{ "schema_version": 1, "targets": [ { "id": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）", "url": "http://127.0.0.1:18100", "model": "deepseek-v4.1-flash-1017", "environment": { "CT_OFF": "{\"thinking\": false}", "CT_ON": "{\"thinking\": true}", "MAX_LEN": "1048576" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env" ], "verify_model": true }, { "id": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）", "url": "http://127.0.0.1:18112", "model": "qwen3.8-27b", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "verify_model": true }, { "id": "dsv41cap", "label": "dsv41cap（配置端口 18110）", "config_name": "dsv41cap", "url": "http://127.0.0.1:18110", "model": "deepseek-v4.1-flash", "environment": { "CT_OFF": "{\"thinking\": false}", "CT_ON": "{\"thinking\": true}", "MAX_LEN": "1048576" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/dsv41cap.conf" }, { "id": "glm53", "label": "glm53（配置端口 18102）", "config_name": "glm53", "url": "http://127.0.0.1:18102", "model": "glm-5.3", "environment": { "CT_OFF": "{\"reasoning_effort\": \"low\"}", "CT_ON": "{}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/glm53.conf" }, { "id": "mimo", "label": "mimo（配置端口 18104）", "config_name": "mimo", "url": "http://127.0.0.1:18104", "model": "mimo-v2.6-pro", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/mimo.conf" }, { "id": "qwen27b", "label": "qwen27b（配置端口 18103）", "config_name": "qwen27b", "url": "http://127.0.0.1:18103", "model": "qwen3.5-27b", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen27b.conf" }, { "id": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）", "config_name": "qwen27btp8", "url": "http://127.0.0.1:18103", "model": "qwen3.5-27b", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen27btp8.conf" }, { "id": "qwen38cap", "label": "qwen38cap（配置端口 18107）", "config_name": "qwen38cap", "url": "http://127.0.0.1:18107", "model": "qwen3.8-27b", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38cap.conf" }, { "id": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）", "config_name": "qwen38fp8", "url": "http://127.0.0.1:18105", "model": "qwen3.8-27b-fp8", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8.conf" }, { "id": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）", "config_name": "qwen38fp8cap", "url": "http://127.0.0.1:18108", "model": "qwen3.8-27b-fp8", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8cap.conf" }, { "id": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）", "config_name": "qwen38fp8mtpcap", "url": "http://127.0.0.1:18109", "model": "qwen3.8-27b-fp8", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8mtpcap.conf" }, { "id": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）", "config_name": "qwen38fp8tp8", "url": "http://127.0.0.1:18110", "model": "qwen3.8-27b-fp8", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8tp8.conf" }, { "id": "sglang38", "label": "sglang38（配置端口 18106）", "config_name": "sglang38", "url": "http://127.0.0.1:18106", "model": "qwen3.8-27b", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/sglang38.conf" }, { "id": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）", "config_name": "sglang38fp8", "url": "http://127.0.0.1:18106", "model": "qwen3.8-27b-fp8", "environment": { "CT_OFF": "{\"enable_thinking\": false}", "CT_ON": "{\"enable_thinking\": true}", "MAX_LEN": "262144" }, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "config_file": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/sglang38fp8.conf" } ], "profiles": [ { "id": "gpu-suite", "source_group": "company-gpu-suite", "entry": "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/xas-gpu-test.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0", "arguments": [], "parameters": [], "environment": {}, "credential_files": [], "impact": { "level": "service", "title": "进入 GPU 验收原菜单，可能影响整机与模型服务", "details": [ "菜单包含 GPU 压力、驱动/环境安装、模型启停、固件切换和重启相关操作；按所选菜单继续确认影响。", "原入口会初始化既有数据目录，并可能通过 tmux 创建持久会话；取消控制台不证明菜单产生的容器或 tmux 会话已结束。" ], "requires_confirmation": true }, "interactive": true, "timeout_seconds": 259200, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/nvidia-smi", "/usr/bin/docker", "/usr/bin/tmux" ], "required_files": [ "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/xas-gpu-test.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/report.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/stress_report.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/detail_export.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/stress_metrics.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/db.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/html_to_pdf.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/ai_bench.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/model_catalog.json", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/report_templates.