Update README.md
Browse files
README.md
CHANGED
|
@@ -16,6 +16,58 @@ model-index:
|
|
| 16 |
results: []
|
| 17 |
---
|
| 18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
## Llamacpp imatrix Quantizations of miniclaus-qw1.5B-UNAMGS
|
| 20 |
|
| 21 |
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b4058">b4058</a> for quantization.
|
|
|
|
| 16 |
results: []
|
| 17 |
---
|
| 18 |
|
| 19 |
+
# miniclaus-qw1.5B-UNAMGS
|
| 20 |
+
|
| 21 |
+
Trained with `Magpie-Align/Magpie-Pro-MT-300K-v0.1`
|
| 22 |
+
|
| 23 |
+
Using MGS & UNA (MLP) on this tiny but powerful model.
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
|
| 27 |
+
|
| 28 |
+
It achieves the following results on the evaluation set:
|
| 29 |
+
- Loss: 0.7193
|
| 30 |
+
|
| 31 |
+
## Training procedure
|
| 32 |
+
|
| 33 |
+
### Training hyperparameters
|
| 34 |
+
|
| 35 |
+
The following hyperparameters were used during training:
|
| 36 |
+
- train_batch_size: 1
|
| 37 |
+
- seed: 42
|
| 38 |
+
- distributed_type: multi-GPU
|
| 39 |
+
- num_devices: 8
|
| 40 |
+
- total_train_batch_size: 128
|
| 41 |
+
- total_eval_batch_size: 8
|
| 42 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 43 |
+
- num_epochs: 1
|
| 44 |
+
|
| 45 |
+
### Training results
|
| 46 |
+
|
| 47 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
| 48 |
+
|:-------------:|:------:|:----:|:---------------:|
|
| 49 |
+
| 1.1641 | 0.0007 | 1 | 0.8514 |
|
| 50 |
+
| 0.9246 | 0.0503 | 76 | 0.7921 |
|
| 51 |
+
| 0.8791 | 0.1006 | 152 | 0.7727 |
|
| 52 |
+
| 0.8507 | 0.1509 | 228 | 0.7611 |
|
| 53 |
+
| 0.8376 | 0.2012 | 304 | 0.7534 |
|
| 54 |
+
| 0.793 | 0.2515 | 380 | 0.7467 |
|
| 55 |
+
| 0.7834 | 0.3018 | 456 | 0.7421 |
|
| 56 |
+
| 0.7807 | 0.3521 | 532 | 0.7384 |
|
| 57 |
+
| 0.764 | 0.4023 | 608 | 0.7359 |
|
| 58 |
+
| 0.7738 | 0.4526 | 684 | 0.7320 |
|
| 59 |
+
| 0.7425 | 0.5029 | 760 | 0.7300 |
|
| 60 |
+
| 0.7519 | 0.5532 | 836 | 0.7279 |
|
| 61 |
+
| 0.7461 | 0.6035 | 912 | 0.7255 |
|
| 62 |
+
| 0.7489 | 0.6538 | 988 | 0.7245 |
|
| 63 |
+
| 0.7614 | 0.7041 | 1064 | 0.7222 |
|
| 64 |
+
| 0.7576 | 0.7544 | 1140 | 0.7222 |
|
| 65 |
+
| 0.7303 | 0.8047 | 1216 | 0.7209 |
|
| 66 |
+
| 0.7332 | 0.8550 | 1292 | 0.7199 |
|
| 67 |
+
| 0.7541 | 0.9053 | 1368 | 0.7202 |
|
| 68 |
+
| 0.7369 | 0.9556 | 1444 | 0.7193 |
|
| 69 |
+
|
| 70 |
+
|
| 71 |
## Llamacpp imatrix Quantizations of miniclaus-qw1.5B-UNAMGS
|
| 72 |
|
| 73 |
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b4058">b4058</a> for quantization.
|