1026 lines
27 KiB
Markdown
1026 lines
27 KiB
Markdown
## config files
|
||
|
||
### tmux
|
||
|
||
```
|
||
bind-key b send-prefix
|
||
|
||
set -g default-terminal "tmux-256color"
|
||
set -ag terminal-overrides ",xterm-256color:RGB"
|
||
set -g status-right ""
|
||
|
||
unbind '%'
|
||
bind '|' split-window -h -c "#{pane_current_path}"
|
||
|
||
unbind '"'
|
||
bind '-' split-window -v -c "#{pane_current_path}"
|
||
|
||
bind -r m resize-pane -Z
|
||
|
||
set -g base-index 1
|
||
|
||
set-window-option -g mode-keys vi
|
||
bind-key -T copy-mode-vi 'v' send -X begin-selection
|
||
bind-key -T copy-mode-vi 'y' send -X copy-selection
|
||
unbind -T copy-mode-vi MouseDragEnd1Pane
|
||
```
|
||
|
||
### .bashrc
|
||
|
||
```bash
|
||
# export HOME=/mnt/debugger/wjh
|
||
|
||
[ -z "$PS1" ] && return
|
||
|
||
export TERM=xterm-256color
|
||
|
||
if [ -f ~/.dircolors ]; then
|
||
eval "$(dircolors -b ~/.dircolors)"
|
||
else
|
||
eval "$(dircolors -b)"
|
||
fi
|
||
|
||
alias ls='ls --color=auto -F'
|
||
alias ll='ls -lh --color=auto'
|
||
alias la='ll -a'
|
||
alias grep='grep --color=auto'
|
||
alias df='df -h'
|
||
alias du='du -h'
|
||
|
||
export PS1='\[\e[01;32m\]\u@\h\[\e[00m\]:\[\e[01;34m\]\w\[\e[00m\]\$ '
|
||
|
||
HISTSIZE=10000
|
||
HISTFILESIZE=20000
|
||
shopt -s histappend
|
||
export HISTCONTROL=ignoredups:erasedups
|
||
|
||
if [ -f /etc/bash_completion ] && ! shopt -oq posix; then
|
||
. /etc/bash_completion
|
||
fi
|
||
|
||
unset color_prompt force_color_prompt
|
||
|
||
use() {
|
||
source /mnt/debugger/wjh/.venv/$1/bin/activate
|
||
}
|
||
|
||
proxyOn() {
|
||
p="http://ipads:ipads123@127.0.0.1:11235"
|
||
export http_proxy=$p https_proxy=$p all_proxy=$p
|
||
}
|
||
|
||
proxyOff() {
|
||
unset http_proxy https_proxy all_proxy
|
||
}
|
||
```
|
||
|
||
### create user
|
||
|
||
```bash
|
||
#!/bin/bash
|
||
# 使用方法: sudo ./create_user.sh <用户名> <密码> <home目录路径>
|
||
|
||
# 检查是否以root权限运行
|
||
if [ "$(id -u)" != "0" ]; then
|
||
echo "此脚本必须以root权限运行。请使用sudo执行。" 1>&2
|
||
exit 1
|
||
fi
|
||
|
||
# 检查参数数量
|
||
if [ $# -lt 3 ]; then
|
||
echo "错误: 需要提供至少三个参数 - 用户名、密码和home目录路径"
|
||
echo "用法: $0 <用户名> <密码> <home目录路径> [<uid>]"
|
||
exit 1
|
||
fi
|
||
|
||
USERNAME=$1
|
||
PASSWORD=$2
|
||
HOME_DIR=$3
|
||
USER_ID=$4
|
||
|
||
# 创建用户并指定home目录
|
||
useradd -m -d "$HOME_DIR" -s /bin/bash "$USERNAME" -u $USER_ID
|
||
echo "用户 $USERNAME 已创建，home目录为 $HOME_DIR"
|
||
|
||
# 设置用户密码
|
||
echo "$USERNAME:$PASSWORD" | chpasswd
|
||
echo "用户 $USERNAME 的密码已设置"
|
||
|
||
# 赋予sudo权限 (适用于Ubuntu/Debian系统)
|
||
usermod -aG sudo "$USERNAME"
|
||
echo "用户 $USERNAME 已添加到sudo组，拥有管理员权限"
|
||
|
||
# 对于CentOS/RHEL系统，应使用以下命令代替上面的命令
|
||
# usermod -aG wheel "$USERNAME"
|
||
|
||
echo "用户创建完成！可以使用以下命令验证:"
|
||
echo "su - $USERNAME"
|
||
echo "sudo -l"
|
||
```
|
||
|
||
## AIBank
|
||
|
||
申请理由：
|
||
```
|
||
服务百炼在线模型推理任务，缩短推理服务的 TTFT 与 TPOT，节省推理资源。
|
||
|
||
1. 基于 vLLM 测试 MoE 的 expert 负载不均衡特性
|
||
2. 优化 vLLM 的 MoE 部分的性能
|
||
3. 测试 MoE 的性能，分析 MoE 部分的提升效果
|
||
```
|
||
|
||
```
|
||
服务百炼在线模型推理任务，实现动态的模型资源池切换，节省推理资源。
|
||
|
||
1. 在异构硬件（H20、H800、5090 等）上测试 vLLM serve 不同模型（30B、235B 等）的性能并分析
|
||
2. 依据性能分析结果，在异构硬件上做模型推理的服务配置调优，让任一模型在任一硬件上能够得到最优的推理配置
|
||
3. 为异构资源池上模型流量的切换与动态调整做指导，为不同模型在给定流量需求下分配最优的硬件资源池，让不同模型的总资源需求量尽可能减少，从而优化整体的资源利用率
|
||
```
|
||
|
||
### setup
|
||
|
||
```
|
||
curl -LsSf https://astral.sh/uv/install.sh | sh
|
||
```
|
||
|
||
|
||
|
||
```
|
||
IdentityFile ~/.ssh/gahow_main
|
||
ServerAliveInterval 180
|
||
ServerAliveCountMax 3
|
||
|
||
Host ali
|
||
HostName 47.83.188.3
|
||
User gahow
|
||
```
|
||
|
||
|
||
```
|
||
ssh -f -N -D 1080 ali
|
||
|
||
curl --socks5 127.0.0.1:1080 https://ifconfig.me
|
||
|
||
export SSL_CERT_FILE=$(python -m certifi)
|
||
|
||
curl http://ipinfo.io
|
||
```
|
||
|
||
|
||
|
||
``` bash
|
||
cd "$HOME/.vscode-server/extensions/openai.chatgpt-0.4.15/bin/linux-x86_64"
|
||
|
||
# keep the original
|
||
