jpalmae/qwen35b-espanol-abl-merged-v3

TEXT GENERATIONConcurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 19, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

jpalmae/qwen35b-espanol-abl-merged-v3 is a 35.1 billion parameter multimodal Qwen3.5/3.6 model developed by jpalmae, specifically fine-tuned for neutral and Chilean Spanish. This model incorporates a unique anti-Rioplatense filtering during training, making it distinctively optimized for specific Spanish dialects. It maintains multimodal capabilities, including vision configuration and support for image/video tokens, making it suitable for applications requiring nuanced Spanish language understanding and multimodal input processing.

Loading preview...

Model Overview

The jpalmae/qwen35b-espanol-abl-merged-v3 is a 35.1 billion parameter multimodal language model, built upon the Qwen3.5/3.6 35B-A3B architecture. Developed by jpalmae, this model is a merged version that has undergone specialized training to cater to neutral and Chilean Spanish, with a notable anti-Rioplatense filtering applied during its dataset preparation.

Key Capabilities & Features

  • Multimodal Architecture: Retains full multimodal capabilities, including vision_config and support for image/video tokens, allowing for processing of diverse input types.
  • Spanish Dialect Specialization: Fine-tuned with a custom dataset of 200,000 examples, including 60,000 Chilean, 60,000 neutral, and 80,000 anti-Rioplatense filtered examples, ensuring high relevance for target dialects.
  • Training Details: Utilizes LoRA (r=64, alpha=128) for efficient fine-tuning over 2,500 steps, achieving a final loss of approximately 2.11.
  • Deployment Ready: Designed for deployment with vLLM, indicating optimized inference performance.

When to Use This Model

This model is particularly well-suited for applications requiring:

  • Accurate Spanish Generation: Specifically for neutral and Chilean Spanish contexts, where dialectal nuances are important.
  • Multimodal Understanding: For tasks that involve both text and visual inputs in Spanish.
  • Regional Content Creation: Ideal for generating content, chatbots, or virtual assistants tailored to Chilean or general neutral Spanish-speaking audiences, avoiding Rioplatense linguistic characteristics.