---
license: apache-2.0
base_model: VextLabsinc/juwel-beryl
library_name: peft
pipeline_tag: text-generation
tags:
  - juwel
  - gem
  - gem-ruby
  - code
  - lora
  - vext
---

> **Both halves of this model are on the Hub.** The adapter weights live in this repo and the required base is [`VextLabsinc/juwel-beryl`](https://huggingface.co/VextLabsinc/juwel-beryl) (80-layer, BF16). Load the base with `AutoModelForImageTextToText`, then apply this adapter.

# GEM Ruby · code specialist

<!--
Template for every public open-weight release.
Fill on the GPU pod before huggingface-cli upload.
Never claim first/all weights / AGI / never-forgets / cheapest-ever.
-->

**Model id:** `VextLabsinc/gem-ruby`  
**Org:** Vext Labs, Inc.  
**License:** Apache-2.0 (see `LICENSE`) — **AS IS, no warranty**  
**Library:** `peft`  <!-- peft | transformers -->  
**Pipeline:** text-generation  

## Base model

- **Required base:** `VextLabsinc/juwel-beryl`  
- **Architecture notes:** Base = JUWEL Beryl (80-layer, hidden 5120, 64 heads, intermediate 25600; internal lineage theron-base-v9 + one CIP rung). This PEFT LoRA is **native to that 80-layer geometry** (`layers_to_transform` 0-79, r=64, alpha=128) — it is not padded and not drop-in on any other JUWEL base. Base architecture is `Qwen3VLForConditionalGeneration` (image-text-to-text), so load it with `AutoModelForImageTextToText`, not `AutoModelForCausalLM`.  
- **Base license / attribution:** see `NOTICE`  

If this is a **LoRA / PEFT adapter**, load base first, then:

```python
from transformers import AutoModelForImageTextToText
from peft import PeftModel
base = AutoModelForImageTextToText.from_pretrained("VextLabsinc/juwel-beryl", torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(base, "VextLabsinc/gem-ruby")
```

## Intended use

- Research, education, and **authorized** professional workflows in the **code** domain.  
- Integration into systems where a human remains responsible for outcomes.  

## Out of scope / prohibited uses

You may **not** use this model for:

- Unauthorized access to computer systems, networks, or accounts  
- Development or deployment of malware, ransomware, or fraud  
- Child sexual abuse material or any illegal content  
- Weapons development or violent crime  
- Any use that violates applicable law or third-party rights  
- For security-related models: testing only on systems you **own** or have **explicit written permission** to assess  

Vext Labs does **not** endorse misuse. Publishing weights is **not** permission to break the law.

## Limitations

- Outputs can be wrong, biased, or unsafe if misused.  
- Not a substitute for licensed professionals (medical, legal, financial, security).  
- Not guaranteed to refuse harmful requests; apply your own filters and policies.  
- Domain specialist adapter; quality varies by prompt.

## Training data (summary)

- Domain specialist continued training / LoRA on Theron lineage; see lab training docs. No customer confidential data intended.  
- **No** customer confidential data is intentionally included in this release package.  
- Downstream fine-tunes by third parties are **not** controlled by Vext Labs.

## Evaluation

- **Status:** PENDING — no reproducible benchmark published yet. Evals run on our OWN BF16 weights on a GPU pod (never a hosted API), then posted here with a full audit trail (raw responses, test cases, timestamps, model version). We publish no score we cannot reproduce on our own stack.  <!-- e.g. internal rubric 2026-04-02 | pending | public harness -->  
- **Numbers:** None yet — honest placeholder, not a hidden result. Benchmarks were blocked until 2026-07-28 by a base-identity error: these cards named a 144-layer base for an 80-layer adapter, so the adapter was not loadable as trained. The base identity is now corrected and evaluation is under way.  
- Do **not** treat internal rubrics as standardized public leaderboards (MMLU/HELM/etc.) unless re-run under a named public harness.

## Files / integrity

- Weight files: see repository file list  
- Checksums: `SHA256SUMS`  
- R2 source (internal): `s3://vext-theron-fleet/v10_loras_v9_padded/code/` (not a public download URL)

## Liability

These weights are provided **AS IS** under the `LICENSE`.  
To the maximum extent permitted by law, Vext Labs, Inc. disclaims all warranties and is **not liable** for damages arising from use or misuse of this model.  
You are solely responsible for compliance with law and for authorized use only.  
See also `USE_POLICY.md`.

## Contact

- Product / lab: https://vextlabs.ai · https://juwel.ai  
- Disclosure / questions: info@vextlabs.ai  


<!-- legal_tier: T1 -->

## Weights — download

- **Adapter (this model):** hosted in this HF repo — `adapter_model.safetensors` + `adapter_config.json`. Loads with `PeftModel.from_pretrained(base, "VextLabsinc/gem-ruby")`. (R2 mirror: https://pub-a6ae0476e46849f98f1746a61dc4c106.r2.dev/gem-ruby/)
- **Base (`juwel-beryl`, 80L, BF16):** on the Hub at https://huggingface.co/VextLabsinc/juwel-beryl — loads directly by repo id, no manual download needed.