SKIS-AI-Research/ODIN-C1

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:May 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ODIN-C1 (On-Device accelerated Intelligent Network-Code 1) is a 1.2 billion parameter small language model developed by SKIS-AI-Research, based on the LFM architecture. Engineered for on-device programming, it offers hardware-agnostic inference capabilities. This model is specifically optimized for code generation tasks, providing an efficient solution for embedded and edge computing environments.

Loading preview...

Overview

SKIS-AI-Research's ODIN-C1 is a 1.2 billion parameter small language model (sLM) built upon the LFM architecture. Its name, "On-Device accelerated Intelligent Network-Code 1," highlights its core design principle: efficient code generation for on-device applications. A key differentiator of ODIN-C1 is its hardware-agnostic inference affinity, which allows for flexible deployment across various hardware platforms.

Key Capabilities

  • On-Device Programming: Specifically engineered for generating code in resource-constrained environments.
  • Hardware-Agnostic Inference: Designed to provide consistent performance regardless of the underlying hardware.
  • Code Generation: Optimized for tasks requiring the creation of programming scripts, as demonstrated by its ability to generate Python code for sequences like Fibonacci.

Training Details

ODIN-C1 was developed using Supervised Fine-Tuning (SFT) techniques. The training process leveraged a mix of datasets including unsloth/alpaca-cleaned, uniquealexx/Kimi-K2.6-Thinking-200x, nvidia/Nemotron-Cascade-2-SFT-Data, iamtarun/python_code_instructions_18k_alpaca, Entity-27th/ODIN-C1-SFT-Mix, and Entity-27th/ODIN-C1-Realign. The model was trained efficiently using Unsloth and Huggingface's TRL library on a single AMD Instinct MI300X accelerator and a GeForce RTX 4070 Laptop GPU.

Good For

  • Developers working on embedded systems or edge devices requiring local code generation capabilities.
  • Applications where efficient, hardware-flexible code generation is critical.
  • Tasks involving the creation of Python scripts and similar programming instructions.

Finetunes
2 models