Skip to content

llm_setup.md

Latest commit

 

History

History
131 lines (95 loc) · 2.88 KB

llm_setup.md

File metadata and controls

131 lines (95 loc) · 2.88 KB

LLM Setup

This project supports one optional provider-backed LLM workflow:

  • table1-parser review-variable-plausibility

The deterministic table1-parser parse command does not call an LLM.

Required Variables

When LLM_PROVIDER=openai, these must be set:

  • LLM_PROVIDER
  • OPENAI_API_KEY
  • OPENAI_MODEL

When LLM_PROVIDER=qwen, these must be set:

  • LLM_PROVIDER
  • DASHSCOPE_API_KEY
  • QWEN_MODEL

Optional:

  • LLM_TEMPERATURE
  • LLM_TIMEOUT_SECONDS
  • LLM_MAX_RETRIES
  • LLM_DEBUG
  • LLM_SDK_DEBUG
  • QWEN_BASE_URL

Example Setup On macOS/Linux

OpenAI:

export LLM_PROVIDER=openai
export OPENAI_API_KEY=your_api_key_here
export OPENAI_MODEL=gpt-4.1-mini
export LLM_TEMPERATURE=0
export LLM_TIMEOUT_SECONDS=60
export LLM_MAX_RETRIES=2
export LLM_DEBUG=false
export LLM_SDK_DEBUG=false

Qwen:

export LLM_PROVIDER=qwen
export DASHSCOPE_API_KEY=your_api_key_here
export QWEN_MODEL=qwen-plus
export QWEN_BASE_URL=https://dashscope.aliyuncs.com/api/v1
export LLM_TEMPERATURE=0
export LLM_TIMEOUT_SECONDS=60
export LLM_MAX_RETRIES=2
export LLM_DEBUG=false

Example Setup On Windows PowerShell

OpenAI:

$env:LLM_PROVIDER = "openai"
$env:OPENAI_API_KEY = "your_api_key_here"
$env:OPENAI_MODEL = "gpt-4.1-mini"
$env:LLM_TEMPERATURE = "0"
$env:LLM_TIMEOUT_SECONDS = "60"
$env:LLM_MAX_RETRIES = "2"
$env:LLM_DEBUG = "false"
$env:LLM_SDK_DEBUG = "false"

Qwen:

$env:LLM_PROVIDER = "qwen"
$env:DASHSCOPE_API_KEY = "your_api_key_here"
$env:QWEN_MODEL = "qwen-plus"
$env:QWEN_BASE_URL = "https://dashscope.aliyuncs.com/api/v1"
$env:LLM_TEMPERATURE = "0"
$env:LLM_TIMEOUT_SECONDS = "60"
$env:LLM_MAX_RETRIES = "2"
$env:LLM_DEBUG = "false"

Meaning of the two debug flags:

  • LLM_DEBUG=true write timestamped variable-plausibility debug JSON artifacts to disk during review-variable-plausibility
  • LLM_SDK_DEBUG=true enable verbose provider/SDK logging in the terminal

Install Requirement

The configured OpenAI client requires the OpenAI Python SDK:

python3 -m pip install -e '.[dev]'

The configured Qwen client uses the Python standard library HTTP stack.

If provider setup is missing, review-variable-plausibility skips provider calls with a clear setup warning and writes an empty review artifact.

Running The Configured LLM Path

table1-parser review-variable-plausibility testpapers/cobaltpaper.pdf

Debug Artifacts

If you want per-table review debug artifacts, enable:

export LLM_DEBUG=true

Then run table1-parser review-variable-plausibility .... The parser writes a timestamped debug directory under:

outputs/papers/<paper_stem>/llm_variable_plausibility_debug/

Security

  • Do not hardcode API keys in source files.
  • Do not commit API keys to the repository.
  • Prefer setting credentials in the shell environment or a local untracked secrets file loaded by your shell.