Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,13 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
|
| 5 |
# Welcome to a new generation of Minecraft
|
|
@@ -21,4 +29,29 @@ Andy 3.5 is designed to be used with MindCraft
|
|
| 21 |
|
| 22 |
## How was model trained?
|
| 23 |
|
| 24 |
-
The
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- Sweaterdog/Andy-3.5
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
base_model:
|
| 8 |
+
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
|
| 9 |
+
tags:
|
| 10 |
+
- Minecraft
|
| 11 |
---
|
| 12 |
|
| 13 |
# Welcome to a new generation of Minecraft
|
|
|
|
| 29 |
|
| 30 |
## How was model trained?
|
| 31 |
|
| 32 |
+
The model was trained on a dataset of ~10,000 messages coming directly from MindCraft, ensuring quality data
|
| 33 |
+
|
| 34 |
+
## What are capabilities and Limitations?
|
| 35 |
+
|
| 36 |
+
The smaller model *(The preview ones at least)* had 1/3 of the parameters tuned, the larger preview model has not been released yet.
|
| 37 |
+
Andy-3.5 was trained on EVERYTHING regarding Minecraft and MindCraft, it knows how to use commands natively without a system prompt.
|
| 38 |
+
Andy-3.5 also knows how to build / use !newAction to perform commands, it was trained on lots of building, as well as, using !newAction to do tasks like manually making something or strip mining.
|
| 39 |
+
|
| 40 |
+
****Know this is a PREVIEW model, it is NOT finished!****
|
| 41 |
+
|
| 42 |
+
## Why a preview model?
|
| 43 |
+
|
| 44 |
+
Andy-3.5-preview was made to test the intelligence of a Minecraft Ai with the current dataset, it was meant to see the progress of the training and what area's are needed for the future
|
| 45 |
+
DO NOT expect this model to be able to do everything perfectly, it only knows as much as the dataset told it, as well as the other 2/3 of the untouched parameters allow.
|
| 46 |
+
The model *may* experience bugs, such as not saying your name, getting previous messages confused, or other small things.
|
| 47 |
+
|
| 48 |
+
## Important notes and considerations
|
| 49 |
+
|
| 50 |
+
The preview model of Andy-3.5-mini *(Andy-3.5-mini-preview)* was trained on a context length of 4096, this was meant to speed up training and VRAM usage.
|
| 51 |
+
|
| 52 |
+
The Base model of Andy-3.5-mini-preview was a distilled version of Deepseek-R1, which was a tuned model of Qwen-2.5-1.5b
|
| 53 |
+
|
| 54 |
+
Since a context window of 4096 is not nearly enough for MindCraft, you can go higher, Qwen-2.5-1.5b was trained on a context length of 64,000, the distilled version of Deepseek-R1 was trained on a length of 128,000, the usable length may be closer to 32,000 tokens for Andy-3.5-mini-preview
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
When the full versions of Andy-3.5 and Andy-3.5-preview release, they will both be trained on a context length of 128,000 to ensure proper usage during playing.
|