README.md

---
license: apache-2.0
base_model: MainStack/marvy-1-14B
base_model_relation: quantized
pipeline_tag: text-generation
language:
  - en
tags:
  - servicenow
  - itsm
  - csdm
  - delivery
  - gguf
  - llama.cpp
  - ollama
  - quantized
  - qwen2.5
---

# marvy-1-14B-GGUF

**GGUF quants of marvy-1-14B, the first open LLM for the full ServiceNow delivery lifecycle. Run it locally and privately on Apple Silicon, LM Studio, or Ollama.**

GGUF quantizations of [`MainStack/marvy-1-14B`](https://huggingface.co/MainStack/marvy-1-14B)
for use with [llama.cpp](https://github.com/ggerganov/llama.cpp),
[Ollama](https://ollama.com), [LM Studio](https://lmstudio.ai), and compatible runtimes.

> Released under **Apache-2.0**. Built with Qwen — see `NOTICE`.

## Files

| File | Quant | Size (approx) | Use when |
|---|---|---|---|
| `marvy-1-14B-Q4_K_M.gguf` | Q4_K_M | ~9 GB | Default — best size/quality balance, laptops |
| `marvy-1-14B-Q8_0.gguf`   | Q8_0   | ~16 GB | Highest fidelity, near-FP16 quality |

## Quick start

### Ollama

```bash
ollama run hf.co/MainStack/marvy-1-14B-GGUF:Q4_K_M
```

### llama.cpp

```bash
./llama-cli -hf MainStack/marvy-1-14B-GGUF:Q4_K_M \
  -p "Write a ServiceNow user story with acceptance criteria for P1 SLA escalation." \
  --temp 0.4
```

### LM Studio

1. In the model browser, search `MainStack/marvy-1-14B-GGUF` and download a quant
   (`Q4_K_M` recommended), **or** drop the `.gguf` into
   `~/.lmstudio/models/MainStack/marvy-1-14B-GGUF/`.
2. Load it, set the system prompt below, temperature ~0.4.
3. To use from code/OpenCode, start the local server:
   ```bash
   lms server start          # OpenAI-compatible on http://localhost:1234/v1
   ```

### Use in OpenCode

Point OpenCode at the local LM Studio (or llama.cpp) server as an
OpenAI-compatible provider — see **[`USAGE.md`](./USAGE.md)** for the exact
`opencode.json` snippet.

### Recommended system prompt

```
You are a senior ServiceNow delivery consultant. You produce precise, implementation-grade
artifacts: business analyses, requirements, solution design documents, user stories with
acceptance criteria, test cases, and validation reviews. You favor out-of-the-box
capabilities, cite concrete tables/plugins/sys_ids when relevant, and write in clear
professional English.
```

📖 **Full usage** (all runtimes + OpenCode wiring): [`USAGE.md`](./USAGE.md) ·
**Validate it works:** [`VALIDATION.md`](./VALIDATION.md)

## Provenance & limitations

See the [merged model card](https://huggingface.co/MainStack/marvy-1-14B) for the
full training data, anonymization methodology, evaluation (test ppl 13.107 on a
project-disjoint split), and limitations. Quantization adds the usual minor
quality reduction versus the FP16 model.

## License & attribution

Dual-licensed: **weights Apache-2.0**, **MainStack contributions (cards, docs,
benchmark) CC-BY-4.0** — see [`LICENSING.md`](./LICENSING.md). **If you use
marvy-1-14B as a baseline, fine-tune it, distill from it, or evaluate against
it, please credit MainStack** and link to
https://huggingface.co/MainStack/marvy-1-14B. Keep the `NOTICE` file intact
(required by Apache-2.0 §4) and cite the entry on the
[merged model card](https://huggingface.co/MainStack/marvy-1-14B#citation).