--- title: README emoji: 🌴 colorFrom: green colorTo: yellow sdk: static app_file: home.html pinned: false short_description: Jamaica's independent AI research lab. ---

CorX Labs. Architecting the future, from the ground up.

**CorX Labs is Jamaica's independent AI research lab.** We build Jamaican AI from the ground up: five open models so far, from CorX3.8, a 27-billion-parameter Jamaican Patois assistant, to NatureV1, a weather and hurricane model. Most were trained from random weights on a single GPU, and where one is a fine-tune, we say so. | Models built | Released with weights | Technical reports | Licence | |:-:|:-:|:-:|:-:| | **5** | **4** | **5** | **Apache-2.0** |

### Models | Model | What it is | Size | Built from | Report | |---|---|:-:|---|:-:| | [**CorX3.8-27B**](https://huggingface.co/CorxLabs/CorX3.8-27B) | Jamaican Patois conversational assistant. Patois by default, English on request. | 27B | Fine-tune of Qwen3.8-27B, fully merged | [03](https://corxlabs-papers.static.hf.space/corx3.8-27b-model-report.html) | | [**CorX1.5**](https://huggingface.co/CorxLabs/CorX1.5-160M) | Decoder-only language model with learned role embeddings | 157.8M | From scratch, tokenizer included | [01](https://corxlabs-papers.static.hf.space/corx1.5-technical-report.html) | | [**TriStream-SVS**](https://huggingface.co/CorxLabs/TriStream-SVS-300M) | Singing-voice synthesis: pitch, timbre and breath in three separate streams | 321.8M | From scratch | [02](https://corxlabs-papers.static.hf.space/tristream-svs-technical-report.html) | | [**NatureV1**](https://huggingface.co/CorxLabs/NatureV1-500) | Probabilistic weather and hurricane model, attention in kilometres on the sphere | 89.45M | From scratch, on the iHELIX engine | [05](https://corxlabs-papers.static.hf.space/naturev1-hurricane-physics-report.html) | | [**HELIX**](https://pypi.org/project/helix-lm/) | Linear-scaling sequence architecture that keeps exact long-range recall | configurable | From scratch; no trained checkpoints | [04](https://corxlabs-papers.static.hf.space/helix-technical-report.html) | Browse by collection: [Text](https://huggingface.co/collections/CorxLabs/text-models-6ad400811d7073d767ebe7f1) · [Singing voice](https://huggingface.co/collections/CorxLabs/svs-models-6ad40082a6e143550bd7aaa6) · [Weather](https://huggingface.co/collections/CorxLabs/weather-models-6ad40082a150e48c970f176a) ### Papers Read them on the web or download the PDFs: **[CorX Labs Papers](https://huggingface.co/spaces/CorxLabs/papers)**

1. **[CorX1.5: a 158M-parameter language model trained from random weights on one GPU.](https://corxlabs-papers.static.hf.space/corx1.5-technical-report.html)** ([PDF](https://corxlabs-papers.static.hf.space/corx1.5-technical-report.pdf)) The architecture, the three-stage training, and the memory arithmetic that fits three billion tokens through one card. 2. **[TriStream-SVS: building the source-filter model of the voice into a singing synthesiser.](https://corxlabs-papers.static.hf.space/tristream-svs-technical-report.html)** ([PDF](https://corxlabs-papers.static.hf.space/tristream-svs-technical-report.pdf)) Three streams that cannot share information until a late fusion trunk, with zero-shot cloning and a 32-step flow-matching decoder. 3. **[CorX3.8-27B: giving an open 27B model a Jamaican voice.](https://corxlabs-papers.static.hf.space/corx3.8-27b-model-report.html)** ([PDF](https://corxlabs-papers.static.hf.space/corx3.8-27b-model-report.pdf)) What a rank-64 LoRA on 19 million tokens changed, and what it did not. 4. **[HELIX: separating memory capacity from bandwidth in a linear-scaling sequence model.](https://corxlabs-papers.static.hf.space/helix-technical-report.html)** ([PDF](https://corxlabs-papers.static.hf.space/helix-technical-report.pdf)) N^1.00 attention pairs, and recall that ties full attention while the same model without its index falls to chance. 5. **[Hurricane tracks without a neural network.](https://corxlabs-papers.static.hf.space/naturev1-hurricane-physics-report.html)** ([PDF](https://corxlabs-papers.static.hf.space/naturev1-hurricane-physics-report.pdf)) A physics consensus that beats extrapolation and CLIPER at every lead on 4,447 forecasts from four seasons it never saw. ### Measured, not promised | Hurricane track error (km), 2018 and 2020-2022, all basins | 24 h | 48 h | 72 h | 120 h | |---|:-:|:-:|:-:|:-:| | Straight-line extrapolation (strongest per lead) | 179 | 442 | 751 | 1,417 | | CLIPER | 163 | 378 | 589 | 914 | | **NatureV1 hurricane physics, no neural network** | **127** | **301** | **515** | **884** | Global AI models such as GraphCast and Pangu-Weather remain far better beyond 12 hours, and the report says so. The same rule runs through every model here: results are measured on data the model never trained on, "trained from scratch", "fine-tuned" and "not trained yet" are kept apart, and nothing is claimed that has not been measured. TriStream-SVS, for example, has no listening tests yet, so it reports no quality scores.

[corx-labs.com](https://www.corx-labs.com/) · [Chat with CorX3.8](https://www.corx-labs.com/chat/) · [Documentation](https://www.corx-labs.com/documentation/) · [Blog](https://www.corx-labs.com/blog/) · [Benchmarks](https://www.corx-labs.com/benchmarks/) · PyPI: [naturev1](https://pypi.org/project/naturev1/), [ihelix](https://pypi.org/project/ihelix/), [helix-lm](https://pypi.org/project/helix-lm/) _{CorX Labs · Westmoreland, Jamaica · Architecting the future, from the ground up.}