minsore/pepper-1-preview

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 29, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Pepper 1 Preview by Minsore is a 1.5 billion parameter instruction-following code assistant built on Qwen2.5-Coder-1.5B, designed specifically for generating Python functions from natural language descriptions. With a 32K context length, it excels at interactive Python code generation, outperforming its base model on HumanEval by 7 points. This compact model is optimized for consumer hardware and is released under the Apache 2.0 license.

Loading preview...

Overview

Pepper 1 Preview is a 1.5 billion parameter code assistant developed by Minsore, specifically fine-tuned for instruction-following Python code generation. Built upon the Qwen2.5-Coder-1.5B architecture, it focuses on creating clean, idiomatic Python functions from natural language prompts. This model is designed for real-time, interactive use on consumer hardware, requiring approximately 1 GB of VRAM for full offload.

Key Capabilities & Performance

  • Instruction-tuned: Generates Python functions from plain English descriptions.
  • Python-focused: Produces clean, idiomatic Python code.
  • Compact & Efficient: At 1.5B parameters, it's designed for interactive use and runs efficiently on consumer-grade hardware.
  • Strong HumanEval Performance: Achieves 82.0% on HumanEval@50, outperforming its base Qwen 1.5B model by 7 points.
  • Apache 2.0 Licensed: Offers flexibility for commercial and research use.

Limitations

  • Python-only: Not trained for other programming languages.
  • No FIM Support: Does not support fill-in-the-middle tasks.
  • Not an Agent: Lacks tool calling or multi-step planning capabilities.
  • Weak on Algorithmic Tasks: Shows lower performance on LiveCodeBench compared to the base Qwen model.
  • Limited Context: Optimized for 4K inference context, with longer files potentially truncated.

Ideal Use Cases

  • Developers needing a lightweight, fast Python code generation assistant.
  • Applications requiring instruction-following to create Python functions.
  • Environments with limited computational resources where a compact, specialized model is beneficial.

Finetunes
1 models