upload_instructions

README.md

@@ -2,24 +2,101 @@
 license: apache-2.0
 ---
 
+# BoAmps open data repository
+
+## What is this repository?
+
+This is the **open database of BoAmps reports**: a public collection of measurements of the energy consumption and environmental impact of AI workloads (inference, training, validation, etc.).
+
+[BoAmps](https://github.com/Boavizta/BoAmps) is an open project by [Boavizta](https://boavizta.org/) that defines a **standard format** for reporting these measurements, so that data coming from different teams, hardware, and models can be compared and aggregated. Each entry in this repository is a report following that format: a JSON file describing *what* was run, on *which* hardware, and *how much energy* it actually measured.
+
+A key principle: every report is based on **real, physical measurements** (via `nvidia-smi`, CodeCarbon, power meters, etc.). This is what makes the dataset a reliable, transparent basis for studying the footprint of AI.
+
+The rest of this document is a guide for contributors who want to add their own reports.
+
 # Guide: How to share your data on the BoAmps repository
 
-This guide explains step by step how to share BoAmps format reports on this public Hugging Face repository. 
+This guide explains step by step how to share BoAmps format reports on this public Hugging Face repository.
+
+It follows the journey of a contributor: first checking whether your data is eligible, then producing and validating a report, structuring your contribution, and finally submitting it.
 
 ## Table of Contents
+- [BoAmps open data repository](#boamps-open-data-repository)
+  - [What is this repository?](#what-is-this-repository)
 - [Guide: How to share your data on the BoAmps repository](#guide-how-to-share-your-data-on-the-boamps-repository)
   - [Table of Contents](#table-of-contents)
-  - [Prerequisites](#prerequisites)
-  - [Method 1: Hugging Face Web Interface](#method-1-hugging-face-web-interface)
-  - [Method 2: Git (Command Line)](#method-2-git-command-line)
+  - [Step 1: Is my data eligible?](#step-1-is-my-data-eligible)
+  - [Step 2: Create and validate a report](#step-2-create-and-validate-a-report)
+  - [Step 3: Structure your contribution](#step-3-structure-your-contribution)
+  - [Step 4: Submit your contribution](#step-4-submit-your-contribution)
+    - [Method 1: Hugging Face Web Interface](#method-1-hugging-face-web-interface)
+    - [Method 2: Git (Command Line)](#method-2-git-command-line)
+  - [If BoAmps doesn't fit your data](#if-boamps-doesnt-fit-your-data)
+  - [Useful links](#useful-links)
+
+## Step 1: Is my data eligible?
+
+Before investing time in a conversion, check that your data is a good fit for BoAmps:
+
+- **The energy must be measured, not estimated.** BoAmps needs real measurements taken with tools such as `nvidia-smi`, CodeCarbon, a power meter, etc. If the energy consumption is *estimated* or *calculated* (for example with EcoLogits) rather than physically measured, it is **not** compatible with BoAmps.
+- **The data must relate to AI.** The measurement should concern an AI workload: inference, training, validation, and so on.
+
+If your data does not meet both criteria, it is unfortunately not a fit for this repository.
+
+## Step 2: Create and validate a report
+
+Producing a valid report happens in three phases: mapping, conversion, and validation.
+
+**1. Mapping.** On paper or in a working file, before writing any report:
+- **Decide the scope of a report.** Depending on the source, it is often better to produce several reports with finer granularity — especially when the configuration (hardware, model) changes. Conversely, when there are many small consecutive measurements, a single report may be more appropriate. Helpful boundaries: the AI model changes, the dataset changes, the hardware changes, or the day changes (for a service measured over the long term, one report per day is reasonable). Sometimes it is as simple as *"one row of the CSV = one report"*.
+- **Map each source field to a BoAmps field.** Note any conversion or computation needed (in particular unit conversions for energy or duration).
+- **Identify static fields.** Some BoAmps fields may be identical across all your reports (e.g. hardware, publisher). Note them so you can reuse them.
+
+The BoAmps field definitions live in the [`model/` folder of the BoAmps repository](https://github.com/Boavizta/BoAmps/tree/main/model) — each JSON file defines part of the schema.
 
-## Prerequisites
+**2. Conversion.** From the mapping above, convert your source data into BoAmps reports using one of:
+- **Method 1 (recommended): write a Python conversion script.** See the [examples on GitHub](https://github.com/Boavizta/BoAmps/tree/main/examples). Keeping the script lets us regenerate the reports if the BoAmps model evolves.
+- **Method 2: write the reports manually** or via the [online form](https://huggingface.co/spaces/soprasteria/BoAmps_report_creation).
+- **Method 3:** if you are taking the measurements yourself, produce the reports directly from your measurement pipeline.
 
-Before starting, make sure you have:
-- A Hugging Face account
-- The files you want to upload
+**3. Validation.** Check that your reports are well-formed with the [JSON schema validator](https://github.com/Boavizta/BoAmps/tree/main/tools/schema_validator) before submitting. This avoids back-and-forth during review.
 
