Update README.md
Browse files
README.md
CHANGED
|
@@ -10,4 +10,29 @@ tags:
|
|
| 10 |
- agent
|
| 11 |
- lora
|
| 12 |
- finetune
|
| 13 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
- agent
|
| 11 |
- lora
|
| 12 |
- finetune
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
Testing a QLoRA adaptor for [allenai/MolmoE-1B-0924](https://huggingface.co/allenai/MolmoE-1B-0924),
|
| 17 |
+
|
| 18 |
+
Targets top 10 experts that are activated when pointing is involved and image pooling and projection layers of Vision backbone
|
| 19 |
+
|
| 20 |
+
Trained on 47 screenshots of a low-poly video game with ragdoll casualties
|
| 21 |
+
|
| 22 |
+
Evaluated on 44 screenshots of aforementioned video game
|
| 23 |
+
|
| 24 |
+
Molmo has an edge case where it declares there are no humans in an image:
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
This custom QLoRA successfully reduces the occurance of these cases
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
However, pointing to non-human objects is observed to increase.
|
| 31 |
+
|
| 32 |
+
Comparison of Model performance with and without QLora on Eval dataset
|
| 33 |
+
|Model| MolmoE-1B | MolmoE-1B w/ QLora |
|
| 34 |
+
|----------|------|------|
|
| 35 |
+
| Precision | 82.4 | 81.5 |
|
| 36 |
+
| Recall | 63.5 | 72.1 |
|
| 37 |
+
|
| 38 |
+
Dataset: [reubk/RavenfieldDataset](https://huggingface.co/datasets/reubk/RavenfieldDataset)
|