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/report_template_menu.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/gpu_compat.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/ai_runtime.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/ai_environment.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/ai_environment.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/model_manager.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/model_discovery.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/model_downloads.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/llm_planner.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/llm_batch.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/llm_model_environment.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/llm_report.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/llm_menu.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/token_economics.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/token_prices.json", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/ai_menu.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/display_menu.sh", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/bandwidth_audit.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/bandwidth_ab.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/display_switch.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/display_tool.py", "/XAS/SH/XAS-GPU-Test/XAS-GPU-Test-v3.5.0/display_probe.py", "/etc/xas-gpu-test/config.env" ], "check_ports": [], "output_paths": [ "/XAS/GPUTest/runs", "/XAS/GPUTest/reports" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "原 config.env 的 DATA_ROOT=/XAS/GPUTest。只打开原菜单，不自动输入或选择任何测试项目。", "原工具创建的 tmux 持久任务可能脱离本次控制台 cgroup；平台取消后必须核查原工具会话和模型容器，不能承诺全部停止。", "runs/reports 使用原工具的时间戳产物；只收集本次运行新增或发生变化的文件，不预拷整个历史 runs/reports。", "2026-10-20：服务器最高已安装版本仍为 v3.5.0，已登记的 224 个源码文件摘要未变；当前 GPU 0–14 已分配给模型/图像/语音服务。" ], "available": true }, { "id": "deepseek-suite", "source_group": "dsv41-trial-20261010", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/run-b-suite.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial", "arguments": [ "{run_label}", "{mode}" ], "parameters": [ { "key": "mode", "label": "原测试模式", "type": "select", "default": "full", "options": [ { "value": "full", "label": "完整（原默认）" }, { "value": "quick", "label": "快速" } ], "required": true } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 14400, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/run-b-suite.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/spec_prompts.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/multiturn_cache.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/toolcall_stress.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/curves.py" ], "check_ports": [ 18101 ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/logs/test-{run_label}-*" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "原入口硬编码 B 服务 18101，不会切到 A 的 18100；B 未运行时预检应阻止启动。" ], "available": true }, { "id": "model-suite", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-suite.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_config}", "{run_label}", "{mode}" ], "parameters": [ { "key": "target", "label": "原服务配置", "type": "select", "default": "sglang38fp8", "options": [ { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "mode", "label": "原测试模式", "type": "select", "default": "full", "options": [ { "value": "full", "label": "完整（原默认）" }, { "value": "quick", "label": "快速" }, { "value": "load", "label": "仅并发负载" } ], "required": true } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 14400, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-suite.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/{target_config}.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_prompts.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/multiturn_cache.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/toolcall_stress.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/functional_check.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/needle.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/curves.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/{target_config}-{run_label}-*" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "Shell 继续读取 configs/{target_config}.conf 的 SERVED、PORT、GPUS、CT_OFF、CT_ON 等原字段；控制台不覆盖这些配置。" ], "available": true }, { "id": "functional", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/functional_check.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/functional_check.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。" ], "available": true }, { "id": "curves", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/curves.