Sara Han Díaz

I guess the tag is generated during completion. However, this might depend on the prompt, the max number of tokens, and how the inference is performed.

Hi @3rica ! Did you check the examples folder, for instance, this one?: https://github.com/argilla-io/synthetic-data-generator/blob/main/examples/ollama-different-model-for-completion.py. Let me know if you still have issues and think that a more explanatory document would make sense.

When setting up the HF_TOKEN, did you ensure you granted access to the Inference Endpoints?

Hi @Socialmediaprophet ! Sorry that I missed this message. I see that it's running now 🙌. There were some days when the Hub was a bit unstable, so that might have been the root cause of the connection error.

Hi @beketm ! The main reason is that we directly used the generated completions, but it's true that I missed writing the initial tag.

Hi! @Aristo2333 ! Regarding DeepSeek, we're using the distilled version, which is available through the Serverless Inference API (https://huggingface.co/docs/api-inference/index), which means that further configuration is not required. Below the Provider indicates that it's available.

The Llama issue is raised cause you need to go to the original repository and request access. After approval (in general, quite quick) you'll be able to use the model via the Serverless Inference API too.