mv codex codex.real
|
||
|
||
# wrapper that forces a proxy and then calls the original
|
||
cat > codex <<'EOF'
|
||
#!/usr/bin/env bash
|
||
export HTTPS_PROXY="http://localhost:7890"
|
||
export HTTP_PROXY="http://localhost:7890"
|
||
export NO_PROXY="http://localhost:7890"
|
||
# ------------------------------------------------
|
||
HERE="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
|
||
exec "$HERE/codex.real" "$@"
|
||
EOF
|
||
chmod +x codex
|
||
```
|
||
|
||
|
||
|
||
```
|
||
networksetup -setv6off Wi-Fi
|
||
curl ipinfo.io
|
||
|
||
|
||
<html><body><p align="center"><a href="http://portal.wifi.alibaba-inc.com/loginPage-custom?random=c9ca5299-e5cf-4750-8a97-93010063568b&ap_mac=40%3A1A%3A58%3A2A%3A81%3A70&sign_method=hmac_sha256&ssid=alibaba-guest&sta_mac=9A%3A89%3A66%3ABB%3A08%3A5A×tamp=1761019558780&version=1.0&sign=5C396F5000AA6AF0CCC4EE9C0BC0F95B68E208D9522F23CA97AC3FA4AE56C8CF">如果浏览器不能自动跳转，请点击这里.</a><br><a href="http://portal.wifi.alibaba-inc.com/loginPage-custom?random=c9ca5299-e5cf-4750-8a97-93010063568b&ap_mac=40%3A1A%3A58%3A2A%3A81%3A70&sign_method=hmac_sha256&ssid=alibaba-guest&sta_mac=9A%3A89%3A66%3ABB%3A08%3A5A×tamp=1761019558780&version=1.0&sign=5C396F5000AA6AF0CCC4EE9C0BC0F95B68E208D9522F23CA97AC3FA4AE56C8CF">If the browser cannot be redirected automatically, please click here.</a></body></html>
|
||
```
|
||
|
||
|
||
|
||
|
||
```
|
||
pmset -g
|
||
|
||
sudo pmset -a sleep 0
|
||
sudo pmset -a disablesleep 1
|
||
|
||
sudo pmset -a disablesleep 0
|
||
sudo pmset -a sleep 10
|
||
|
||
pmset displaysleepnow
|
||
```
|
||
|
||
|
||
|
||
```bash
|
||
sudo apt install rclone
|
||
rclone config
|
||
rclone lsf mydav:
|
||
rclone mkdir mydav:/newfoler/xx
|
||
rclone copy /path/to/file mydav:/folder/
|
||
rclone sync /path/to/localdir mydav:/folder
|
||
```
|
||
|
||
|
||
# Dash vllm
|
||
|
||
`unset VLLM_CACHE_ROOT`
|
||
|
||
|
||
|
||
```json
|
||
{
|
||
"memory_cache": "100G",
|
||
"rdma": "1",
|
||
"enable-nvidia-ibgda": "true",
|
||
"enable-nvidia-gdrcopy": "true",
|
||
"global_dynamic_batching": {
|
||
"enabled": true,
|
||
"scheduler_type": "cache-aware-batching",
|
||
"scheduler_config": {
|
||
"max_batch_size": 128,
|
||
"max_token_num": 1000000,
|
||
"request_seize_threshold": 20,
|
||
"prefix_match_weight": 10,
|
||
"load_weight": 20,
|
||
"prefill_concurrency": 2,
|
||
"long_context": {
|
||
"long_context_threshold": 30000,
|
||
"long_context_proportion": 0.1,
|
||
"max_long_context_proportion": 0.3
|
||
},
|
||
"enable_global_batching_optimization": true
|
||
},
|
||
"request_transmission": "grpc",
|
||
"response_transmission": "grpc",
|
||
"preferred_batch_size": 0,
|
||
"token_capacity": 0,
|
||
"max_queue_delay": 100,
|
||
"bucket_strategy": "",
|
||
"bucket_count": 1,
|
||
"bucket_thresholds": []
|
||
},
|
||
"turbo": {
|
||
"env": {
|
||
"DashScope_Service_GracefulShutdown_WaitSeconds": 300,
|
||
"DashScope_Api_Cache_Aware_EnableLengthIsolated": true
|
||
}
|
||
},
|
||
"termination_grace_period": 300
|
||
}
|
||
```
|
||
|
||
|
||
# Trace
|
||
|
||
Some Observations
|
||
- 很多请求在 tool call 之后，携带 tool response 的 content 作为 input
|
||
```
|
||
<|im_start|>system\n{user_sys}{bailian_sys_prompt}<|im_end|>
|
||
<|im_start|>user\n{user_content}<|im_end|>
|
||
<|im_start|>assistant\n<think><|im_end|>
|
||
|
||
<|im_start|>user\n<tool_response>\n{tool_content}<\tool_response><|im_end|>
|
||
```
|
||
- thinking 下 input_length < 8k，但 input_length + output_length > 8k 的占比不小，多轮对话少（因为只从一个 cluster 内采集）
|
||
```
|
||
thinking: cnt=1845, cnt / len(data)=0.16394170961435933
|
||
coder: cnt=1871, cnt / len(data)=0.012262580450654747
|
||
```
|
||
|
||
|
||
|
||
字段：
|
||
|
||
- thinking
|
||
|
||
```
|
||
__tag__:__hostname__
|
||
__tag__:__pack_id__
|
||
__tag__:__path__
|
||
__tag__:__receive_time__
|
||
__tag__:__service_name__
|
||
__tag__:__user_defined_id__
|
||
__tag__:_container_ip_
|
||
__tag__:_container_name_
|
