---
license: mit
language:
- en
- code
library_name: gguf
pipeline_tag: text-generation
tags:
- svelte
- sveltekit
- svelte-5
- runes
- code-generation
- gguf
- qwen3
- lora
base_model: Qwen/Qwen3-8B
base_model_relation: finetune
---

# Svelte Coder 8B (v0.9.0)

A Svelte 5 / SvelteKit 2 specialist coding model — **8B variant**.
Free to use under MIT. Built by [rockypod](https://rockypod.com) on a
homelab RTX 3090 Ti using continuous retrieval-augmented fine-tuning
(RAFT) and a correction-stream methodology.

This is the **8B variant** for hardware where the 14B doesn't fit.
For best benchmark results, use the [14B variant](https://huggingface.co/rockypod/svelte-coder)
when the hardware allows.

**[14B (recommended)](https://huggingface.co/rockypod/svelte-coder)** ·
**[4B (lightweight)](https://huggingface.co/rockypod/svelte-coder-4b)** ·
**[GitHub — exam, integration guides, transparency](https://github.com/rockypod/svelte-coder)**

## Benchmark

| Instrument | Score |
|---|---|
| 30Q spot exam | **82.8%** (36.0 / 43.5 weighted) |
| 204Q in-scope (rescored) | 74.68% (145 / 190 raw) |

For comparison, the 14B variant scores 100% / 70.11% on the same
instruments. The 30Q is the cleaner grader; the 204Q has known
keyword-matching artifacts. See the
[main README](https://huggingface.co/rockypod/svelte-coder/blob/main/README.md)
for the full two-exams discussion.

## Hardware requirements

- **VRAM:** ~5 GB (Q4_K_M GGUF), runs on most consumer GPUs
  (RTX 3060 12GB, RTX 4060 8GB+ with offloading, Apple Silicon 8GB+)
- **Context length:** 8192
- **Recommended use case:** systems where the 14B variant (~8.4 GB)
  doesn't fit in available VRAM

## Files

- `svelte-coder-v0.9.0-8b-q4_k_m.gguf` — 4-bit quantized weights (~5 GB)

## Usage

### Ollama

```bash
ollama pull rockypod/svelte-coder:8b
ollama run rockypod/svelte-coder:8b "Write a Svelte 5 counter with $state and $derived"
```

### LM Studio / llama.cpp

Download `svelte-coder-v0.9.0-8b-q4_k_m.gguf` and load with the
production parameters: temperature 0.2, num_ctx 8192, num_predict 1500,
repeat_penalty 1.5. Use the ChatML template:

```
<|im_start|>system
You are SvelteCoder, an expert Svelte 5 / SvelteKit 2 coding assistant. Answer the question with complete, production-quality code.<|im_end|>
<|im_start|>user
Your question<|im_end|>
<|im_start|>assistant
<think>
```

## Limitations specific to the 8B

- **Svelte 4 echo trap is more frequent than on the 14B.** The 8B has
  less capacity to override Qwen3-8B's pretrained Svelte 4 reflexes,
  particularly on T1 (Runes) and T13 (DaisyUI) fix-this-snippet
  questions. Review output for `export let`, `on:click`, `<slot>`
  patterns when modernizing Svelte 4 code.
- All other limitations from the [main README](https://huggingface.co/rockypod/svelte-coder/blob/main/README.md)
  apply.

## Apple Silicon note

MLX builds for Apple Silicon are not included in v0.9.0 for the 8B and
4B variants. Apple Silicon users are recommended to use the 14B variant,
which includes MLX 4-bit weights.

## License & Attribution

**Fine-tuning work licensed under the MIT License** — see
[LICENSE](LICENSE) in the GitHub repo.

**Base model and teacher model are licensed under Apache 2.0** — see
[LICENSE-APACHE](LICENSE-APACHE) and [NOTICE](NOTICE):

- Base: [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) — © Alibaba Cloud
- Teacher: [Qwen3-Coder-Next 80B](https://huggingface.co/Qwen/Qwen3-Coder-Next) — © Alibaba Cloud

The 8B Svelte Coder weights are a derivative work of Qwen3-8B,
fine-tuned via LoRA adapters on the v1.5 Svelte 5 / SvelteKit 2
specialist dataset (1,508 entries).