Kysaddy/Koda-v1

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Koda-v1 is a 1.5 billion parameter coding and conversational model developed by Kysaddy. This experimental model is a bfloat16 merged checkpoint, fine-tuned from Qwen/Qwen2.5-Coder-1.5B-Instruct. It is designed for code generation and general conversational tasks, offering direct loading without bitsandbytes quantization for standard Hugging Face Transformers.

Loading preview...

Koda-v1 Overview

Koda-v1 is an experimental 1.5 billion parameter model developed by Kysaddy, focusing on both coding and conversational capabilities. It is derived from a 100-step CPU QLoRA fine-tune of the Qwen/Qwen2.5-Coder-1.5B-Instruct base model.

Key Characteristics

  • Architecture: Based on Qwen/Qwen2.5-Coder-1.5B-Instruct, fine-tuned on HuggingFaceH4/Bespoke-Stratos-17k.
  • Parameter Count: 1.5 billion parameters, offering a balance between performance and efficiency.
  • Format: Provided as a standard bfloat16 merged checkpoint, allowing direct loading with Hugging Face Transformers without requiring bitsandbytes quantization.
  • Context Length: Supports a context window of 32768 tokens.

Intended Use and Limitations

Koda-v1 is an experimental checkpoint and not a fully converged production model. Users should:

  • Review generated code: Always verify code produced by the model before execution.
  • Adhere to Qwen license: Retain the original Qwen license, notices, and usage conditions.

This model is suitable for developers exploring code generation and conversational AI with a smaller, experimental model, particularly those working within the Hugging Face ecosystem.

Finetunes
1 models