Qwen3.6-35B-A3B-Slovenian-LoRA

Qwen3.6-35B-A3B — Slovenian CPT+SFT LoRA adapter

Trained by MediaAtlas — LLM fine-tuning on your own data, trained in the EU, weights delivered. Pricing · All our models

Na kratko: Slovenski jezikovni model (35B MoE, 3B aktivnih) za pogovor, znanje in prevajanje EN ⇄ SL, z razmišljanjem. Glede na neuglašeno osnovo: slovenski nabor acc_norm 0,623 → 0,654, prevajanje BLEU EN→SL 23,8 → 26,3 in SL→EN 30,9 → 35,0; adapter LoRA. Sprejme tudi slike.

Vision / slike: the adapter was trained on text only; after merging, the base model's vision encoder still handles image input. Image understanding was not evaluated in Slovenian.

LoRA adapter for Qwen/Qwen3.6-35B-A3B adding Slovenian fluency, knowledge, and en↔sl translation (continued-pretraining + SFT, both baked into this one adapter). Merging it into the base reproduces the full model.

On held-out evals vs the untuned base: Slovenian-LLM-Eval acc_norm 0.623 → 0.654; translation BLEU en→sl 23.8 → 26.3, sl→en 30.9 → 35.0. Reasoning model.

💡 For plug-and-play chat, a GGUF build (LM Studio / llama.cpp) is at texdata/Qwen3.6-35B-A3B-Slovenian-GGUF.

How to run — load in 4-bit

This qwen3_5_moe arch has a broken fp16/bf16 forward in current transformers (zero hidden state → constant logits → garbage). Load in 4-bit (bitsandbytes nf4) — the quantization it was trained under. (vLLM does not run this arch on Blackwell sm_120 GPUs.)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel

bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                         bnb_4bit_compute_dtype=torch.bfloat16, bnb_4bit_use_double_quant=True)
tok  = AutoTokenizer.from_pretrained("Qwen/Qwen3.6-35B-A3B", trust_remote_code=True)
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.6-35B-A3B", quantization_config=bnb,
                                            trust_remote_code=True, device_map="auto")
model = PeftModel.from_pretrained(base, "texdata/Qwen3.6-35B-A3B-Slovenian-LoRA")

# reasoning model: enable_thinking=False for direct answers/translation (else it spends tokens on <think>)
msg  = [{"role": "user", "content": "Prevedi v slovenščino: Good morning!"}]
text = tok.apply_chat_template(msg, add_generation_prompt=True, tokenize=False, enable_thinking=False)
out  = model.generate(**tok(text, return_tensors="pt").to(model.device), max_new_tokens=128)
print(tok.decode(out[0], skip_special_tokens=True))

License & data provenance

Base Qwen/Qwen3.6-35B-A3B is Apache-2.0. CPT data is open-licensed → this repo is license: other:

Attribute the above CPT sources and comply with their terms before commercial use.

Citation

Cite this work (Tadej Fius, MediaAtlas Ltd):

@misc{fius2026qwen35bsllora,
  title        = {Qwen3.6-35B-A3B Slovenian (LoRA)},
  author       = {Fius, Tadej},
  year         = {2026},
  publisher    = {MediaAtlas Ltd},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/texdata/Qwen3.6-35B-A3B-Slovenian-LoRA}
}

Upstream / source citations:

@misc{qwen3, title={Qwen3 Technical Report}, author={{Qwen Team}}, year={2025}, url={https://huggingface.co/Qwen}}
@misc{wikipedia_sl, title={Slovenian Wikipedia}, author={{Wikimedia Foundation}}, note={CC BY-SA 4.0}, url={https://sl.wikipedia.org}}
@misc{penedo2024fineweb2, title={FineWeb2: A sparkling update with 1000s of languages}, author={Penedo, G. and others}, year={2024}, url={https://huggingface.co/datasets/HuggingFaceFW/fineweb-2}}

CPT data: Slovenian Wikipedia (CC BY-SA 4.0) and FineWeb2-sl (ODC-BY 1.0).