SaidzodaEng/Ameena_Qwen3-8B_e3

Ameena-Qwen3-8B-Tajik-v3

A fully fine-tuned Qwen3-8B model trained on 10B+ tokens of native Tajik, Uzbek, and Russian educational content over 3 epochs. Powers the core AI tutor in Ameena.tj — Central Asia’s first sovereign learning platform.

Model Details

• Developed by: Saidzoda AI Research Lab (IT Park Tajikistan)
• Base model: Qwen/Qwen3-8B
• Training data: 10B+ tokens from textbooks, exams, literature, and technical manuals
• Languages: Tajik (primary), Uzbek, Russian, English
• Epochs: 3
• License: Apache 2.0

Use Case

• AI-powered course generation
• Step-by-step tutoring (explains, doesn’t solve)
• Homework analysis & certification
• Offline inference on consumer devices

Training Infrastructure

• Hardware: Nvidia H200
• Framework: Hugging Face Transformers + Accelerate
• Precision: BF16 mixed precision
• Duration: ~220 hours

Environmental Impact

• Cloud Provider: Runpod
• Region: europe-west4
• Carbon Emitted: ~2.2 kg CO₂eq
  (Estimated via ML CO2 Impact Calculator)

How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "SaidzodaEng/Ameena_Qwen3-8B_e3"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto"
)

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_Qwen3-8B_e3. Hugging Face.

BibTeX

@misc{saidzoda_ameena_qwen3_2025,
  author = {Saidzoda AI Research Lab},
  title = {Ameena-Qwen3-8B-Tajik-v3},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/SaidzodaEng/Ameena_Qwen3-8B_e3}}
}