README.md

---
license: apache-2.0
base_model: Qwen/Qwen3-1.7B
tags:
  - vibescript
  - code-compression
  - lora
  - gguf
  - qwen3
language:
  - en
pipeline_tag: text-generation
---

# VibeScript - Code to DSL Converter

**vibecoder-discern** converts natural language and code into VibeScript - a compact symbolic DSL for expressing programming concepts.

## What is VibeScript?

VibeScript compresses verbose code into symbolic notation:

| Code | VibeScript |
|------|------------|
| `function add(a, b) { return a + b; }` | `Ω> add!(a, b)` |
| `const users = await db.query(...)` | `δ.m.p.query()` |
| `app.get('/api/users', ...)` | `θ.m.route(θ.e, ζ.x)` |
| `if (error) { throw new Error(...) }` | `~system~γ#error!` |

## Model Variants

| Path | Format | Size | Use Case |
|------|--------|------|----------|
| `/lora-adapter/` | LoRA | ~13MB | Merge with your own Qwen3-1.7B |
| `/merged-model/` | HuggingFace | ~3.4GB | Ready-to-use transformers |
| `/gguf/` | GGUF Q4_K_M | ~1.1GB | llama.cpp / Ollama |

## Quick Start

### llama.cpp (GGUF)

```bash
# Download
wget https://huggingface.co/calebboud/vibescript/resolve/main/gguf/vibecoder-discern-1.7B-Q4_K_M.gguf

# Run
llama-cli -m vibecoder-discern-1.7B-Q4_K_M.gguf \
  -p "Convert this to vibescript: function multiply(x, y) { return x * y; }" \
  -n 100 --temp 0.7
```

### Transformers (Merged Model)

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("calebboud/vibescript", subfolder="merged-model")
tokenizer = AutoTokenizer.from_pretrained("calebboud/vibescript", subfolder="merged-model")

prompt = "Convert this to vibescript: console.log('Hello World')"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

### LoRA Adapter (Merge Yourself)

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-1.7B")
model = PeftModel.from_pretrained(base, "calebboud/vibescript", subfolder="lora-adapter")
merged = model.merge_and_unload()
```

## Training Details

- **Base Model:** Qwen/Qwen3-1.7B
- **Method:** LoRA (r=8, alpha=16)
- **Target Modules:** q_proj, k_proj, v_proj, o_proj
- **Dataset:** 885 code → vibescript examples
- **Task:** CAUSAL_LM

## VibeScript Symbols

| Symbol | Meaning |
|--------|---------|
| `Ω>` | Function definition |
| `Σ` | Route/scaffold |
| `δ` | Database operations |
| `θ` | HTTP/API |
| `γ` | Error handling |
| `ζ` | Structure/scaffold |
| `α` | Analysis |
| `ε` | Dependencies |

## Coming Soon

- **vibecoder-expand**: VibeScript → Code (reverse direction)

## License

Apache 2.0