Davizig10jojo/BlazerNano-V2-Again
Davizig10jojo/BlazerNano-V2-Again is a 0.8 billion parameter instruction-tuned language model based on Qwen3-0.6B, developed by Davizig10jojo. It is fine-tuned primarily for Portuguese (PT-BR) conversations, incorporating a mix of conversational and distillation datasets. This model is optimized for text generation tasks, particularly in Portuguese, offering a compact solution for localized applications.
Loading preview...
Overview
Davizig10jojo/BlazerNano-V2-Again is a compact 0.8 billion parameter language model, fine-tuned from the Qwen3-0.6B base model. It is specifically designed for text generation, with a strong emphasis on the Portuguese language, particularly Brazilian Portuguese conversations.
Key Capabilities
- Portuguese Language Proficiency: The model's training data composition, with 50% from
Polygl0t/gigaverbo-v2-sft(PT-BR conversations), makes it highly capable in generating Portuguese text. - Instruction Following: Fine-tuned with a mix of conversational and distillation datasets, including
ManusagentsandDavizig10jojo/BlazerNano_V2, enhancing its ability to follow instructions. - Efficient Deployment: Trained using LoRA (r=4, alpha=8) and available in quantized GGUF formats (Q6_K, Q4_K_M) for efficient deployment on consumer hardware, including
llama.cpp.
Training Details
The model underwent a single epoch of training on 2x T4 GPUs with a learning rate of 0.0003. Its base model is Qwen/Qwen3-0.6B.
Good For
- Applications requiring text generation in Portuguese, especially conversational agents or content creation.
- Edge device deployment or scenarios with limited computational resources due to its small size and available GGUF quantizations.
- Developers looking for a specialized, instruction-tuned model for Portuguese language tasks.