||
__tag__:_image_name_
|
||
__tag__:_namespace_
|
||
__tag__:_pod_name_
|
||
__tag__:_pod_uid_
|
||
__tag__:eci_id
|
||
__time__
|
||
__topic__
|
||
cluster_name
|
||
cluster_type
|
||
data_name
|
||
data_type
|
||
data_version
|
||
env
|
||
framework_type
|
||
framework_version
|
||
ip
|
||
model_name
|
||
model_version
|
||
program_name
|
||
program_version
|
||
region
|
||
request_count
|
||
request_data_size
|
||
request_data_url
|
||
request_id
|
||
request_params
|
||
request_params.header
|
||
request_params.header.attributes
|
||
request_params.header.attributes.app_id
|
||
request_params.header.attributes.baggage
|
||
request_params.header.attributes.model
|
||
request_params.header.attributes.subuser_id
|
||
request_params.header.attributes.traceparent
|
||
request_params.header.attributes.traceparent_new
|
||
request_params.header.attributes.tracestate
|
||
request_params.header.attributes.user_id
|
||
request_params.header.attributes.workspace_id
|
||
request_params.header.attributes.x-dashscope-apikeyid
|
||
request_params.header.attributes.x-dashscope-apikeyloc
|
||
request_params.header.attributes.x-dashscope-app
|
||
request_params.header.attributes.x-dashscope-bwid
|
||
request_params.header.attributes.x-dashscope-inner-enableestimatedusage
|
||
request_params.header.attributes.x-dashscope-inner-flow-control
|
||
request_params.header.attributes.x-dashscope-inner-logging-policy
|
||
request_params.header.attributes.x-dashscope-inner-model-type
|
||
request_params.header.attributes.x-dashscope-inner-request-priority
|
||
request_params.header.attributes.x-dashscope-inner-requestreadytime
|
||
request_params.header.attributes.x-dashscope-inner-streammode
|
||
request_params.header.attributes.x-dashscope-inner-timeout
|
||
request_params.header.attributes.x-dashscope-inner-user-meta
|
||
request_params.header.attributes.x-dashscope-inner-user-meta.dashscope_connector_user_type
|
||
request_params.header.attributes.x-dashscope-requestid
|
||
request_params.header.attributes.x-dashscope-service
|
||
request_params.header.attributes.x-dashscope-sse
|
||
request_params.header.attributes.x-dashscope-subuid
|
||
request_params.header.attributes.x-dashscope-uid
|
||
request_params.header.attributes.x-dashscope-workspace
|
||
request_params.header.attributes.x-ds-request-priority
|
||
request_params.header.attributes.x-ds-virtual-service
|
||
request_params.header.attributes.x-ds-virtual-service-version
|
||
request_params.header.request_id
|
||
request_params.payload
|
||
request_params.payload.input
|
||
request_params.payload.input.messages
|
||
request_params.payload.input.messages.content
|
||
request_params.payload.input.messages.name
|
||
request_params.payload.input.messages.reasoning_content
|
||
request_params.payload.input.messages.role
|
||
request_params.payload.input.messages.tool_call_id
|
||
request_params.payload.input.messages.tool_calls
|
||
request_params.payload.input.messages.tool_calls.function
|
||
request_params.payload.input.messages.tool_calls.function.arguments
|
||
request_params.payload.input.messages.tool_calls.function.arguments.code
|
||
request_params.payload.input.messages.tool_calls.function.name
|
||
request_params.payload.input.messages.tool_calls.id
|
||
request_params.payload.input.messages.tool_calls.index
|
||
request_params.payload.input.messages.tool_calls.type
|
||
request_params.payload.parameters
|
||
request_params.payload.parameters.enable_thinking
|
||
request_params.payload.parameters.extra_body
|
||
request_params.payload.parameters.extra_body.ignore_datainspection_error
|
||
request_params.payload.parameters.ignore_datainspection_error
|
||
request_params.payload.parameters.incremental_output
|
||
request_params.payload.parameters.max_tokens
|
||
request_params.payload.parameters.presence_penalty
|
||
request_params.payload.parameters.result_format
|
||
