from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ghzlmc/nemotron-3-nano-30b-unified"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)

Training

Fine-tuned from nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 with TRL SFT and exported as merged safetensors shards.

The repository includes the custom Nemotron-H configuration/modeling files required by trust_remote_code=True.