/p800/dsv4_p800_sglang_tp_dp_matrix/config.env
View File

@ -59,8 +59,8 @@ DOCKER_IMAGE="${DOCKER_IMAGE:-iregistry.baidu-int.com/xpu/sglang-p800-pd-disagg-
# copy (also used by the adaptive search for capacity checks); it is mounted
# into the container at CONTAINER_DATASET_PATH, which is what the benchmark
# client references. Default matches the H20 experiments; the file is
# gitignored, download it into /data1/yy/sskj/datasets/ first.
DATASET_PATH="${DATASET_PATH:-/data1/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
# gitignored, download it into ${ROOT_DIR}/datasets/ first.
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
CONTAINER_DATASET_PATH="${CONTAINER_DATASET_PATH:-/workspace/ShareGPT_V3_unfiltered_cleaned_split.json}"
# Matrix and concurrency rules are defined in matrix.json by default.

4
experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/config.env
View File

@ -48,9 +48,9 @@ DOCKER_IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}
# To use ShareGPT, set BENCH_DATASET_NAME=random and DATASET_PATH explicitly.
BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random}"
DATASET_PATH="${DATASET_PATH:-/data/yy/sskj/dataset/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/dataset/ShareGPT_V3_unfiltered_cleaned_split.json}"
SGLANG_BENCH_MODULE="${SGLANG_BENCH_MODULE:-sglang.benchmark.serving}"
CACHE_DIR="${CACHE_DIR:-/data/yy/sskj/sglang_sm120_cache}"
CACHE_DIR="${CACHE_DIR:-${ROOT_DIR}/sglang_sm120_cache}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

4
experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/config.env
View File

@ -49,9 +49,9 @@ DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6
# To use ShareGPT, set BENCH_DATASET_NAME=random and DATASET_PATH explicitly.
BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random}"
DATASET_PATH="${DATASET_PATH:-/data/yy/sskj/dataset/ShareGPT_V3_unfiltered_cleaned_split.json}"
DATASET_PATH="${DATASET_PATH:-${ROOT_DIR}/dataset/ShareGPT_V3_unfiltered_cleaned_split.json}"
SGLANG_BENCH_MODULE="${SGLANG_BENCH_MODULE:-sglang.benchmark.serving}"
CACHE_DIR="${CACHE_DIR:-/data/yy/sskj/vllm_sm120_cache}"
CACHE_DIR="${CACHE_DIR:-${ROOT_DIR}/vllm_sm120_cache}"
# Matrix and concurrency rules are defined in matrix.json by default.
MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

2
platforms/kunlun_p800.env
View File

@ -24,4 +24,4 @@ CONTAINER_PYTHON="/root/miniconda/envs/python310_torch25_cuda/bin/python"
# Host mount points for patches required by the P800 SGLang image
# These are baked into the container start command in server_docker.sh.
PATCH_ROOT="${PATCH_ROOT:-/data1/yy/sskj/platforms/patches/kunlun_p800}"
PATCH_ROOT="${PATCH_ROOT:-${ROOT_DIR}/platforms/patches/kunlun_p800}"

6
platforms/nvidia_h20.env
View File

@ -16,9 +16,9 @@ MODEL_ROOT="/data1/hf_models"
# Virtual environments on the host (used by native H20 scripts).
# Override these if your H20 machine uses different paths.
VENV_VLLM="${VENV_VLLM:-/data1/yy/sskj/envs/vllm}"
VENV_SGLANG="${VENV_SGLANG:-/data1/yy/sskj/envs/sglang}"
VENV_CLIENT="${VENV_CLIENT:-/data1/yy/sskj/envs/sglang}"
VENV_VLLM="${VENV_VLLM:-${ROOT_DIR}/envs/vllm}"
VENV_SGLANG="${VENV_SGLANG:-${ROOT_DIR}/envs/sglang}"
VENV_CLIENT="${VENV_CLIENT:-${ROOT_DIR}/envs/sglang}"
# Default server start script for the legacy benchmark grid.
SERVER_START_SCRIPT="${SERVER_START_SCRIPT:-${ROOT_DIR}/scripts/start_dsv4_dspark_8card.sh}"