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{mode}", "{output_file}", "{lengths}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "mode", "label": "曲线类型", "type": "select", "default": "context", "options": [ { "value": "context", "label": "输入长度曲线" }, { "value": "output", "label": "输出长度曲线" } ], "required": true }, { "key": "lengths", "label": "输入长度档位", "type": "select", "default": "1000,8000,32000,128000,256000", "options": [ { "value": "1000,8000,32000,128000,256000", "label": "原默认 1千 / 8千 / 3.2万 / 12.8万 / 25.6万" }, { "value": "1000,8000,32000", "label": "1千 / 8千 / 3.2万" }, { "value": "32000", "label": "3.2万" } ], "required": true, "help": "输出曲线固定生成 8192 token；该模式忽略输入长度档位。" } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/curves.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "context 长度默认与原脚本一致；超过服务上下文上限会按原脚本返回错误。" ], "available": true }, { "id": "language", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/qwen38fp8tp8-01-prompts.json", "{concurrency}", "{requests}", "{output_file}", "--max-tokens", "{max_tokens}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "concurrency", "label": "并发数", "type": "integer", "default": 1, "min": 1, "max": 512, "required": true }, { "key": "requests", "label": "正式请求数", "type": "integer", "default": 32, "min": 1, "max": 100000, "required": true, "help": "另外有原脚本固定的 4 次预热请求。" }, { "key": "max_tokens", "label": "每次最大生成 token", "type": "integer", "default": 768, "min": 1, "max": 32768, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/qwen38fp8tp8-01-prompts.json" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "复用原 logs/qwen38fp8tp8-01-prompts.json；本轮仅核对路径及文件大小，未读取题目内容。", "保留原温度 0.6、关闭思考、正常采样流程；原 --greedy / --thinking / --logprobs 未在本方案开放。" ], "available": true }, { "id": "conversation-cache", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/multiturn_cache.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/multiturn_cache.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。" ], "available": true }, { "id": "needle", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/needle.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{input_tokens}", "{depth}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "input_tokens", "label": "目标上下文 token", "type": "integer", "default": 32000, "min": 1000, "max": 1000000, "required": true }, { "key": "depth", "label": "目标信息位置比例", "type": "number", "default": 0.5, "min": 0, "max": 1, "required": true }, { "key": "max_tokens", "label": "最大生成 token", "type": "integer", "default": 4096, "min": 1, "max": 32768, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}", "NEEDLE_MAX_TOKENS": "{max_tokens}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/needle.py" ], "check_ports": [ "{target_port}" ], "output_paths": [], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原入口只向标准输出打印 JSON，不接受结果路径；结果保存在本次控制台运行日志。" ], "available": true }, { "id": "concurrent-input", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/concurrent_prefill.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{concurrency}", "{input_tokens}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "concurrency", "label": "并发数", "type": "integer", "default": 4, "min": 1, "max": 256, "required": true }, { "key": "input_tokens", "label": "每请求输入 token", "type": "integer", "default": 32000, "min": 1000, "max": 1000000, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/concurrent_prefill.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 的数字参数为必填；表单初值为控制台建议值，不冒充原脚本默认值。" ], "available": true }, { "id": "cache-revisit", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/kv_revisit.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{sessions}", "{input_tokens}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "sessions", "label": "会话数", "type": "integer", "default": 8, "min": 1, "max": 256, "required": true }, { "key": "input_tokens", "label": "每会话输入 token", "type": "integer", "default": 32000, "min": 1000, "max": 1000000, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/kv_revisit.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 的数字参数为必填；表单初值为控制台建议值，不冒充原脚本默认值。" ], "available": true }, { "id": "interference", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/prefill_interference.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{input_tokens}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "input_tokens", "label": "长输入 token", "type": "integer", "default": 128000, "min": 1000, "max": 1000000, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/prefill_interference.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 的数字参数为必填；表单初值为控制台建议值，不冒充原脚本默认值。" ], "available": true }, { "id": "tool-calls", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/toolcall_stress.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{concurrency}", "{repeats}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "concurrency", "label": "并发数", "type": "integer", "default": 16, "min": 1, "max": 256, "required": true }, { "key": "repeats", "label": "每题重复次数", "type": "integer", "default": 4, "min": 1, "max": 1000, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/toolcall_stress.