-## Method 1: Hugging Face Web Interface
+**File naming.** Name each report file clearly, following this format:
+
+`report_<publisher>_<taskStage>_<taskFamily>_<infraType>_<reportID>.json`
+
+- `<publisher>`: the name of the publisher (useful for gathering reports written by the same person — you may use a pseudonym if you wish to remain anonymous)
+- `<taskStage>`: mandatory field of the report
+- `<taskFamily>`: mandatory field of the report
+- `<infraType>`: mandatory field of the report
+- `<reportID>`: optional
+
+## Step 3: Structure your contribution
+
+Create a new subfolder inside `data/`, named after your source (for a public source, the name of the paper; for a private source, the organisation name plus a unique identifier for this dataset). That subfolder should contain:
+
+```
+data/<source-name>/
+├── reports/          (MANDATORY) one or more reports in BoAmps format
+├── source_info.md    (MANDATORY) a short description of the source
+├── source_data/      (OPTIONAL)  the raw source data before conversion
+└── conversion.py     (OPTIONAL)  the conversion script, if you used one
+```
+
+- **`reports/`** - the report(s) in BoAmps format, named as described above. **(mandatory)**
+- **`source_info.md`** - a brief description of: **(mandatory)**
+  - the type of data that was converted,
+  - the original authors of the data / attribution,
+  - the person(s) who performed the conversion,
+  - any conversion issues related to BoAmps (missing fields, data not detailed enough, etc.).
+- **`source_data/`** - the source data before conversion (CodeCarbon CSV, the paper's dataset, data extracted from figures and tables, etc.). *(optional)*
+- **`conversion.py`** - the conversion script, if the conversion was scripted. *(optional)*
+
+Including the source data and the conversion script helps us regenerate the reports if the BoAmps model changes.
+
+## Step 4: Submit your contribution
+
+### Method 1: Hugging Face Web Interface
 
 1. Log in to Hugging Face
 2. Go to [the boamps dataset](https://huggingface.co/datasets/boavizta/open_data_boamps)
@@ -27,25 +104,29 @@ Before starting, make sure you have:
 ![Access to files](screenshots/01-access-repository.png)
 ![Access to data folder](screenshots/02-access-data-files.png)
 4. Click on "Contribute" then "Upload files"
-![Contribute](screenshots/03-upload-files.png) 
-
-5. Drop your files in BoAmps format (please name them clearly) and give a name to the PR (e.g. 10 reports on image classification). You can add an extended description but this is optional.
-The name of the files should follow this format : 
-"report_\<publisher>_\<taskStage>_\<taskFamily>_\<infraType>_\<reportID>.json"
-- \<publisher>: the name of the publisher (this can be useful for gathering reports written by the same person, but you can of course create a pseudonym if you wish to remain anonymous)
-- \<taskStage>: mandatory field of the report
-- \<taskFamily>: mandatory field of the report
-- \<infraType>: mandatory field of the report
-- \<reportID>: optionnal 
+![Contribute](screenshots/03-upload-files.png)
 
-1. At the bottom of the page, click on "Open a Pull Request".
-2. You should see your PR created in "Community" > "Pull request". Now just wait for our team to validate your PR, thank you very much for your participation and your commitment to more frugal AI, in full transparency!
-![Consult](screenshots/04-consult-PR.png) 
+5. Upload your files following the structure from [Step 3](#step-3-structure-your-contribution). To place a file inside your source subfolder, prefix its name with the folder path (e.g. `<source-name>/reports/report_....json`) in the upload dialog. Give a name to the PR (e.g. *10 reports on image classification*). You can add an extended description, but this is optional.
+6. At the bottom of the page, click on "Open a Pull Request".
+7. You should see your PR created in "Community" > "Pull request". Now just wait for our team to validate your PR. Thank you very much for your participation and your commitment to more frugal AI, in full transparency!
+![Consult](screenshots/04-consult-PR.png)
 
-
-## Method 2: Git (Command Line)
+### Method 2: Git (Command Line)
 
 1. Clone the repository
 2. Create a branch
-3. Add your files
-4. Create a PR 
+3. Add your files, following the folder structure from [Step 3](#step-3-structure-your-contribution)
+4. Commit, push, and open a Pull Request
+
+## If BoAmps doesn't fit your data
+
+If, during a conversion, at least one software, hardware, algorithmic, or other characteristic (data characterisation, etc.) does not seem to map onto an existing BoAmps field, please [contact us](https://github.com/Boavizta/BoAmps/issues) to discuss possible evolutions of the model.
+
+## Useful links
+
+- **BoAmps model & tools (GitHub):** https://github.com/Boavizta/BoAmps
+- **Field definitions (schema):** https://github.com/Boavizta/BoAmps/tree/main/model
+- **Example reports & conversion scripts:** https://github.com/Boavizta/BoAmps/tree/main/examples
+- **JSON schema validator:** https://github.com/Boavizta/BoAmps/tree/main/tools/schema_validator
+- **Documentation (WIP):** https://boavizta.github.io/BoAmps/
+- **Online report creation form:** https://huggingface.co/spaces/soprasteria/BoAmps_report_creation