GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Sup2Doggie
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Sup2Doggie
Model Tree
google/gemma-4-31B-it
Input Modalities
Output Modalities
Supported Functionality
Experimental Indonesian/English assistant for the MindSeek AI project.
Built-in demonstration set (~35 short hand-written Indonesian/English examples incl. identity, math, coding, summarization, structured output, refusal and uncertainty). Too small for meaningful capability gains.
Smoke test of 9 prompts, base vs fine-tuned, greedy decoding. Final train loss 3.7856, val loss 3.2065. These are not benchmark scores. Lower loss does not imply broader capability.
May produce incorrect or fabricated information; fine-tuning does not guarantee factual accuracy. No tool or internet access.
python
import torchfrom transformers import AutoModelForMultimodalLM, AutoTokenizer, BitsAndBytesConfigfrom peft import PeftModelbnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.bfloat16)base = AutoModelForMultimodalLM.from_pretrained("google/gemma-4-31B-it", quantization_config=bnb, dtype=torch.bfloat16, device_map="auto")model = PeftModel.from_pretrained(base, "Sup2Doggie/mindseek-prime-4.6")tok = AutoTokenizer.from_pretrained("Sup2Doggie/mindseek-prime-4.6")msgs = [{"role": "user", "content": "Halo, kamu siapa?"}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False, return_tensors="pt", return_dict=True)["input_ids"].to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=200)[0][ids.shape[1]:], skip_special_tokens=True))