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 的数字参数为必填；表单初值为控制台建议值，不冒充原脚本默认值。" ], "available": true }, { "id": "throughput", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_config}", "{run_label}-throughput.json", "--dataset-name", "random", "--random-input-len", "{input_tokens}", "--random-output-len", "{output_tokens}", "--random-range-ratio", "{range_ratio}", "--num-prompts", "{requests}", "--max-concurrency", "{concurrency}", "--seed", "{seed}" ], "parameters": [ { "key": "target", "label": "原服务配置", "type": "select", "default": "sglang38fp8", "options": [ { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "input_tokens", "label": "平均输入 token", "type": "integer", "default": 1024, "min": 1, "max": 1000000, "required": true }, { "key": "output_tokens", "label": "平均生成 token", "type": "integer", "default": 512, "min": 1, "max": 32768, "required": true }, { "key": "range_ratio", "label": "随机长度浮动比例", "type": "number", "default": 0, "min": 0, "max": 1, "required": true }, { "key": "requests", "label": "请求数", "type": "integer", "default": 32, "min": 1, "max": 1000000, "required": true }, { "key": "concurrency", "label": "最大并发数", "type": "integer", "default": 4, "min": 1, "max": 2048, "required": true }, { "key": "seed", "label": "随机种子", "type": "integer", "default": 20261013, "min": 0, "max": 2147483647, "required": true } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 14400, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/{target_config}.conf" ], "check_ports": [ "{target_port}" ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/{run_label}-throughput.json" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "Shell 继续读取 configs/{target_config}.conf 的 SERVED、PORT、GPUS、CT_OFF、CT_ON 等原字段；控制台不覆盖这些配置。", "原 Shell 透传的 vllm bench serve 参数限定为上述登记字段；不接受任意额外参数。表单初值是控制台建议值。" ], "available": true }, { "id": "math", "source_group": "dsv41-trial-20261010", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/math_check.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial", "arguments": [ "{concurrency}", "{samples}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "concurrency", "label": "并发数", "type": "integer", "default": 4, "min": 1, "max": 256, "required": true }, { "key": "samples", "label": "每题采样次数", "type": "integer", "default": 1, "min": 1, "max": 1000, "required": true, "help": "原脚本固定生成 32 道题，总请求数为 32 × 采样次数。" } ], "environment": { "PYTHONUNBUFFERED": "1" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/math_check.py" ], "check_ports": [ 18100 ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 的数字参数为必填；表单初值为控制台建议值，不冒充原脚本默认值。", "原脚本硬编码 DeepSeek 模型和 18100 地址；本方案固定目标 A。", "2026-10-20：当前 DeepSeek 服务主名为 deepseek-v4.1-flash-1017；元数据接口仍提供 deepseek-v4.1-flash 兼容别名，原脚本保持不变。" ], "available": true }, { "id": "vision", "source_group": "dsv41-trial-20261010", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/vision_check.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial", "arguments": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/logs/test-34-vision-input.png", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/vision_check.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/logs/test-34-vision-input.png" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "复用原 run-d7v-d8.sh 引用的 test-34-vision-input.png；本轮仅核对文件名及大小，未读取或复制图像内容。", "原入口分别发送关闭/开启 thinking 的两次看图请求；目标服务需已启用视觉。未自动修改或重启模型。", "2026-10-20：当前 DeepSeek 服务主名为 deepseek-v4.1-flash-1017；元数据接口仍提供 deepseek-v4.1-flash 兼容别名，原脚本保持不变。" ], "available": true }, { "id": "robust", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/robust_check.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "storm_n", "label": "断连并发数", "type": "integer", "default": 64, "min": 1, "max": 256, "required": true }, { "key": "storm_rounds", "label": "断连轮数", "type": "integer", "default": 5, "min": 1, "max": 100, "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}", "STORM_N": "{storm_n}", "STORM_ROUNDS": "{storm_rounds}" }, "credential_files": [], "impact": { "level": "resource", "title": "执行断连风暴及超长异常请求", "details": [ "原默认是 64 路并发、5 轮断连，并发送超过目标上下文上限的输入/输出等异常请求，可能短时影响目标服务。", "MAX_LEN 来自登记目标的原启动配置，运行前需核对服务实际配置是否仍一致。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/robust_check.