i-Coder/iCoder-27B-SFT

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 25, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

iCoder/iCoder-27B-SFT is a 27 billion parameter language model, an intermediate checkpoint derived from Qwen3.6-27B, specifically from the supervised fine-tuning (SFT) stage of the iCoder-27B training pipeline. This model is designed for research into RTL design and GPU kernel optimization, representing a foundational step before further optimization stages. It is intended for pipeline reproduction, ablation studies, and measuring the impact of subsequent training phases. Its primary strength lies in its specialized training for hardware description languages and GPU code.

Loading preview...

iCoder-27B-SFT: An Intermediate Checkpoint for Hardware Optimization

iCoder-27B-SFT is a 27 billion parameter model developed by i-Coder, serving as an intermediate checkpoint within the broader iCoder-27B training pipeline. This specific version is the result of the supervised fine-tuning (SFT) stage, building upon the base model Qwen3.6-27B. The full iCoder-27B model is ultimately designed for advanced tasks in RTL design and GPU kernel optimization, with its development driven by an agent that iteratively refines its own training process.

Key Characteristics

  • Pipeline Stage: Represents the initial supervised fine-tuning phase of the iCoder-27B development.
  • Base Model: Inherits its foundation from Qwen3.6-27B.
  • Specialization Focus: Part of a larger effort to create models proficient in hardware description languages (like Verilog) and GPU code optimization.
  • Context Length: Supports a context window of 32768 tokens.

Intended Use Cases

This checkpoint is primarily intended for research purposes related to the iCoder-27B training pipeline. Specific use cases include:

  • Reproducing the supervised fine-tuning stage.
  • Ablation studies to understand the contribution of this stage.
  • Measuring the incremental value added by subsequent training phases (OPSD and RLVR) that lead to the final iCoder-27B model.

It is important to note that iCoder-27B-SFT is a mid-pipeline artifact and has not undergone deployment preparation.

Finetunes
1 models