julienp79/occitan-gemma-4-12b-it-lora

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 5, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

julienp79/occitan-gemma-4-12b-it-lora is a 12 billion parameter Gemma-4-IT model fine-tuned by julienp79 for the Occitan language. This model, the largest Occitan Gemma variant to date, was optimized using LoRA for efficient training on consumer hardware and offers enhanced reasoning and linguistic nuance in Occitan. It is designed for generating and processing text in Occitan, leveraging a balanced corpus of literary, journalistic, grammatical, and encyclopedic texts.

Loading preview...

Occitan Gemma-4-12B-IT (LoRA Merged)

This model is a fine-tuned version of Google's Gemma-4-12B-IT, specifically optimized for the Occitan language. Developed by julienp79, it represents the largest Occitan Gemma model available, offering significant improvements in reasoning and linguistic nuance for Occitan text generation.

Key Capabilities & Features

  • Occitan Language Specialization: Fine-tuned on a balanced corpus of Occitan texts, including literary, journalistic, grammatical, and encyclopedic content.
  • Efficient Training: Utilizes LoRA (Low-Rank Adaptation) for efficient fine-tuning, enabling training of a 12B model on an RTX 3060 (12GB VRAM) through surgical optimizations like paged_adamw_8bit, 4-bit NormalFloat quantization, and gradient checkpointing.
  • Gemma 4 Architecture: Built upon Google's Gemma 4 family, inheriting its strong reasoning capabilities.
  • Flexible Deployment: Available in full merged Safetensors weights for transformers and accelerate, as well as various GGUF quantized versions (Q2_K, Q4_K_M, Q5_K_M, Q8_0, f16) for local inference via tools like LM Studio, Ollama, or llama.cpp.

When to Use This Model

  • Occitan Text Generation: Ideal for applications requiring high-quality text generation, translation, or content creation in Occitan.
  • Linguistic Research: Useful for researchers studying the Occitan language, leveraging its enhanced linguistic nuance.
  • Resource-Constrained Environments: The optimized training and availability of quantized versions make it suitable for deployment on hardware with limited VRAM.

Finetunes
1 models