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。" ], "available": true }, { "id": "agent", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/agent_sim.py", "interpreter": "/usr/bin/python3", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_url}", "{target_model}", "{sessions}", "{turns}", "{context_tokens}", "{new_tokens}", "{output_tokens}", "{seed}", "{output_file}" ], "parameters": [ { "key": "target", "label": "目标服务", "type": "select", "default": "deepseek-a", "options": [ { "value": "deepseek-a", "label": "DeepSeek V4.1 Flash 1017（GPU 0–7 · 18100）" }, { "value": "qwen38-g14", "label": "Qwen3.8-27B（GPU 14 · 18112）" }, { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "sessions", "label": "每地址会话数", "type": "integer", "default": 4, "min": 1, "max": 512, "required": true }, { "key": "turns", "label": "每会话轮数", "type": "integer", "default": 4, "min": 1, "max": 1000, "required": true }, { "key": "context_tokens", "label": "固定上下文 token", "type": "integer", "default": 8192, "min": 0, "max": 1000000, "required": true }, { "key": "new_tokens", "label": "每轮新增 token", "type": "integer", "default": 1024, "min": 0, "max": 1000000, "required": true }, { "key": "output_tokens", "label": "每轮生成 token", "type": "integer", "default": 256, "min": 1, "max": 32768, "required": true }, { "key": "seed", "label": "随机种子", "type": "integer", "default": 20261013, "min": 0, "max": 2147483647, "required": true }, { "key": "agent_mode", "label": "上下文增长方式", "type": "select", "default": "replace", "options": [ { "value": "replace", "label": "替换新增段（原默认）" }, { "value": "append", "label": "逐轮追加" } ], "required": true } ], "environment": { "PYTHONUNBUFFERED": "1", "SPEC_URLS": "{target_url}", "MODEL": "{target_model}", "AGENT_MODE": "{agent_mode}" }, "credential_files": [], "impact": { "level": "standard", "title": "向已运行服务发送测试请求", "details": [ "原脚本按既定用例调用目标接口，可能影响该服务的响应时间。" ], "requires_confirmation": false }, "interactive": false, "timeout_seconds": 7200, "required_tools": [ "/usr/bin/python3" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/agent_sim.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "{output_file}" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "通过所选目标的原认证文件在运行时加载 VLLM_API_KEY；不在参数或日志中显示值。", "原 CLI 数字参数全部必填；表单初值是控制台建议值。", "输入使用原脚本生成的随机 token 编号；append 模式每轮增长，超过目标上下文限制会失败。" ], "available": true }, { "id": "replicas", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-replicas.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_config}", "{run_label}", "8" ], "parameters": [ { "key": "target", "label": "原实例配置", "type": "select", "default": "qwen38fp8cap", "options": [ { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" } ], "required": true, "help": "请选择当前 18110–18117 上已有 8 份服务对应的原配置。" } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 14400, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout", "/usr/bin/vmstat" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-replicas.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/{target_config}.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_prompts.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-replicas.sh" ], "check_ports": [ 18110, 18111, 18112, 18113, 18114, 18115, 18116, 18117 ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/{target_config}-{run_label}-*" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "Shell 继续读取 configs/{target_config}.conf 的 SERVED、PORT、GPUS、CT_OFF、CT_ON 等原字段；控制台不覆盖这些配置。", "保留原默认 8 份、GPU_LIST=8 9 10 11 12 13 14 15、端口 18110–18117，以及原 NAT_LEVELS/RAND_LEVELS；控制台本方案固定此组合。", "只检测并使用已启动的 8 个实例，不调用 serve-replicas.sh。所选配置自身 PORT 不代表此入口的实例端口。" ], "available": true }, { "id": "engine-comparison", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-compare.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_config}", "{run_label}" ], "parameters": [ { "key": "target", "label": "原服务配置", "type": "select", "default": "sglang38fp8", "options": [ { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 14400, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-compare.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/{target_config}.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_prompts.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/curves.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/{target_config}-{run_label}-*" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "Shell 继续读取 configs/{target_config}.conf 的 SERVED、PORT、GPUS、CT_OFF、CT_ON 等原字段；控制台不覆盖这些配置。", "保留原 NAT_LEVELS、THINK_N=16、RAND_LEVELS、LONG=32000、SEED_BASE=5000；本入口测试所选已运行配置，并不自行切换引擎。" ], "available": true }, { "id": "soak", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/soak.