Update README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,14 @@
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
---
|
|
|
|
|
|
|
| 4 |
https://huggingface.co/chargoddard/llama2-22b-blocktriangular trained one one epoch of 52k rows of Stanford Alpaca. About 11 hours on a 3090.
|
| 5 |
|
| 6 |
I had trouble with training using the other 22b method with `BLOCK_DIAGONAL=True`, but with this method, this is the first time I've been able to target all modules without breaking the output.
|
| 7 |
|
| 8 |
`target_modules = ["q_proj", "k_proj", "v_proj", "o_proj", "up_proj", "gate_proj", "down_proj"]`
|
| 9 |
|
| 10 |
-
Trained at 5e-5 with r=32. For more info see https://wandb.ai/nkpz/huggingface/runs/3oy5nbtv/workspace?workspace=user-nkpz
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
---
|
| 4 |
+
**There is no official 22b model, this is just a weird experiment, and any potential benefits of hacking on the architecture have not been validated in any formal manner**
|
| 5 |
+
|
| 6 |
https://huggingface.co/chargoddard/llama2-22b-blocktriangular trained one one epoch of 52k rows of Stanford Alpaca. About 11 hours on a 3090.
|
| 7 |
|
| 8 |
I had trouble with training using the other 22b method with `BLOCK_DIAGONAL=True`, but with this method, this is the first time I've been able to target all modules without breaking the output.
|
| 9 |
|
| 10 |
`target_modules = ["q_proj", "k_proj", "v_proj", "o_proj", "up_proj", "gate_proj", "down_proj"]`
|
| 11 |
|
| 12 |
+
Trained at 5e-5 with r=32. For more info see https://wandb.ai/nkpz/huggingface/runs/3oy5nbtv/workspace?workspace=user-nkpz
|
| 13 |
+
|
| 14 |
+
It's been responding coherently enough that I would need to run some objective benchmarks to determine if this is better/worse than stock llama 13b
|