request_params.payload.parameters.temperature
|
||
request_params.payload.parameters.tools
|
||
request_params.payload.parameters.tools.function
|
||
request_params.payload.parameters.tools.function.description
|
||
request_params.payload.parameters.tools.function.name
|
||
request_params.payload.parameters.tools.function.parameters
|
||
request_params.payload.parameters.tools.function.parameters.properties
|
||
request_params.payload.parameters.tools.function.parameters.properties.code
|
||
request_params.payload.parameters.tools.function.parameters.properties.code.description
|
||
request_params.payload.parameters.tools.function.parameters.properties.code.type
|
||
request_params.payload.parameters.tools.function.parameters.properties.query
|
||
request_params.payload.parameters.tools.function.parameters.properties.query.description
|
||
request_params.payload.parameters.tools.function.parameters.properties.query.type
|
||
request_params.payload.parameters.tools.function.parameters.properties.url
|
||
request_params.payload.parameters.tools.function.parameters.properties.url.description
|
||
request_params.payload.parameters.tools.function.parameters.properties.url.type
|
||
request_params.payload.parameters.tools.function.parameters.required
|
||
request_params.payload.parameters.tools.function.parameters.type
|
||
request_params.payload.parameters.tools.type
|
||
request_params.payload.parameters.top_p
|
||
request_params.payload.resources
|
||
response_count
|
||
response_data_size
|
||
response_params
|
||
response_params.header
|
||
response_params.header.attributes
|
||
response_params.header.attributes.request_uuid
|
||
response_params.header.attributes.x-ds-backend-first-request-time
|
||
response_params.header.attributes.x-ds-backend-first-response-time
|
||
response_params.header.finished
|
||
response_params.header.request_id
|
||
response_params.header.status_code
|
||
response_params.header.status_message
|
||
response_params.header.status_name
|
||
response_params.payload
|
||
response_params.payload.output
|
||
response_params.payload.output.choices
|
||
response_params.payload.output.choices.finish_reason
|
||
response_params.payload.output.choices.index
|
||
response_params.payload.output.choices.message
|
||
response_params.payload.output.choices.message.content
|
||
response_params.payload.output.choices.message.reasoning_content
|
||
response_params.payload.output.choices.message.role
|
||
response_params.payload.output.choices.message.tool_calls
|
||
response_params.payload.output.choices.message.tool_calls.function
|
||
response_params.payload.output.choices.message.tool_calls.function.arguments
|
||
response_params.payload.output.choices.message.tool_calls.function.arguments.code
|
||
response_params.payload.output.choices.message.tool_calls.function.name
|
||
response_params.payload.output.choices.message.tool_calls.id
|
||
response_params.payload.output.choices.message.tool_calls.index
|
||
response_params.payload.output.choices.message.tool_calls.type
|
||
response_params.payload.usage
|
||
response_params.payload.usage.cached_tokens
|
||
response_params.payload.usage.input_tokens
|
||
response_params.payload.usage.output_tokens
|
||
response_params.payload.usage.output_tokens_details
|
||
response_params.payload.usage.output_tokens_details.reasoning_tokens
|
||
response_params.payload.usage.total_tokens
|
||
service_group
|
||
service_id
|
||
service_name
|
||
session_error
|
||
session_id
|
||
status_code
|
||
status_message
|
||
status_name
|
||
time
|
||
total_cost_time
|
||
__source__
|
||
```
|
||
|
||
- thinking-info
|
||
```
|
||
=== 发现的字段列表 ===
|
||
__tag__:__hostname__
|
||
__tag__:__pack_id__
|
||
__tag__:__path__
|
||
__tag__:__receive_time__
|
||
__tag__:__service_name__
|
||
__tag__:__user_defined_id__
|
||
__tag__:_container_ip_
|
||
__tag__:_container_name_
|
||
__tag__:_image_name_
|
||
__tag__:_namespace_
|
