Ramikan-BR
commited on
Commit
•
cd095f1
1
Parent(s):
27d7c2d
Update README.md
Browse files
README.md
CHANGED
@@ -22,4 +22,43 @@ base_model: unsloth/tinyllama-bnb-4bit
|
|
22 |
|
23 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
24 |
|
25 |
-
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
24 |
|
25 |
+
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
26 |
+
|
27 |
+
==((====))== Unsloth - 2x faster free finetuning | Num GPUs = 1
|
28 |
+
\\ /| Num examples = 967 | Num Epochs = 1
|
29 |
+
O^O/ \_/ \ Batch size per device = 2 | Gradient Accumulation steps = 16
|
30 |
+
\ / Total batch size = 32 | Total steps = 30
|
31 |
+
"-____-" Number of trainable parameters = 100,925,440
|
32 |
+
[30/30 26:26, Epoch 0/1]
|
33 |
+
Step Training Loss
|
34 |
+
1 1.737000
|
35 |
+
2 1.738000
|
36 |
+
3 1.384700
|
37 |
+
4 1.086400
|
38 |
+
5 1.009600
|
39 |
+
6 0.921000
|
40 |
+
7 0.830400
|
41 |
+
8 0.808900
|
42 |
+
9 0.774500
|
43 |
+
10 0.759900
|
44 |
+
11 0.736100
|
45 |
+
12 0.721200
|
46 |
+
13 0.733200
|
47 |
+
14 0.701000
|
48 |
+
15 0.711700
|
49 |
+
16 0.701400
|
50 |
+
17 0.689500
|
51 |
+
18 0.678800
|
52 |
+
19 0.675200
|
53 |
+
20 0.680500
|
54 |
+
21 0.685800
|
55 |
+
22 0.681200
|
56 |
+
23 0.672000
|
57 |
+
24 0.679900
|
58 |
+
25 0.675500
|
59 |
+
26 0.666600
|
60 |
+
27 0.687900
|
61 |
+
28 0.653600
|
62 |
+
29 0.672500
|
63 |
+
30 0.660900
|
64 |
+
|