Update README.md
Browse files
README.md
CHANGED
|
@@ -38,7 +38,8 @@ Our pretrained model series, trained from the ground up rather than fine-tuned.
|
|
| 38 |
|---|---|---|---|---|
|
| 39 |
| **Hydrion v1** | 114M | Base | Released | [Hugging Face β](https://huggingface.co/OPENGCM/Hydrion-v1-Base) |
|
| 40 |
| **Hydrion v1-SFT** | 114M | Chat-ready | Released | [Hugging Face β](https://huggingface.co/OPENGCM/Hydrion-v1-SFT) |
|
| 41 |
-
| **Hydrion
|
|
|
|
| 42 |
|
| 43 |
|
| 44 |
---
|
|
@@ -76,7 +77,7 @@ Our current flagship GTM model is GTM-v2-chat.
|
|
| 76 |
Small models don't have to mean small ambitions. We're building on a few principles:
|
| 77 |
|
| 78 |
- **Open** β every weight, dataset, and training method we use is public, not just the final model
|
| 79 |
-
- **Efficient** β
|
| 80 |
- **Customizable** β models built to be forked, fine-tuned, and taken apart
|
| 81 |
- **Accessible** β no gatekeeping behind closed weights or paywalled APIs
|
| 82 |
|
|
@@ -89,9 +90,6 @@ We're currently focused on:
|
|
| 89 |
- **Pretraining from scratch** β Hydrion, our from-scratch language model series, and the data/architecture decisions that go into it
|
| 90 |
- **Fine-tuning** β GCM, our code-focused chat models built on existing open-weight bases
|
| 91 |
- **Evaluation** β benchmarking against public leaderboards and tracking what actually moves the needle
|
| 92 |
-
- **Local deployment** β making sure what we build can run on real, non-datacenter hardware
|
| 93 |
-
|
| 94 |
-
Follow along as it happens in our [Research space](https://huggingface.co/spaces/OPENGCM/Research).
|
| 95 |
|
| 96 |
---
|
| 97 |
|
|
|
|
| 38 |
|---|---|---|---|---|
|
| 39 |
| **Hydrion v1** | 114M | Base | Released | [Hugging Face β](https://huggingface.co/OPENGCM/Hydrion-v1-Base) |
|
| 40 |
| **Hydrion v1-SFT** | 114M | Chat-ready | Released | [Hugging Face β](https://huggingface.co/OPENGCM/Hydrion-v1-SFT) |
|
| 41 |
+
| **Hydrion v2** | ~250M | Base | In training |
|
| 42 |
+
| **???????** | ~250M | Base | In training |
|
| 43 |
|
| 44 |
|
| 45 |
---
|
|
|
|
| 77 |
Small models don't have to mean small ambitions. We're building on a few principles:
|
| 78 |
|
| 79 |
- **Open** β every weight, dataset, and training method we use is public, not just the final model
|
| 80 |
+
- **Efficient** β our models are trained on powerful GPUs, but can be ran on consumer GPUs
|
| 81 |
- **Customizable** β models built to be forked, fine-tuned, and taken apart
|
| 82 |
- **Accessible** β no gatekeeping behind closed weights or paywalled APIs
|
| 83 |
|
|
|
|
| 90 |
- **Pretraining from scratch** β Hydrion, our from-scratch language model series, and the data/architecture decisions that go into it
|
| 91 |
- **Fine-tuning** β GCM, our code-focused chat models built on existing open-weight bases
|
| 92 |
- **Evaluation** β benchmarking against public leaderboards and tracking what actually moves the needle
|
|
|
|
|
|
|
|
|
|
| 93 |
|
| 94 |
---
|
| 95 |
|