Qwen3.6-35B-A3B Tool Slo — 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 model (35B MoE, 3B aktivnih) za zanesljive klice orodij skozi cel večkoračni pogovor, v Qwen XML formatu. Teče lokalno (llama.cpp, LM Studio) po združitvi; adapter LoRA. Sprejme tudi slike.

In short (EN): LoRA adapter (rank 128) for Qwen/Qwen3.6-35B-A3B: Slovenian continued pretraining + instruction SFT + agentic multi-turn tool-calling SFT (APIGen-MT, ToolACE, Hermes). Native Qwen3.6 XML tool calls. GGUF build: texdata/Qwen3.6-35B-A3B-Tool-Slo-GGUF.

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.

QLoRA adapter za slovenski model z zanesljivim klicanjem orodij (function / tool calling). Nanesi ga na Qwen/Qwen3.6-35B-A3B (rank 128, lora_target all, 4-bit bnb).

Kumulativna prilagoditev: slovenski CPT + instrukcijski/prevajalski SFT + multi-turn agentski tool-calling SFT na:

  • Salesforce / APIGen-MT-5k (multi-turn agentic)
  • Team-ACE / ToolACE
  • NousResearch Hermes function-calling

Native tool-calling v Qwen3.6 XML formatu; ohranja slovenščino skozi retencijski nabor.

Uporaba

LLaMA-Factory merge:

model_name_or_path: Qwen/Qwen3.6-35B-A3B
adapter_name_or_path: texdata/Qwen3.6-35B-A3B-Tool-Slo-LoRA
template: qwen3_5
finetuning_type: lora

ali PEFT: PeftModel.from_pretrained(base, "texdata/Qwen3.6-35B-A3B-Tool-Slo-LoRA")

GGUF (llama.cpp / LM Studio): texdata/Qwen3.6-35B-A3B-Tool-Slo-GGUF.

License

Base Qwen/Qwen3.6-35B-A3B is Apache-2.0. Training data carries its own terms: Slovenian Wikipedia (CC BY-SA 4.0), FineWeb2-sl (ODC-BY 1.0), APIGen-MT, ToolACE and Hermes function-calling datasets. Marked license: other; resolve data terms before commercial use.

Citation

Cite this work (Tadej Fius, MediaAtlas Ltd):

@misc{fius2026qwen35btoollora,
  title        = {Qwen3.6-35B-A3B Tool-Slo (LoRA)},
  author       = {Fius, Tadej},
  year         = {2026},
  publisher    = {MediaAtlas Ltd},
  howpublished = {Hugging Face},
  url          = {https://huggingface.co/texdata/Qwen3.6-35B-A3B-Tool-Slo-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}, author={Penedo, G. and others}, year={2024}, url={https://huggingface.co/datasets/HuggingFaceFW/fineweb-2}}
@article{liu2024apigen, title={APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets}, author={Liu, Zuxin and others}, journal={NeurIPS}, year={2024}}
@misc{hermes_fc, title={Hermes Function-Calling Datasets}, author={{NousResearch}}, url={https://huggingface.co/NousResearch}}

CPT: Slovenian Wikipedia (CC BY-SA), FineWeb2-sl (ODC-BY). Tools: Hermes + APIGen/xLAM.

Downloads last month
44
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for texdata/Qwen3.6-35B-A3B-Tool-Slo-LoRA

Adapter
(297)
this model