Update README.md
Browse files
README.md
CHANGED
|
@@ -10,7 +10,7 @@ tags:
|
|
| 10 |
license: other
|
| 11 |
---
|
| 12 |
|
| 13 |
-
# GamePlayer
|
| 14 |
|
| 15 |
This is the frozen broad multi-game SFT mixture used in the EventChunk-FPS
|
| 16 |
research lineage.
|
|
@@ -22,8 +22,7 @@ research lineage.
|
|
| 22 |
- Total: 1,425,263 rows and 11,264,394 image/action supervision turns.
|
| 23 |
- 23,509 train and 1,359 validation trajectory groups, with zero reported
|
| 24 |
train/validation trajectory overlap and zero missing image paths.
|
| 25 |
-
- Action horizons: 8, 12, 16, or 24 atomic actions.
|
| 26 |
-
intended to pause during generation and each chunk executes once.
|
| 27 |
|
| 28 |
The source data comes from [OpenP2P](https://huggingface.co/elefantai/open-p2p), [CrossFPS](https://huggingface.co/datasets/zizhaotong/CrossFPS-train), [CS2](https://huggingface.co/datasets/RekaAI/CS2-10k), [Gaming500](https://huggingface.co/datasets/markov-ai/gaming-500-hours), and a small ViZDoom
|
| 29 |
success-anchor set we created. You can get the source video from these repo, cause we just provide the frames from codec. See `sft_v16_scale100_mixture.manifest.json` and `validation.json` under `archives/metadata.tar` for exact source and game counts.
|
|
|
|
| 10 |
license: other
|
| 11 |
---
|
| 12 |
|
| 13 |
+
# GamePlayer-1.4M
|
| 14 |
|
| 15 |
This is the frozen broad multi-game SFT mixture used in the EventChunk-FPS
|
| 16 |
research lineage.
|
|
|
|
| 22 |
- Total: 1,425,263 rows and 11,264,394 image/action supervision turns.
|
| 23 |
- 23,509 train and 1,359 validation trajectory groups, with zero reported
|
| 24 |
train/validation trajectory overlap and zero missing image paths.
|
| 25 |
+
- Action horizons: 8, 12, 16, or 24 atomic actions.
|
|
|
|
| 26 |
|
| 27 |
The source data comes from [OpenP2P](https://huggingface.co/elefantai/open-p2p), [CrossFPS](https://huggingface.co/datasets/zizhaotong/CrossFPS-train), [CS2](https://huggingface.co/datasets/RekaAI/CS2-10k), [Gaming500](https://huggingface.co/datasets/markov-ai/gaming-500-hours), and a small ViZDoom
|
| 28 |
success-anchor set we created. You can get the source video from these repo, cause we just provide the frames from codec. See `sft_v16_scale100_mixture.manifest.json` and `validation.json` under `archives/metadata.tar` for exact source and game counts.
|