---
license: lgpl-3.0
base_model: Qwen/Qwen2.5-Coder-1.5B
pipeline_tag: text-generation
tags:
  - gguf
  - pascal
  - code-completion
---

# PascalABC.NET Full Line Completion (Qwen/Qwen2.5-Coder-1.5B, Q4_K_M)

FIM (fill-in-the-middle) model fine-tuned for full-line code completion in
PascalABC.NET. Runs locally via llama.cpp (`llama-server`) and powers the AI
full-line completion (ghost text) feature of the PascalABC.NET VS Code
extension.

- **Base model:** Qwen/Qwen2.5-Coder-1.5B
- **Quantization:** Q4_K_M — `completion_qwen_pascal_q4_k_m.gguf` (~940 MiB)
- **Intended use:** short inline suggestions while typing PascalABC.NET code
- **Language:** Pascal (PascalABC.NET dialect)
- **Limitations:** a 1.5B model — expect short, conservative completions;
  Pascal sources only; everything runs offline on the user's machine

## Usage

Installed automatically by the PascalABC.NET VS Code extension
(setting `pascalabc.fullLineCompletion.enabled`), or manually:

```
dotnet FullLineCompletionSetup.dll --model-url https://huggingface.co/PascalABCNET/Qwen-Coder-1.5B_Q4_K_M_full_line_completion/resolve/<revision>/completion_qwen_pascal_q4_k_m.gguf
```

## Versioning

The immutable model version is the git revision (commit SHA) of this
repository. `fullline-model.json` in the repository root records the current
version (file name, revision, sha256, size) and is the integration point for
tooling that needs to check for updates.

## License

LGPL-3.0, the same as PascalABC.NET tooling.