# Low-resource efficient inference

Samastam is an early instruction-tuned variant of the Sarvam-1 model, fine-tuned on the Alpaca-cleaned dataset to support multilingual instruction responses.

An inference-optimized model fine-tuned based on Falcon3-Mamba-7B-Instruct, utilizing the Mamba architecture for efficient inference

This is a statically quantized version based on behbudiy/Llama-3.1-8B-Instuct-Uz, supporting Uzbek and English, suitable for various text generation tasks.

CosmosLLaMa-Instruction DPO is a large language model specifically designed for Turkish text generation tasks, capable of continuing text fragments in a coherent and contextually appropriate manner.

TypeScript-specific code generation model fine-tuned from deepseek-coder-1.3b-base, supporting 16K context and Fill-in-the-Middle (FIM) functionality

Continuously trained on the MMLU dataset based on the Llama-2-7b-hf model to improve MMLU metrics while maintaining stability in other indicators