(v3 preview check out there and report me any bugs: https://huggingface.co/14ai/Polando-v3-1.7b-THINKING-PREVIEW)
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
}