State-of-the-art Small Language Coder Model: Mify-Coder

Infosys AI Research | Mify Team · 2025

Read State-of-the-art Small Language Coder Model: Mify-Coder by Infosys AI Research | Mify Team (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-23747.txt.

Description

A very large author list from the Mify project presents Mify-Coder, a code-generating language model with two and a half billion parameters, small by current standards. It was trained on several trillion tokens starting from the Mify-2.5B foundation model, using a compute-optimal strategy designed to squeeze frontier-grade coding ability into a compact network that can run on ordinary desktops once quantized.

The training pipeline mixes carefully curated code with synthetic data produced through agent-designed prompts, filtered for quality by language models and refined against enterprise-grade evaluations. Continual pretraining and supervised fine-tuning were explored within one continuous trajectory, tuning objectives, data mixtures, and sampling together. The authors report accuracy and safety comparable to much larger baselines on standard coding and function-calling benchmarks, arguing that disciplined data and compute choices let small models punch above their size.

Catalog record

CREATORInfosys AI Research | Mify Team
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-23747

Browse the full library · About the Tower of Babel