14ai/Polando-v2-1.5b

Polando 1.5B

Fine-tuned on high-quality Polish conversational datasets (such as Emplocity/Owca), this updated release delivers better polish language and better logic than Polando-v1-1.5b.

Key Improvements

• Better polish than Polando-v1-1b
• The model FINALLY can do greeting like hello or something..
• Really better logic than Polando v1 1b.
• He is a finetuned Qwen so he speak he's name is Qwen.

---

Technical Specifications

Parameter	Specification
Base Architecture	Qwen2.5-1.5B-Instruct
Format	Merged FP16 (Full Precision)
File Size	~3 GB (model.safetensors)
Dependencies	Standard transformers library (Native execution; no PEFT/LoRA required)

---

Recommended Generation Parameters

To achieve the best response quality and maintain optimal performance, we recommend using the following configuration:

generation_config = {
    "max_new_tokens": 512,        # Adjust based on your needs
    "min_new_tokens": 1,
    "do_sample": True,
    "temperature": 0.2,           # Recommended range: 0.1 - 0.3 for precise responses
    "top_p": 0.85,
    "repetition_penalty": 1.15
}