The hardware options for the HF Hub GUI are managed openly on GitHub, so the only way to change them is to modify GitHub or have someone else modify it. Conversely, that’s all it takes to complete the process.
What controls the “My Hardware” dropdown on Hugging Face Hub
That dropdown is backed by a curated hardware list in huggingface/huggingface.js, specifically:
packages/tasks/src/hardware.ts (GitHub)
That file defines a small “hardware spec” object per device. It is intentionally approximate and uses:
So you do not need to add your whole spec sheet. In practice, for a GPU entry you mainly add:
tflops (FP16-ish number they use in this file)memory (VRAM in GB)
What you should add for “AMD Radeon AI PRO R9700”
From AMD’s official datasheet and product page:
- Dedicated memory size: 32 GB
- Peak FP16 vector performance: 95.7 TFLOPS (AMD)
Minimal entry (the shape used in this file is “name → { tflops, memory }”):
"AMD Radeon AI PRO R9700": {
tflops: 95.7,
memory: [32],
},
Why I’m using 95.7:
- The file’s own guidance is “FP16 whenever possible for GPUs.” (Hugging Face Forums)
- AMD publishes multiple “peak” numbers (FP16 vector, FP16 matrix, sparse). The safest match to the file’s “FP16 baseline” intent is the plain FP16 vector value. (AMD)
Optional improvement (often helpful in dropdowns):
- Add an alias if you suspect users might search without “AMD”, e.g. also add
"Radeon AI PRO R9700" with the same numbers. This reduces “can’t find my hardware” reports.
Option 1: Open an Issue (fastest, no coding)
Use this if you want maintainers to confirm naming or placement first.
Steps (GitHub UI)
GitHub’s standard flow is:
Issue template (paste)
Title: Add AMD Radeon AI PRO R9700 to hardware list (My Hardware dropdown)
Context
Hugging Face Hub “My Hardware / Local Apps and Hardware” list is backed by:
packages/tasks/src/hardware.ts in huggingface/huggingface.js
Request
Please add:
"AMD Radeon AI PRO R9700": { tflops: 95.7, memory: [32] }
Why these numbers
hardware.ts uses approximate FP16 for GPUs.
AMD official specs:
- FP16 Vector: 95.7 TFLOPS
- VRAM: 32 GB
Sources
AMD datasheet PDF: https://www.amd.com/content/dam/amd/en/documents/partner-hub/radeon-pro/radeon-ai-pro-r9700-datasheet.pdf
AMD product page: https://www.amd.com/en/products/graphics/workstations/radeon-ai-pro/ai-9000-series/amd-radeon-ai-pro-r9700.html
Hugging Face community members have been pointing people to “submit an issue or PR on GitHub” for adding items to that dropdown. (Hugging Face Forums)
Option 2: Open a PR (the change actually lands)
A. Easiest PR method: edit in the browser
-
Open packages/tasks/src/hardware.ts in the repo. (GitHub)
-
Click the pencil icon Edit (GitHub will ask you to fork if needed).
-
Add the entry in the right section.
- Keep the list ordering consistent (usually alphabetical in these tables).
-
Commit to a new branch in your fork.
-
Click Compare & pull request and submit.
GitHub’s “PR from a fork” flow is documented here. (GitHub Docs)
B. Local dev PR method (useful if CI complains)
If formatting or lint fails in CI, run the repo formatter before committing. The project documents common pnpm scripts including pnpm format. (Hugging Face)
Typical flow:
git clone https://github.com/<you>/huggingface.js.git
cd huggingface.js
git checkout -b add-r9700-hardware
# edit packages/tasks/src/hardware.ts
pnpm install
pnpm format
git add packages/tasks/src/hardware.ts
git commit -m "Add AMD Radeon AI PRO R9700 to hardware list"
git push -u origin add-r9700-hardware
Then open a PR from your branch to huggingface/huggingface.js.
PR description template
Adds AMD Radeon AI PRO R9700 to packages/tasks/src/hardware.ts for the Hub “My Hardware” dropdown.
Entry:
"AMD Radeon AI PRO R9700": { tflops: 95.7, memory: [32] }
Rationale:
hardware.ts uses approximate FP16 for GPUs.
AMD specs list FP16 Vector = 95.7 TFLOPS and VRAM = 32GB.
Sources:
https://www.amd.com/content/dam/amd/en/documents/partner-hub/radeon-pro/radeon-ai-pro-r9700-datasheet.pdf
https://www.amd.com/en/products/graphics/workstations/radeon-ai-pro/ai-9000-series/amd-radeon-ai-pro-r9700.html
Notes on your big spec list
Your details (bus width, Infinity Cache, bandwidth, ray accelerators, INT4 TOPS, sparsity) are real specs, but the Hub hardware dropdown list is not trying to model all of that. It’s mainly:
You can include the extra specs in the issue body as “additional context”, but keep the actual code change small.
Quick summary
- Edit
huggingface.js → packages/tasks/src/hardware.ts. (GitHub)
- Add:
"AMD Radeon AI PRO R9700": { tflops: 95.7, memory: [32] } using AMD official specs. (AMD)
- Open Issue via repo → Issues → New issue. (GitHub Docs)
- Or open a PR from a fork (browser edit is simplest). (GitHub Docs)