Skip to content

Add Mila Docs context to AI assistants

all_docs_condensed.md is a single-file distillation of the full Mila cluster documentation — covering cluster access, Slurm, storage, GPU selection, Python environments, experiment tracking, and more. Providing this file to any LLM chat assistant gives that assistant an accurate, Mila-specific knowledge base, making it possible to resolve cluster issues on the spot without manually searching the documentation.

Before you begin

Requirements

What this guide covers

  • Download the condensed reference file
  • Load the file into a chat LLM session
  • Phrase effective queries for accurate answers
  • Interpret and validate the assistant's responses

Download the condensed reference

all_docs_condensed.md is published at the root of the mila-docs repository. Download it once and keep it locally — it covers all major cluster topics and fits within the context window of all current major LLMs.

Download with curl:

1
curl -LO https://raw.githubusercontent.com/mila-iqia/mila-docs/master/all_docs_condensed.md

Or download manually on GitHub:

  1. Open the mila-docs repository.
  2. Click all_docs_condensed.md in the file list.

  3. Click the Download raw file button, on the upper right corner of the file.

    Download Github file

Load the file into a chat session

Most LLM chat assistants accept either file uploads or large text pastes.

  1. Open a new conversation at claude.ai.

  2. Click the paperclip icon. Provide file to Claude LLM

  3. Select all_docs_condensed.md from the local filesystem.

  4. Add a context message alongside the file upload:

    1
2
3
    This file is the Mila cluster documentation condensed into a
single reference. Use it to answer my questions about the
Mila cluster.

  5. Ask the first question in the same message or as a follow-up.

  1. Open a new conversation at chatgpt.com using GPT-4o or a later model.
  2. Click the paperclip icon. Provide file to ChatGPT
  3. Select all_docs_condensed.md from the local filesystem.

  4. Add a context message alongside the upload:

    1
2
3
    The attached file is the Mila cluster documentation condensed
into a single reference. Use it to answer my questions about
the Mila cluster.

  5. Ask the first question in the same message or as a follow-up.

  1. Open a new conversation at gemini.google.com.
  2. Click the attachment icon. Provide file to Gemini
  3. Select all_docs_condensed.md from the local filesystem.

  4. Add a context message alongside the upload:

    1
2
3
    The attached file is the Mila cluster documentation condensed
into a single reference. Use it to answer my questions about
the Mila cluster.

  5. Ask the first question in the same message or as a follow-up.

If the assistant does not support file uploads, paste the full content of all_docs_condensed.md directly into the message. The file is approximately 1 000 lines and fits within the context window of all major LLMs.

Paste into the message field and add the context instructions below before asking a question.

1
2
3
The attached file is the Mila cluster documentation condensed
into a single reference. Use it to answer my questions about
the Mila cluster.

Phrase effective queries

Queries that name the specific topic — Slurm commands, partition names, storage paths, or tool names — produce the most accurate responses.

Goal Example query
Look up a command What does the savail command do?
Understand a partition What are the resource limits for the unkillable partition?
Troubleshoot an error My job is stuck in PD state — what does that mean?
Get setup steps How do I set up MFA for the Mila cluster?
Compare options When should I use salloc instead of sbatch?
Check storage policies What is the difference between $SCRATCH and $HOME?

Tip

Include the word Mila or cluster in each query to help the assistant scope its answer to the provided file rather than drawing on general, potentially inaccurate knowledge.

Interpret and validate responses

LLM assistants can occasionally produce responses that look plausible but contain errors — especially for commands with many optional flags or for policies that change over time.

Apply these checks after any response:

  • Verify command flags: run man <command> or <command> --help on a login node to confirm flag names and syntax.
  • Check current partition limits: limits can change — verify with sinfo.
  • Confirm storage paths: quota policies and storage layout are sometimes updated between condensed-file releases.

Key concepts

all_docs_condensed.md
A single-file distillation of the full Mila technical documentation, designed to fit within the context window of any modern LLM assistant. Covers cluster access, Slurm, storage, Python environments, distributed training, and FAQ.
Context window
The maximum amount of text an LLM processes in a single session. Modern assistants support windows large enough to hold the entire condensed file.
Context message
An instruction given at the start of a session that tells the assistant how to interpret the provided material. Including a clear context message improves response accuracy.

Comments