Capabilities

• Replies in 26 languages: Spanish, English, German, French, Italian, Portuguese, Dutch, Polish, Russian, Chinese, Japanese, Korean, Arabic, Hindi, Turkish, Swedish, Danish, Finnish, Czech, Romanian, Hungarian, Greek, Hebrew, Thai, Vietnamese, Indonesian
• Factual QA: Answers questions about capitals, science definitions, geography, and general knowledge
• Preserved reasoning: Retains the original reasoning capabilities of the base model

Languages & Performance

Usage

`python from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('admijgjtjtjtjjg/supra-50m-multilingual') tokenizer = AutoTokenizer.from_pretrained('admijgjtjtjtjjg/supra-50m-multilingual')

prompt = 'Q: What is the capital of France?' inputs = tokenizer(prompt, return_tensors='pt') outputs = model.generate(**inputs, max_new_tokens=50) print(tokenizer.decode(outputs[0])) `

Or use with PEFT (LoRA adapter):

python from peft import PeftModel base = AutoModelForCausalLM.from_pretrained('SupraLabs/Supra-50M-Reasoning') model = PeftModel.from_pretrained(base, 'admijgjtjtjtjjg/supra-50m-multilingual')

Training Details

• Base Model: SupraLabs/Supra-50M-Reasoning (51.8M params)
• Method: LoRA (r=32, alpha=64, dropout=0.1)
• Trainable params: 1.37M (2.59%)
• Dataset: 1855 examples (multilingual QA + factual QA)
• Epochs: 6
• Hardware: CPU
• Best eval_loss: 0.839