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [ "{target_config}", "{run_label}", "{concurrency}", "{requests}", "{long_inputs}" ], "parameters": [ { "key": "target", "label": "原服务配置", "type": "select", "default": "sglang38fp8", "options": [ { "value": "dsv41cap", "label": "dsv41cap（配置端口 18110）" }, { "value": "glm53", "label": "glm53（配置端口 18102）" }, { "value": "mimo", "label": "mimo（配置端口 18104）" }, { "value": "qwen27b", "label": "qwen27b（配置端口 18103）" }, { "value": "qwen27btp8", "label": "qwen27btp8（配置端口 18103）" }, { "value": "qwen38cap", "label": "qwen38cap（配置端口 18107）" }, { "value": "qwen38fp8", "label": "qwen38fp8（配置端口 18105）" }, { "value": "qwen38fp8cap", "label": "qwen38fp8cap（配置端口 18108）" }, { "value": "qwen38fp8mtpcap", "label": "qwen38fp8mtpcap（配置端口 18109）" }, { "value": "qwen38fp8tp8", "label": "qwen38fp8tp8（配置端口 18110）" }, { "value": "sglang38", "label": "sglang38（配置端口 18106）" }, { "value": "sglang38fp8", "label": "sglang38fp8（配置端口 18106）" } ], "required": true, "help": "配置存在不代表服务正在运行；预检仅检查端口，不自动启动模型。" }, { "key": "concurrency", "label": "并发数", "type": "integer", "default": 32, "min": 1, "max": 2048, "required": true }, { "key": "requests", "label": "主负载请求数", "type": "integer", "default": 3200, "min": 1, "max": 10000000, "required": true }, { "key": "long_inputs", "label": "长输入次数上限", "type": "integer", "default": 12, "min": 0, "max": 864, "required": true } ], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "resource", "title": "向原有模型服务施加并发负载", "details": [ "按原流程运行并发请求、性能采样与临时 benchmark 客户端；可能明显影响目标服务的延迟和吞吐。", "使用既有服务；此入口不会自动启动目标模型。原 Docker 客户端不属于控制台 cgroup，取消后需核对残留容器。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 259200, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/soak.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/{target_config}.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/concurrent_prefill.py" ], "check_ports": [ "{target_port}" ], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/{target_config}-{run_label}-*" ], "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "Shell 继续读取 configs/{target_config}.conf 的 SERVED、PORT、GPUS、CT_OFF、CT_ON 等原字段；控制台不覆盖这些配置。", "32 / 3200 / 12 为原 CLI 默认；按请求数决定时长。原遥测 timeout 固定 14400 秒，延长控制台超时不会延长原遥测覆盖。" ], "available": true }, { "id": "capacity", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/cap-8x-chain.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [], "parameters": [], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env" ], "impact": { "level": "service", "title": "停止 DeepSeek A，并进行 16 卡整机满载测试", "details": [ "原链先使用 GPU 8–15，随后停止 DeepSeek A，使用 GPU 0–15，启停多套模型及 16 个 Qwen 实例。", "原 EXIT trap 会尝试按 start-rec.sh 重启 A；失败、超时或取消不能保证恢复原状态。", "原日志和结果名称固定，可能覆盖已有 dsv41cap、glm53、mimo、qwen38fp8tp8、qwen38fp8cap 结果；需逐次确认并保全历史文件。", "2026-10-20 核对：原链末尾 start-rec.sh 默认切回 UNCENSORED 版，不会恢复当前 1017 主模型。GPU 8–14 另有图像/语音/Qwen 服务，执行前须协调资源。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 259200, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout", "/usr/bin/pgrep", "/usr/bin/setsid", "/usr/bin/vmstat" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/cap-8x-chain.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/serve-sets.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/serve-replicas.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/stop-model.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-multi.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-replicas.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/run-compare.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-replicas.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-serve.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_prompts.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/spec_bench.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/curves.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/dsv41cap.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/glm53.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/mimo.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8tp8.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8cap.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/phaseb-levels.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/start-rec.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/scripts/serve.sh" ], "check_ports": [], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/dsv41cap-01-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/dsv41cap-02-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/glm53-11-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/glm53-12-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/mimo-11-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/mimo-12-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/qwen38fp8tp8-01-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/qwen38fp8cap-02-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-dsv41cap-x01-g8.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-dsv41cap-x02-g*.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-glm53-x11-g8.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-glm53-x12-g*.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-mimo-x11-g8.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-mimo-x12-g*.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-qwen38fp8tp8-a1-g8.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-qwen38fp8cap-x16-g*.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/logs/serve-r7-rec-pin.log" ], "preserve_existing_outputs": true, "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "原入口无位置参数；不伪造可配置的编号或输出路径。