# Community GPU grant — Kurdish (Kurmanji) open-source model **URL:** **Author:** ![muzaffercky](https://avatars.discourse-cdn.com/v4/letter/m/e274bd/32.png) [@muzaffercky](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmuzaffercky)%5C **Post date:** [October 23, 2026, 7:59pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fcommunity-gpu-grant-kurdish-kurmanji-open-source-model%2F190958%2F1 "2026-10-23T19:59:43Z") Hi Hugging Face team, I’m an independent researcher building open-source NLP for Kurdish (Kurmanji), a low-resource language. I’m not affiliated with a university or company, but all my work is public and reproducible. Published work ([muzaffercky (Muzaffer Cıkay)](http://huggingface.co/muzaffercky)): - Sorjîn — a Kurmanji instruction-tuned chat model - kurdish-instruct — a Kurmanji instruction-tuning dataset - kurdish-web — a deduplicated web corpus, ~676k documents (Kurmanji, Sorani, Zazaki) - kurdish-theses-yok — 447 Kurdish academic theses - kurdish-kurmanji-news — 271k Kurmanji news articles Pipelines ([cikay (Muzaffer Cıkay) · GitHub](http://github.com/cikay)): kurdish\_scrapy, dedup, voice\_dataset\_pipeline kurdish\_ins I want to host a public demo of Sorjîn on a Space so people can try it, and I’d like to request a GPU community grant for this Space. Two questions: 1. Am I eligible for a grant as an independent (non-university) researcher? 2. Can I call the Space’s API from my own app’s backend to serve my free users, or is the grant only for the Space page itself? Thank you, Muzaffer Cıkay Hugging Face: muzaffercky ---
**Author:** ![John6666](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/john6666/32/27664_2.png) [@John6666](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FJohn6666)%5C **Post date:** [October 26, 2026, 7:49am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fcommunity-gpu-grant-kurdish-kurmanji-open-source-model%2F190958%2F2 "2026-10-26T07:49:52Z")
Although many of the policy questions, especially those concerning external apps, are ultimately for Hugging Face to decide, a few relevant details are already documented publicly, so here is what I found about those: * * * **1. Applying as an independent researcher** The [HF-maintained Community GPU grants guide](https://github.com/huggingface/skills/blob/main/skills/huggingface-spaces/references/grants.md) explicitly includes a **Personal project** category, with open research demos among the examples. It does not make university or company affiliation a prerequisite for applying. That is a route to _request_ support, of course, rather than a promise of approval. There is also a fairly recent [Personal-project example from July 2026](https://huggingface.co/spaces/kehong/MoCapAnythingV2/discussions/1) in which HF staff confirmed a ZeroGPU assignment. It is a useful illustration of the category, not a precedent for what hardware or terms another project would receive. **2. Calling a Space from your own app** For a **Gradio Space** , the technical part is supported: HF’s [Spaces as API endpoints](https://huggingface.co/docs/hub/spaces-api-endpoints) documentation covers Python, JavaScript, HTTP and an example of calling a Space from a **FastAPI backend**. The Space’s **Use via API** tab shows the actual endpoints and parameters, so the app does not have to interact with the demo through its web page. Whether _a particular community grant_ also covers ongoing inference for users of a separate free app is a different question. I couldn’t find a public rule that conclusively answers that specific use case. It seems best to mention the intended external-app use in the grant request and leave the decision to HF, without assuming either permission or prohibition from the existence of an API. **3. Finding the application route** The [Hub’s GPU grant documentation](https://huggingface.co/docs/hub/spaces-gpus#community-gpu-grants) points to the application in the **Space’s Settings / Hardware** area. An [HF staff reply from August 2025](https://huggingface.co/spaces/ZhengGeng/OnePoseviaGen/discussions/1) specifically recommended that button because it notifies the reviewers. The [HF-maintained grants guide](https://github.com/huggingface/skills/blob/main/skills/huggingface-spaces/references/grants.md) also describes making a request in the **Space’s Community tab** , titled `Apply for a GPU community grant: Personal project`. Since both instructions are published, I would check the current Settings application first and use the Space Discussion route if appropriate. I wouldn’t assume that this general Forum thread itself submits a grant application, or that both routes must be used. For context, a small Space linking the [Sorjîn model](https://huggingface.co/muzaffercky/Sorjin1.1-7B-Instruct) and relevant datasets/code, with a brief description of the public demo and its GPU needs, should give the request a concrete subject. The existing public work can be linked rather than repeated in full. > **Two practical distinctions that might save some time** > > **A free ZeroGPU prototype is possible in some accounts, but is not the same thing as a dedicated GPU grant.** The current [ZeroGPU guide](https://huggingface.co/docs/hub/spaces-zerogpu) allows free personal accounts in good standing (verified email, account older than 30 days) to host up to two ZeroGPU Spaces. ZeroGPU currently uses **Gradio** and has its own supported runtime, scheduling and usage limits. Sorjîn’s model card includes a Transformers inference example, but that alone does not establish that this particular model has been tested successfully on ZeroGPU. > > If a free ZeroGPU slot is available, a short generation and one request from **Use via API** could be a low-cost way to check compatibility. If ZeroGPU is unavailable or unsuitable for the model or intended workload, that is simply context for considering a dedicated grant; it need not become an obstacle to asking. The [Spaces Overview](https://huggingface.co/docs/hub/spaces-overview#creating-a-new-space) also notes an easy-to-miss account rule: creating ordinary Gradio/Docker Spaces generally requires a paid plan, with an exception for qualifying free ZeroGPU Spaces. The [grants guide](https://github.com/huggingface/skills/blob/main/skills/huggingface-spaces/references/grants.md) mentions a free Static Space as a possible place to stage an application when the usual compute Space cannot yet be created. > > **If the app uses ZeroGPU, the API caller’s quota matters.** According to the [API documentation](https://huggingface.co/docs/hub/spaces-api-endpoints#zerogpu-spaces), authenticated requests consume the ZeroGPU allowance of the account whose HF token is used. Thus, if an app backend sends every user’s request under one developer-owned token, those requests share that account’s allowance; they do not each gain a separate quota merely because the app has different users. This is a **ZeroGPU-specific** consideration, not a limit that should automatically be attributed to every dedicated grant-supported GPU. The [current usage tiers](https://huggingface.co/docs/hub/spaces-zerogpu#usage-tiers) are worth checking rather than relying on old figures. > > For a first connection test, the Space’s **Use via API** examples are probably enough. A Python backend can also inspect its endpoint names with `gradio_client.Client(...).view_api()`. Keep any necessary HF token server-side, rather than exposing it in browser code or a public repository. So the public documentation seems to answer the **personal-project application path** and the **technical API integration** fairly well. The intended scope of the GPU grant for a separate app is the part I would leave to HF. Hopefully these links help with the remaining practical details.