Update README.md
Browse files
README.md
CHANGED
|
@@ -15,3 +15,6 @@ This model card shall provide the information for OpenGPT 5.5, the main model in
|
|
| 15 |
- This involved vector mathematics for a custom tensors and is not, in any way, propietary GPT weights.
|
| 16 |
# The making
|
| 17 |
Natarajan Intelligence Technologies, formerly GoodGoals, took a huge dataset from Hugging Face and asked the same queries in the dataset to GPT 5.2 and made a dataset from GPT 5.2 answers. From here, we used complex mathematics with vectors that improved the model's response through comparing the answers in the GPT dataset to the ones in the large HF dataset. The vectors were made a safetensor file and uploaded to Hugging Face!
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
- This involved vector mathematics for a custom tensors and is not, in any way, propietary GPT weights.
|
| 16 |
# The making
|
| 17 |
Natarajan Intelligence Technologies, formerly GoodGoals, took a huge dataset from Hugging Face and asked the same queries in the dataset to GPT 5.2 and made a dataset from GPT 5.2 answers. From here, we used complex mathematics with vectors that improved the model's response through comparing the answers in the GPT dataset to the ones in the large HF dataset. The vectors were made a safetensor file and uploaded to Hugging Face!
|
| 18 |
+
We didn't use any actual GPT weights and only used vector safetensors that improve the model's performance.
|
| 19 |
+
To verify this, you may view the **.safetensors** file which contains the training examples we used in metadata format.
|
| 20 |
+
You can also make models like this with correction vectors by using our **EfficientTrainer**
|