Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# WIGIP-1 v2
|
| 2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
## Stage 1 – Text Pre-Training (ViT-Style Transformer)
|
| 4 |
|
| 5 |
WIGIP-1 v2 is an experimental research model exploring **Vision Transformer (ViT) style architectures for text modeling**, implemented using **JAX + Flax** with **Fully Sharded Data Parallelism (FSDP)** via `pjit`.
|
|
@@ -78,5 +101,4 @@ This phase is intended to test whether **ViT-style inductive biases** can learn
|
|
| 78 |
## ⚠️ Disclaimer
|
| 79 |
|
| 80 |
This is **research code** and an **experimental architecture**.
|
| 81 |
-
Results are preliminary and **not production-ready**.
|
| 82 |
-
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
datasets:
|
| 4 |
+
- allenai/c4
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
pipeline_tag: text-generation
|
| 8 |
+
tags:
|
| 9 |
+
- jax
|
| 10 |
+
- fsdp
|
| 11 |
+
- vision-transformer
|
| 12 |
+
- text-generation
|
| 13 |
+
- experimental
|
| 14 |
+
---
|
| 15 |
+
# WIGIP-1 v2
|
| 16 |
+
|
| 17 |
+
|
| 18 |
# WIGIP-1 v2
|
| 19 |
|
| 20 |
+
> **⚠️ implementation & Training Scripts:**
|
| 21 |
+
> The full source code, JAX training loops, and architecture definitions are available on my GitHub:
|
| 22 |
+
> 🔗 **[Click here to view the Training Scripts on GitHub](https://github.com/nurric-ai/Wigip_v2_1.7B_ViT)**
|
| 23 |
+
|
| 24 |
+
---
|
| 25 |
+
|
| 26 |
## Stage 1 – Text Pre-Training (ViT-Style Transformer)
|
| 27 |
|
| 28 |
WIGIP-1 v2 is an experimental research model exploring **Vision Transformer (ViT) style architectures for text modeling**, implemented using **JAX + Flax** with **Fully Sharded Data Parallelism (FSDP)** via `pjit`.
|
|
|
|
| 101 |
## ⚠️ Disclaimer
|
| 102 |
|
| 103 |
This is **research code** and an **experimental architecture**.
|
| 104 |
+
Results are preliminary and **not production-ready**.
|
|
|