Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,26 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
# trackgpt
|
| 6 |
+
|
| 7 |
+
An LLM designed to play TrackMania using WASD.
|
| 8 |
+
|
| 9 |
+
# Dataset: leafspark/tracktext
|
| 10 |
+
|
| 11 |
+
llama.cpp train command:
|
| 12 |
+
```
|
| 13 |
+
train-text-from-scratch --layer 8 --ctx 17000 --vocab-model ../models/ggml-vocab-llama.gguf --embd 64 --head 8 --checkpoint-in chk-fsd-384x36-LATEST.gguf --checkpoint-out chk-fsd-384x36-ITERATION.gguf --model-out ggml-fsd-384x36-f32-ITERATION.gguf --train-data "wikihow.txt" -t 12 -b 8 --seed 1 --adam-iter 2560 --no-checkpointing --save-every 30 --adam-alpha 0.001
|
| 14 |
+
```
|
| 15 |
+
# Notes
|
| 16 |
+
|
| 17 |
+
Please make sure to remove "0." and " " from your input! This is so that it can fit in the model context window.
|
| 18 |
+
|
| 19 |
+
# Prompt Format
|
| 20 |
+
|
| 21 |
+
```
|
| 22 |
+
{data}
|
| 23 |
+
{{prompt}}
|
| 24 |
+
{action}
|
| 25 |
+
{{response}}
|
| 26 |
+
```
|