nischay185/konkani-qwen2-1.5b-v3-full

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 8, 2026Architecture:Transformer Featherless Exclusive Cold

The nischay185/konkani-qwen2-1.5b-v3-full is a 1.5 billion parameter Qwen2-based language model, developed by nischay185, specifically designed to improve reasoning and task-following capabilities in Konkani. This standalone model integrates V3 training, which uses an English-Pivoted Chain-of-Thought approach, while preserving Konkani language behavior. It offers a simplified deployment experience compared to its adapter-based predecessor, making it suitable for research and experimentation in low-resource Konkani NLP.

Loading preview...

Konkani Qwen2-1.5B V3 Full: Enhanced Reasoning for Konkani

This model, nischay185/konkani-qwen2-1.5b-v3-full, is a 1.5 billion parameter Qwen2-based language model developed by nischay185. It represents the standalone, full-weight version of the V3 model, merging the adapter weights directly into the base model. This eliminates the need to load the original Konkani SFT model or PEFT separately, simplifying deployment and inference.

Key Capabilities & Differentiators

  • Improved Reasoning: The V3 training focuses on enhancing useful reasoning and task-following, inspired by English-Pivoted Chain-of-Thought (CoT) for low-resource languages. This allows the model to use English as an intermediate reasoning language while providing final answers in Konkani.
  • Konkani Language Preservation: Training included a 50% replay of original Konkani SFT examples to ensure the model retains its Konkani language behavior and knowledge.
  • Simplified Deployment: Unlike the V3 adapter release, this full version can be loaded directly, streamlining the setup process for developers.
  • Targeted Training: The model was fine-tuned using LoRA/PEFT with a mixed dataset comprising original Konkani SFT examples and reasoning-oriented CoT examples.

Intended Use Cases

  • Research on low-resource language modeling, particularly for Konkani.
  • Experimentation in Konkani Natural Language Processing (NLP).
  • Studies on reasoning transfer in multilingual contexts.
  • Evaluation of Konkani language capabilities and academic experimentation.
  • Local inference and deployment of a specialized Konkani model.

Finetunes
1 models