原链使用固定历史文件名，再次运行可能覆盖同名结果，应在确认前检查并保全相关历史文件。" ], "available": true }, { "id": "cost", "source_group": "model-trials-20261011", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/value-chain.sh", "interpreter": "/usr/bin/bash", "user": "root", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials", "arguments": [], "parameters": [], "environment": {}, "credential_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env" ], "impact": { "level": "service", "title": "启停 GPU 8–15 上的性能与成本测试模型", "details": [ "原链等待 cap-8x-chain.sh 退出，再检查 GPU 8–15 显存并启动/停止 DeepSeek 与 8 份 Qwen 模型。", "原结果使用固定 dsv41v-*、qwen38v-* 名称，可能覆盖历史结果。", "该入口产生本机性能证据，成本折算和报告仍按原配套脚本处理；不代表本轮已验证价格口径。", "2026-10-20 核对：GPU 8–14 已有图像/语音/Qwen 服务；原测试会在这些 GPU 启动模型，须先协调资源，控制台不自动停止这些新服务。" ], "requires_confirmation": true }, "interactive": false, "timeout_seconds": 259200, "required_tools": [ "/usr/bin/bash", "/usr/bin/python3", "/usr/bin/docker", "/usr/bin/curl", "/usr/bin/nvidia-smi", "/usr/bin/timeout", "/usr/bin/pgrep", "/usr/bin/setsid", "/usr/bin/vmstat" ], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/value-chain.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/vllm.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/serve-sets.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/serve-replicas.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/stop-model.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-multi.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/scripts/bench-one.sh", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/dsv41cap.conf", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/configs/qwen38fp8cap.conf" ], "check_ports": [], "output_paths": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/dsv41v-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/qwen38v-*", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-dsv41cap-xv-g8.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261011-model-trials/logs/serve-qwen38fp8cap-xv8-g*.log" ], "preserve_existing_outputs": true, "notes": [ "2026-10-13 只读核对原源码；启动方案登记不代表本轮业务测试通过。", "原入口无位置参数；不伪造可配置的编号或输出路径。原链使用固定历史文件名，再次运行可能覆盖同名结果，应在确认前检查并保全相关历史文件。" ], "available": true }, { "id": "drama3-asr", "source_group": "drama3-tests-20261015", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_asr.py", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3", "interpreter": "/opt/omni/bin/python", "user": "root", "parameters": [], "arguments": [], "environment": {}, "credential_files": [], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_asr.py" ], "check_ports": [ 18301, 18302 ], "timeout_seconds": 3600, "impact": { "level": "resource", "title": "语音识别文件与实时转写会占用共享模型服务", "details": [ "用现有 Drama3 合成三段录音，核验 Qwen3-ASR 文件转写、实时转写、字错率和响应延迟；会占用共享语音服务。", "生成的音频与标准输出存入本次运行独立目录，原测试目录和历史结果保留。" ], "requires_confirmation": true }, "available": true, "output_paths": [], "notes": [ "2026-10-20 已只读核对原源码、依赖镜像和素材，未执行本项业务测试。", "原脚本只读挂载；测试客户端使用原镜像，无直接 GPU 设备权限，2 CPU / 4 GiB 内存；会通过接口占用既有模型服务。输出使用本次独立目录，不覆盖原历史音频。" ], "interactive": false, "docker": { "image": "sha256:13369a516b4b961e36b1c3c5d47e3e04ec4bfdc03920964c621ccc83232604b8", "mounts": [ { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests", "target": "/t" } ] } }, { "id": "drama3-api", "source_group": "drama3-tests-20261015", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_api.py", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3", "interpreter": "/opt/omni/bin/python", "user": "root", "parameters": [], "arguments": [], "environment": {}, "credential_files": [], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_api.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api/keys.txt", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs/api-requests.log", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/voices-src/female-1.wav", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/voices-src/female-2.wav", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/voices-src/male-1.wav", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/voices-src/male-2.wav" ], "check_ports": [ 18300, 18301 ], "timeout_seconds": 3600, "impact": { "level": "resource", "title": "语音合成网关兼容性与质量回归会占用共享模型服务", "details": [ "核验本地语音网关的多格式、四种音色、流式、变速、SDK 兼容和八路并发；保留原 selftest 凭据读取方式。", "生成的音频与标准输出存入本次运行独立目录，原测试目录和历史结果保留。" ], "requires_confirmation": true }, "available": true, "output_paths": [], "notes": [ "2026-10-20 已只读核对原源码、依赖镜像和素材，未执行本项业务测试。", "原脚本只读挂载；测试客户端使用原镜像，无直接 GPU 设备权限，2 CPU / 4 GiB 内存；会通过接口占用既有模型服务。