from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Mathieu-Thomas-JOSSET/qwen2.5-3b-instruct-text-merged-16bit-20260128_033527", trust_remote_code=True)
tok = AutoTokenizer.from_pretrained("Mathieu-Thomas-JOSSET/qwen2.5-3b-instruct-text-merged-16bit-20260128_033527", trust_remote_code=True)

Inference Providers (serverless)

This works ONLY if the model appears in the router catalog (GET https://router.huggingface.co/v1/models). If not listed, use a dedicated deployment (Inference Endpoints) or self-hosted inference server.