||
__tag__:_pod_name_
|
||
__tag__:_pod_uid_
|
||
__tag__:eci_id
|
||
__time__
|
||
__topic__
|
||
chunk_index
|
||
context
|
||
context.extra_params
|
||
context.extra_params.baggage
|
||
context.extra_params.cache_control
|
||
context.extra_params.ds_header_attributes
|
||
context.extra_params.ds_header_attributes.X-DashScope-APP-EXTRA-INFO
|
||
context.extra_params.ds_header_attributes.X-DashScope-EUID
|
||
context.extra_params.ds_header_attributes.X-DashScope-Inner-StreamMode
|
||
context.extra_params.ds_header_attributes.X-DashScope-Inner-Timeout
|
||
context.extra_params.ds_header_attributes.app_id
|
||
context.extra_params.ds_header_attributes.baggage
|
||
context.extra_params.ds_header_attributes.model
|
||
context.extra_params.ds_header_attributes.model_service_id
|
||
context.extra_params.ds_header_attributes.subuser_id
|
||
context.extra_params.ds_header_attributes.traceparent
|
||
context.extra_params.ds_header_attributes.traceparent_new
|
||
context.extra_params.ds_header_attributes.tracestate
|
||
context.extra_params.ds_header_attributes.user_id
|
||
context.extra_params.ds_header_attributes.workspace_id
|
||
context.extra_params.ds_header_attributes.x-dashscope-apikeyid
|
||
context.extra_params.ds_header_attributes.x-dashscope-apikeyloc
|
||
context.extra_params.ds_header_attributes.x-dashscope-app
|
||
context.extra_params.ds_header_attributes.x-dashscope-app-extra-info
|
||
context.extra_params.ds_header_attributes.x-dashscope-avgtpsrequestsnum
|
||
context.extra_params.ds_header_attributes.x-dashscope-bwid
|
||
context.extra_params.ds_header_attributes.x-dashscope-datainspection
|
||
context.extra_params.ds_header_attributes.x-dashscope-euid
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-batch-id
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-enableestimatedusage
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-flow-control
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-logging-policy
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-model-type
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-request-priority
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-requestreadytime
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-streammode
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-timeout
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-timeout-heartbeat-interval
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-user-meta
|
||
context.extra_params.ds_header_attributes.x-dashscope-inner-user-meta.dashscope_connector_user_type
|
||
context.extra_params.ds_header_attributes.x-dashscope-loggingconsent
|
||
context.extra_params.ds_header_attributes.x-dashscope-requestid
|
||
context.extra_params.ds_header_attributes.x-dashscope-service
|
||
context.extra_params.ds_header_attributes.x-dashscope-sse
|
||
context.extra_params.ds_header_attributes.x-dashscope-subuid
|
||
context.extra_params.ds_header_attributes.x-dashscope-uid
|
||
context.extra_params.ds_header_attributes.x-dashscope-workspace
|
||
context.extra_params.ds_header_attributes.x-dashscope-workspaceid
|
||
context.extra_params.ds_header_attributes.x-ds-llm-input-tokens
|
||
context.extra_params.ds_header_attributes.x-ds-llm-response-format
|
||
context.extra_params.ds_header_attributes.x-ds-llm-thinking
|
||
context.extra_params.ds_header_attributes.x-ds-request-priority
|
||
context.extra_params.ds_header_attributes.x-ds-virtual-service
|
||
context.extra_params.ds_header_attributes.x-ds-virtual-service-version
|
||
context.extra_params.incremental_output
|
||
context.extra_params.service_id
|
||
context.extra_params.stop
|
||
context.extra_params.stop_token_ids
|
||
context.extra_params.traceparent
|
||
context.extra_params.tracestate
|
||
context.input_text
|
||
context.model_serving_params
|
||
context.model_serving_params.do_sample
|
||
context.model_serving_params.early_stopping
|
||
