SaidzodaEng/Ameena_Qwen3-0.6B_e8

license: apache-2.0 base_model: Qwen/Qwen3-0.6B library_name: transformers pipeline_tag: text-generation tags: - tajik - central-asian - education - ameena - sovereign-ai - qwen3 - lightweight - edge-ai - offline - text-generation

Ameena-Qwen3-0.6B-Tajik-v8

A lightweight, fully fine-tuned Qwen3-0.6B model trained on 2B+ tokens of native Tajik educational content over 4 epochs. Optimized for offline, low-resource devices (4GB RAM) in rural Central Asia. Powers **Ameena.tj**’s mobile-first AI tutor.

Model Details

• Developed by: Saidzoda AI Research Lab (IT Park Tajikistan)
• Base model: Qwen/Qwen3-0.6B
• Training data: 2B+ tokens from textbooks, exams, literature, and technical manuals
• Languages: Tajik (primary), with cross-lingual support for Uzbek, Russian, English
• Epochs: 4
• License: Apache 2.0

Use Case

• Offline AI tutoring on Android phones
• Homework explanation in low-connectivity regions
• Lightweight course generation for teachers
• Edge deployment via llama.cpp or ONNX Runtime

Training Infrastructure

• Hardware: NVIDIA H100 (80GB VRAM)
• Provider: RunPod
• Precision: BF16 mixed precision
• Duration: ~60 hours

Environmental Impact

• Cloud Provider: RunPod (EU data centers)
• Region: Europe
• Carbon Emitted: ~0.6 kg CO₂eq
  (Estimated via ML CO2 Impact Calculator)

How to Use

With Transformers (online)

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "SaidzodaEng/Ameena_Qwen3-0.6B_e8"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto"
)
For Offline (GGUF version coming soon)

This model is being quantized to GGUF format for CPU-only inference. Follow our org for updates.

Out-of-Scope Use

Not for medical, legal, or financial advice
Not intended for high-stakes decision-making
Citation

APA:
Saidzoda AI Research Lab. (2025). Ameena_Qwen3-0.6B_e8. Hugging Face.

BibTeX:
@misc{saidzoda_ameena_qwen3_06b_2025,
  author = {Saidzoda AI Research Lab},
  title = {Ameena-Qwen3-0.6B-Tajik-v8},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/SaidzodaEng/Ameena_Qwen3-0.6B_e8}}
}