qaztech-platform

QAZTECH-AZ104-Tutor-8B

README

License: apache-2.0

Quick start

Ollama: ollama run qaztech-platform/az104-tutor

Transformers:

python

from transformers import AutoModelForCausalLM, AutoTokenizerm = "qaztech-platform/QAZTECH-AZ104-Tutor-8B"tok = AutoTokenizer.from_pretrained(m)model = AutoModelForCausalLM.from_pretrained(m, torch_dtype="auto", device_map="auto")msgs = [{"role": "system", "content": "You are an expert engineering tutor. Reply in the student's language."},        {"role": "user", "content": "Give me a practice question on Azure RBAC"}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False, return_tensors="pt").to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=400)[0][ids.shape[1]:], skip_special_tokens=True))

Use enable_thinking=False: the model was trained without reasoning traces.

Training

Base: Qwen/Qwen3-8B (Apache-2.0). LoRA (r=16, alpha=32, all linear layers), 3 epochs, lr 1e-4, bf16, merged into the base weights.
Data: 5,107 synthetic tutoring dialogues (4,907 train / 200 held out), generated by Qwen3-30B-A3B-Instruct-2507 from Microsoft Azure documentation on role-based access control, virtual networks, Azure Resource Manager and App Service.
Modes: explanation 1,804; misconception correction 1,613; certification-style practice questions 1,690.
Languages: ru 1,149; kk 1,023; es 1,011; en 980; ar 944.
Filters: question and answer in the same language, no references to unseen source text, no mixed scripts, deduplication.
Held-out eval loss 0.795, token accuracy 0.773.

Limitations

Facts can be wrong: incorrect permission names and behaviours were observed in testing. Use with retrieval over official documentation (RAG) and verify against Microsoft Learn.
Kazakh and Arabic output has not yet been reviewed by native speakers; Kazakh phrasing can be awkward.
Coverage is limited to the four Azure documentation areas above.
Contains no real Microsoft exam questions. Not affiliated with or endorsed by Microsoft.

Licence and attribution

Weights: Apache-2.0. Base model: Qwen3-8B by Alibaba Cloud (Apache-2.0). Training data was derived from Microsoft Azure documentation (github.com/MicrosoftDocs/azure-docs), licensed under CC BY 4.0. Microsoft and Azure are trademarks of the Microsoft group of companies.

Trademark

QAZTECH is a registered trade mark of AISC Technologies LTD in the United Kingdom (UK IPO nos. UK00004263700, UK00004317681, UK00004373195). The Apache-2.0 licence does not grant rights to use the QAZTECH name or logo (section 6). Derivative models must not use "QAZTECH" in their name; describe them as "based on QAZTECH AZ-104 Tutor".

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

qaztech-platform

Model Tree

Base

Qwen/Qwen3-8B

Fine-tuned

this model

Input Modalities

Text

Output Modalities