context.model_serving_params.eos_token_id
|
||
context.model_serving_params.head_response_per_tokens
|
||
context.model_serving_params.input_ids
|
||
context.model_serving_params.length_penalty
|
||
context.model_serving_params.logprobs
|
||
context.model_serving_params.max_new_think_tokens
|
||
context.model_serving_params.max_new_tokens
|
||
context.model_serving_params.min_length
|
||
context.model_serving_params.n
|
||
context.model_serving_params.no_repeat_ngram_size
|
||
context.model_serving_params.num_beams
|
||
context.model_serving_params.presence_penalty
|
||
context.model_serving_params.repetition_penalty
|
||
context.model_serving_params.seed
|
||
context.model_serving_params.temperature
|
||
context.model_serving_params.top_k
|
||
context.model_serving_params.top_logprobs
|
||
context.model_serving_params.top_p
|
||
ds_service_id
|
||
ds_service_name
|
||
interval
|
||
interval.cost
|
||
interval.type
|
||
message
|
||
request_id
|
||
step
|
||
time
|
||
timestamp
|
||
total_chunks
|
||
user_id
|
||
__source__
|
||
```
|
||
|
||
|
||
- coder
|
||
|
||
```
|
||
===== Chat Session Analysis (FAST) =====
|
||
Mode: strict
|
||
Total requests: 151817
|
||
Total sessions: 75146
|
||
Mean turns per session: 2.020
|
||
|
||
===== Turn Distribution =====
|
||
turns= 1 sessions=65024
|
||
turns= 2 sessions=3367
|
||
turns= 3 sessions=1814
|
||
turns= 4 sessions=997
|
||
turns= 5 sessions=725
|
||
turns= 6 sessions=487
|
||
turns= 7 sessions=334
|
||
turns= 8 sessions=262
|
||
turns= 9 sessions=198
|
||
turns= 10 sessions=164
|
||
turns= 11 sessions=148
|
||
turns= 12 sessions=132
|
||
turns= 13 sessions=81
|
||
turns= 14 sessions=70
|
||
turns= 15 sessions=78
|
||
turns= 16 sessions=62
|
||
turns= 17 sessions=72
|
||
turns= 18 sessions=67
|
||
turns= 19 sessions=55
|
||
turns= 20 sessions=52
|
||
turns= 21 sessions=35
|
||
turns= 22 sessions=51
|
||
turns= 23 sessions=40
|
||
turns= 24 sessions=37
|
||
turns= 25 sessions=42
|
||
turns= 26 sessions=28
|
||
turns= 27 sessions=28
|
||
turns= 28 sessions=28
|
||
turns= 29 sessions=22
|
||
turns= 30 sessions=24
|
||
turns= 31 sessions=31
|
||
turns= 32 sessions=17
|
||
turns= 33 sessions=23
|
||
turns= 34 sessions=11
|
||
turns= 35 sessions=15
|
||
turns= 36 sessions=23
|
||
turns= 37 sessions=12
|
||
turns= 38 sessions=11
|
||
turns= 39 sessions=20
|
||
turns= 40 sessions=18
|
||
turns= 41 sessions=10
|
||
turns= 42 sessions=20
|
||
turns= 43 sessions=19
|
||
turns= 44 sessions=21
|
||
turns= 45 sessions=19
|
||
turns= 46 sessions=11
|
||
turns= 47 sessions=7
|
||
turns= 48 sessions=12
|
||
turns= 49 sessions=13
|
||
turns= 50 sessions=6
|
||
turns= 51 sessions=8
|
||
turns= 52 sessions=6
|
||
turns= 53 sessions=13
|
||
turns= 54 sessions=14
|
||
turns= 55 sessions=7
|
||
turns= 56 sessions=12
|
||
turns= 57 sessions=12
|
||
turns= 58 sessions=10
|
||
turns= 59 sessions=4
|
||
turns= 60 sessions=3
|
||
turns= 61 sessions=11
|
||
turns= 62 sessions=5
|
||
turns= 63 sessions=4
|
||
turns= 64 sessions=6
|
||
turns= 65 sessions=6
|
||
turns= 66 sessions=6
|
||
turns= 67 sessions=6
|
||
turns= 68 sessions=7
|
||
turns= 69 sessions=3
|
||
turns= 70 sessions=7
|
||
turns= 71 sessions=5
|
||
turns= 72 sessions=9
|
||
turns= 73 sessions=3
|
||
turns= 74 sessions=9
|
||
turns= 75 sessions=8
|
||
turns= 76 sessions=2
|
||
turns= 77 sessions=3
|
||
turns= 78 sessions=2
|
||
turns= 79 sessions=3
|
||
turns= 80 sessions=4
|
||
turns= 81 sessions=3
|
||
turns= 82 sessions=3
|
||
turns= 83 sessions=6
|
||
turns= 84 sessions=2
|
||
turns= 85 sessions=5
|
||
turns= 86 sessions=6
|
||
turns= 87 sessions=6
|
||
turns= 88 sessions=1
|
||
turns= 89 sessions=3
|
||
turns= 91 sessions=4
|
||
turns= 92 sessions=3
|
||
turns= 93 sessions=4
|
||
turns= 94 sessions=2
|
||
turns= 96 sessions=2
|
||
turns= 97 sessions=3
|
||
turns= 98 sessions=2
|
||
turns= 99 sessions=4
|
||
turns=100 sessions=4
|
||
turns=101 sessions=2
