Spaces:

---

Duplicated from  agents-course/First_agent

agents-course

/

First_agent_template

Like 762

Running

yeah i got the same problem

yes same here,the CPU basic version is not available. Can someone help out with this?

I have the same issue, the duplicate button is grayed out and no free cpu version

Same here

Same issue. Are there any updates? Thanks

So as an update, I set up the environment locally, using Ollama, later using a free Gemini key (get one for free on Google AI Studio),just check the models, and how many API requests you can do each day. Hope this one helps a few.

Six people in this thread, so here is what actually changed, measured rather than guessed.

Free CPU has not been removed, but it has nearly disappeared for new copies of this template.

I have a census of all 1,458,692 public Hugging Face Spaces, so I can find every copy of First_agent_template (28,689 of them) and re-query its hardware live. Sampling 30 copies from each of the last three months:

In August, 24 of 30 were on zero-a10g (ZeroGPU) instead. So this is not imagined and not user error, but it is also not absolute — a small number of copies are still landing on free CPU, which is why a few people report it working.

This causes a second failure that looks unrelated. The template is plain CPU smolagents code with no @spaces.GPU decorated function. ZeroGPU refuses to start such a Space, with No @spaces.GPU function detected during startup — that is the error being reported separately in #737. The breakage rate of new copies tracks the hardware shift almost exactly:

Of 35 recently-broken copies I sampled, 33 were on ZeroGPU. Of 35 that work, 21 were on cpu-basic.

What you can do

1. Open your Space Settings and try to set hardware to CPU basic. Two of thirty August copies have it, so it is worth checking before assuming.
2. If you are stuck on ZeroGPU, adding any trivially decorated function satisfies the startup check:

import spaces

@spaces.GPU
def _noop():
    return None

3. @csongorosz 's local route above (Ollama, or a free Gemini key) is a legitimate way to finish Unit 1 without a Space.

What I cannot tell you: why the hardware options changed. I only observe what the API reports after the fact, so I cannot say whether this is a pricing change, a quota change, a rollout, or a bug. That question needs someone at Hugging Face.

I have reported the measurements upstream at https://github.com/huggingface/agents-course/issues/733 so the course maintainers have the numbers.

I do not work for Hugging Face. Method and code: https://github.com/ashishsinha1602/dataset-integrity-audit