---
base_model: Qwen/Qwen3-Coder-30B-A3B-Instruct
library_name: llama.cpp
pipeline_tag: text-generation
license: apache-2.0
tags:
- gguf
- qwen3-coder
- coding
- software-engineering
- moe
- q8_0
- q4_k_m
- tiny-pickle
---

# Tiny Pickle v3 Coder — GGUF

Quantized GGUF releases of Tiny Pickle v3 Coder.

Tiny Pickle v3 Coder was produced by fine-tuning
`Qwen/Qwen3-Coder-30B-A3B-Instruct` with the LoRA adapter published at
`vsan/tiny-pickle-v3-coder-LoRA`, then merging and converting the resulting model with
llama.cpp.

## Files

| File | Quantization | Approximate size |
|---|---|---:|
| `tiny-pickle-v3-coder-q8_0.gguf` | Q8_0 | 31G |
| `tiny-pickle-v3-coder-q4_k_m.gguf` | Q4_K_M | 18G |

Q8_0 retains greater numerical fidelity but requires more storage and
memory. Q4_K_M is smaller and more practical for local inference.

## Run with llama.cpp

```bash
llama-cli \
  -m tiny-pickle-v3-coder-q4_k_m.gguf \
  -ngl 99 \
  -c 8192 \
  -p "Write a robust Python LRU cache with unit tests."
```

## Intended use

- Code generation
- Debugging
- Code review
- Implementation planning
- Test generation
- Software-engineering assistance

## Limitations

Tiny Pickle v3 Coder is experimental and has not yet been proven superior
to its base model on independent benchmarks. Quantization may reduce model
quality. Generated code can be incorrect, insecure, incomplete, or
non-functional and must be reviewed and tested.