|
||
turns=102 sessions=3
|
||
turns=103 sessions=2
|
||
turns=106 sessions=2
|
||
turns=107 sessions=1
|
||
turns=110 sessions=2
|
||
turns=112 sessions=3
|
||
turns=113 sessions=1
|
||
turns=116 sessions=1
|
||
turns=117 sessions=1
|
||
turns=118 sessions=1
|
||
turns=119 sessions=1
|
||
turns=120 sessions=1
|
||
turns=121 sessions=2
|
||
turns=122 sessions=2
|
||
turns=125 sessions=1
|
||
turns=126 sessions=1
|
||
turns=128 sessions=2
|
||
turns=132 sessions=1
|
||
turns=134 sessions=1
|
||
turns=136 sessions=1
|
||
turns=137 sessions=1
|
||
turns=142 sessions=1
|
||
turns=144 sessions=1
|
||
turns=156 sessions=2
|
||
turns=157 sessions=1
|
||
turns=160 sessions=2
|
||
turns=185 sessions=1
|
||
turns=220 sessions=1
|
||
```
|
||
|
||
|
||
```
|
||
===== Chat Session Analysis (FAST) =====
|
||
Mode: strict
|
||
Total requests: 10913
|
||
Total sessions: 10605
|
||
Mean turns per session: 1.029
|
||
|
||
===== Turn Distribution =====
|
||
turns= 1 sessions=10474
|
||
turns= 2 sessions=69
|
||
turns= 3 sessions=25
|
||
turns= 4 sessions=17
|
||
turns= 5 sessions=9
|
||
turns= 6 sessions=2
|
||
turns= 7 sessions=1
|
||
turns= 8 sessions=1
|
||
turns= 9 sessions=2
|
||
turns= 12 sessions=2
|
||
turns= 13 sessions=1
|
||
turns= 14 sessions=1
|
||
turns= 17 sessions=1
|
||
```
|
||
|
||
- Thinking V2
|
||
```
|
||
===== Chat Session Analysis (FAST) =====
|
||
Mode: strict
|
||
Total requests: 10913
|
||
Total sessions: 9611
|
||
Mean turns per session: 1.135
|
||
|
||
===== Turn Distribution =====
|
||
turns= 1 sessions=9083
|
||
turns= 2 sessions=263
|
||
turns= 3 sessions=101
|
||
turns= 4 sessions=49
|
||
turns= 5 sessions=39
|
||
turns= 6 sessions=35
|
||
turns= 7 sessions=14
|
||
turns= 8 sessions=8
|
||
turns= 9 sessions=6
|
||
turns= 10 sessions=3
|
||
turns= 11 sessions=3
|
||
turns= 12 sessions=2
|
||
turns= 14 sessions=2
|
||
turns= 15 sessions=1
|
||
turns= 18 sessions=1
|
||
turns= 36 sessions=1
|
||
```
|
||
|
||
- Coder V2
|
||
```
|
||
turns= 1 sessions=106901
|
||
turns= 2 sessions=3770
|
||
turns= 3 sessions=1897
|
||
turns= 4 sessions=925
|
||
turns= 5 sessions=668
|
||
turns= 6 sessions=417
|
||
turns= 7 sessions=270
|
||
turns= 8 sessions=216
|
||
turns= 9 sessions=142
|
||
turns= 10 sessions=116
|
||
turns= 11 sessions=95
|
||
turns= 12 sessions=86
|
||
turns= 13 sessions=54
|
||
turns= 14 sessions=45
|
||
turns= 15 sessions=53
|
||
turns= 16 sessions=32
|
||
turns= 17 sessions=35
|
||
turns= 18 sessions=31
|
||
turns= 19 sessions=23
|
||
turns= 20 sessions=22
|
||
turns= 21 sessions=14
|
||
turns= 22 sessions=25
|
||
turns= 23 sessions=22
|
||
turns= 24 sessions=16
|
||
turns= 25 sessions=16
|
||
turns= 26 sessions=13
|
||
turns= 27 sessions=11
|
||
turns= 28 sessions=9
|
||
turns= 29 sessions=13
|
||
turns= 30 sessions=8
|
||
turns= 31 sessions=6
|
||
turns= 32 sessions=2
|
||
turns= 33 sessions=3
|
||
turns= 34 sessions=7
|
||
turns= 35 sessions=5
|
||
turns= 36 sessions=4
|
||
turns= 37 sessions=4
|
||
turns= 38 sessions=3
|
||
turns= 39 sessions=8
|
||
turns= 40 sessions=6
|
||
turns= 41 sessions=3
|
||
turns= 42 sessions=6
|
||
turns= 43 sessions=3
|
||
turns= 44 sessions=3
|
||
turns= 45 sessions=4
|
||
turns= 46 sessions=1
|
||
turns= 48 sessions=3
|
||
turns= 49 sessions=4
|
||
turns= 50 sessions=1
|
||
turns= 51 sessions=2
|
||
turns= 52 sessions=3
|
||
turns= 53 sessions=1
|
||
turns= 54 sessions=4
|
||
turns= 55 sessions=5
|
||
turns= 56 sessions=2
|
||
turns= 57 sessions=1
|
||
turns= 58 sessions=1
|
||
turns= 59 sessions=1
|
||
turns= 60 sessions=2
|
||
turns= 61 sessions=1
|
||
turns= 62 sessions=3
|
||
turns= 66 sessions=1
|
||
turns= 68 sessions=1
|
||
turns= 69 sessions=3
|
||
turns= 70 sessions=1
|
||
turns= 71 sessions=1
|
||
turns= 72 sessions=1
|
||
turns= 73 sessions=1
|
||
turns= 80 sessions=1
|
||
turns= 81 sessions=1
|
||
turns= 85 sessions=1
|
||
turns= 98 sessions=1
|
||
turns=100 sessions=1