输出使用本次独立目录，不覆盖原历史音频。", "复用原 selftest 身份；只读检查所需日志是否存在，其内容会持续变化，不把日志变化当作启动配置变更。" ], "interactive": false, "docker": { "image": "sha256:13369a516b4b961e36b1c3c5d47e3e04ec4bfdc03920964c621ccc83232604b8", "mounts": [ { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests", "target": "/t" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api", "target": "/secrets" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/voices-src", "target": "/refs" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs", "target": "/logs" } ] }, "secret_tables": [ { "path": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api/keys.txt", "key_column": 1, "label": "selftest" } ], "mutable_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs/api-requests.log" ] }, { "id": "drama3-voice-chat", "source_group": "drama3-tests-20261015", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_voice_chat.py", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3", "interpreter": "/opt/omni/bin/python", "user": "root", "parameters": [], "arguments": [], "environment": {}, "credential_files": [], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_voice_chat.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api/keys.txt", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs/api-requests.log" ], "check_ports": [ 18300, 18301, 18302, 18343, 18100 ], "timeout_seconds": 3600, "impact": { "level": "resource", "title": "实时语音对话端到端回归会占用共享模型服务", "details": [ "核验输入、回答、打断和四人对话的完整链路；当前语音服务预设仍指向旧 DeepSeek 模型名，待服务方确认后再开放。", "生成的音频与标准输出存入本次运行独立目录，原测试目录和历史结果保留。" ], "requires_confirmation": true }, "available": false, "output_paths": [], "notes": [ "2026-10-20 已只读核对原源码、依赖镜像和素材，未执行本项业务测试。", "原脚本只读挂载；测试客户端使用原镜像，无直接 GPU 设备权限，2 CPU / 4 GiB 内存；会通过接口占用既有模型服务。输出使用本次独立目录，不覆盖原历史音频。", "复用原 selftest 身份；只读检查所需日志是否存在，其内容会持续变化，不把日志变化当作启动配置变更。" ], "interactive": false, "docker": { "image": "sha256:13369a516b4b961e36b1c3c5d47e3e04ec4bfdc03920964c621ccc83232604b8", "mounts": [ { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests", "target": "/t" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api", "target": "/secrets" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs", "target": "/logs" } ] }, "unavailable_reason": "原语音服务的 DeepSeek 预设仍指向 deepseek-v4.1-flash-uncensored，与当前模型目录不一致；须由服务维护者调整预设后再接入。", "secret_tables": [ { "path": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api/keys.txt", "key_column": 1, "label": "selftest" } ], "mutable_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/logs/api-requests.log" ] }, { "id": "drama3-engine", "source_group": "drama3-tests-20261015", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_drama3.py", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3", "interpreter": "/opt/omni/bin/python", "user": "root", "parameters": [], "arguments": [], "environment": {}, "credential_files": [], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/test_drama3.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/admin.env", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/data/refs/example-en.wav" ], "check_ports": [ 18301 ], "timeout_seconds": 3600, "impact": { "level": "resource", "title": "Drama3 引擎合成与流式回归会占用共享模型服务", "details": [ "核验 Drama3 引擎的合成、克隆、流式和并发；依赖原管理凭据及外网参考音频下载，暂未接通执行。", "生成的音频与标准输出存入本次运行独立目录，原测试目录和历史结果保留。" ], "requires_confirmation": true }, "available": false, "output_paths": [], "notes": [ "2026-10-20 已只读核对原源码、依赖镜像和素材，未执行本项业务测试。", "原脚本只读挂载；测试客户端使用原镜像，无直接 GPU 设备权限，2 CPU / 4 GiB 内存；会通过接口占用既有模型服务。输出使用本次独立目录，不覆盖原历史音频。" ], "interactive": false, "docker": { "image": "sha256:13369a516b4b961e36b1c3c5d47e3e04ec4bfdc03920964c621ccc83232604b8", "mounts": [ { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests", "target": "/t" } ] }, "unavailable_reason": "此原测试还需管理凭据适配及外网参考音频下载，尚未核对其调用与输出边界。" }, { "id": "drama3-voice-capacity", "source_group": "drama3-tests-20261015", "entry": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/load_voice.py", "cwd": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3", "interpreter": "/opt/omni/bin/python", "user": "root", "parameters": [], "arguments": [], "environment": {}, "credential_files": [], "required_files": [ "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/load_voice.py", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api/keys.txt", "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests/mock_brain.py" ], "check_ports": [ 18300, 18301, 18302 ], "timeout_seconds": 3600, "impact": { "level": "resource", "title": "语音链路并发容量测试会占用共享模型服务", "details": [ "用模拟大脑测语音链路的并发容量、首声延迟和播放卡顿；需补齐临时辅助容器生命周期与每次资源上限。", "生成的音频与标准输出存入本次运行独立目录，原测试目录和历史结果保留。" ], "requires_confirmation": true }, "available": false, "output_paths": [], "notes": [ "2026-10-20 已只读核对原源码、依赖镜像和素材，未执行本项业务测试。", "原脚本只读挂载；测试客户端使用原镜像，无直接 GPU 设备权限，2 CPU / 4 GiB 内存；会通过接口占用既有模型服务。输出使用本次独立目录，不覆盖原历史音频。" ], "interactive": false, "docker": { "image": "sha256:13369a516b4b961e36b1c3c5d47e3e04ec4bfdc03920964c621ccc83232604b8", "mounts": [ { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/tests", "target": "/t" }, { "source": "/dengyuyun-joy-work-space/workpc-claude-code/20261015-drama3/secrets/api", "target": "/secrets" } ] }, "unavailable_reason": "需要专用模拟大脑容器、原内部网络及并发上限；辅助服务生命周期尚未接入。" } ], "reviewed_at": "2026-10-20" }