|
||
turns=102 sessions=1
|
||
turns=103 sessions=1
|
||
```
|
||
|
||
|
||
# Dashscope
|
||
|
||
```bash
|
||
mkdir -p ~/.ssh
|
||
echo "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFFpGJiuu2lGzQd1cmqSYBAO5sL0gc3+RSQKm9IbMUte Gahow Wang <gahow.wang@gmail.com>" >> ~/.ssh/authorized_keys
|
||
chmod 600 ~/.ssh/authorized_keys
|
||
```
|
||
|
||
```bash
|
||
echo "Port 10022" >> /etc/ssh/sshd_config
|
||
echo "PermitRootLogin yes" >> /etc/ssh/sshd_config
|
||
if [[ -z "$(which service 2>/dev/null)" ]]; then /usr/sbin/sshd; else service ssh restart; fi
|
||
```
|
||
|
||
```bash
|
||
sudo apt update && sudo apt install --no-install-recommends nodejs npm -y
|
||
sudo npm i -g @openai/codex
|
||
```
|
||
|
||
```bash
|
||
sudo rm /usr/lib/python3.12/sitecustomize.py
|
||
```
|
||
|
||
```bash
|
||
echo "IdentityFile ~/.ssh/gahow_main" >> ~/.ssh/config
|
||
echo "sk" > ~/.ssh/gahow_main
|
||
chmod 600 gahow_main
|
||
```
|
||
|
||
|
||
```bash
|
||
proxyOn() {
|
||
p="http://127.0.0.1:7890"
|
||
export http_proxy=$p https_proxy=$p all_proxy=$p
|
||
}
|
||
|
||
proxyOff() {
|
||
unset http_proxy https_proxy all_proxy
|
||
}
|
||
>> ~/.bashrc
|
||
```
|
||
|
||
```bash
|
||
git lfs install
|
||
|
||
git clone git@ipads.se.sjtu.edu.cn:wangjh/auto-tuner.git
|
||
git checkout feat/agent
|
||
git submodule update --init
|
||
```
|
||
|
||
```bash
|
||
curl -LsSf https://astral.sh/uv/install.sh | sh
|
||
export UV_INDEX_URL=https://mirrors.aliyun.com/pypi/simple
|
||
uv sync
|
||
|
||
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
|
||
cargo install ripgrep
|
||
```
|
||
|
||
```bash
|
||
./build_replayer.sh
|
||
```
|
||
|
||
|
||
|
||
|
||
- copy file
|
||
```bash
|
||
#!/usr/bin/env bash
|
||
set -u
|
||
set -o pipefail
|
||
|
||
HOST="29cb1b48db8-cdu78.cn-wulanchabu.nas.aliyuncs.com"
|
||
SRC_ROOT="nfs://${HOST}/wjh"
|
||
DST_ROOT="/dashscope/caches/application/wjh"
|
||
|
||
PARALLEL=256
|
||
RETRY=3
|
||
|
||
mkdir -p "${DST_ROOT}"
|
||
|
||
tmp_root=$(mktemp)
|
||
tmp_auto=$(mktemp)
|
||
tmp_dotdirs=$(mktemp)
|
||
|
||
trap 'rm -f "$tmp_root" "$tmp_auto" "$tmp_dotdirs"' EXIT
|
||
|
||
|
||
copy_one() {
|
||
|
||
rel="$1"
|
||
src="nfs://${HOST}/wjh/${rel}"
|
||
dst="${DST_ROOT}/${rel}"
|
||
|
||
# ---- resume support ----
|
||
if [ -f "$dst" ]; then
|
||
echo "[SKIP] already exists $rel"
|
||
return 0
|
||
fi
|
||
|
||
mkdir -p "$(dirname "$dst")"
|
||
|
||
i=1
|
||
while [ "$i" -le "$RETRY" ]; do
|
||
|
||
echo "[COPY] ${rel} (try $i/$RETRY)"
|
||
|
||
if nfs-cat "$src" > "${dst}.tmp"; then
|
||
mv "${dst}.tmp" "$dst"
|
||
return 0
|
||
fi
|
||
|
||
rm -f "${dst}.tmp"
|
||
|
||
i=$((i+1))
|
||
sleep 1
|
||
|
||
done
|
||
|
||
echo "[FAIL] ${rel}" >&2
|
||
}
|
||
|
||
export HOST DST_ROOT RETRY
|
||
export -f copy_one
|
||
|
||
|
||
echo "===== Step 1: root regular files ====="
|
||
|
||
nfs-ls "${SRC_ROOT}" \
|
||
| awk '$1 ~ /^-/ {print $NF}' \
|
||
| grep -Ev '\.venv|\.cache' \
|
||
> "$tmp_root"
|
||
|
||
|
||
echo "===== Step 2: auto-tuner recursive ====="
|
||
|
||
nfs-ls -R "${SRC_ROOT}/auto-tuner" \
|
||
| awk '$1 ~ /^-/ {print "auto-tuner/" $NF}' \
|
||
| grep -Ev '\.venv|\.cache' \
|
||
> "$tmp_auto"
|
||
|
||
|
||
echo "===== Step 3: dot directories ====="
|
||
|
||
nfs-ls "${SRC_ROOT}" \
|
||
| awk '$1 ~ /^d/ && $NF ~ /^\./ {print $NF}' \
|
||
| grep -Ev '\.venv|\.cache' \
|
||
> "$tmp_dotdirs"
|
||
|
||
|
||
echo "===== Step 4: scan dot dirs ====="
|
||
|
||
while read d
|
||
do
|
||
|
||
echo "[DOTDIR] $d"
|
||
|
||
nfs-ls -R "${SRC_ROOT}/${d}" \
|
||
| awk -v p="$d/" '$1 ~ /^-/ {print p $NF}' \
|
||
| grep -Ev '\.venv|\.cache'
|
||
|
||
done < "$tmp_dotdirs" >> "$tmp_auto"
|
||
|
||
|
||
echo "===== Step 5: copy ====="
|
||
|
||
cat "$tmp_root" "$tmp_auto" \
|
||
| awk '!seen[$0]++' \
|
||
| xargs -P "$PARALLEL" -I {} bash -c 'copy_one "$1"' _ {}
|
||
|
||
|
||
echo "===== DONE ====="
|
||
```
|
||
|
||
|