Commit
•
e1689d1
1
Parent(s):
7cf73f8
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
---
|
| 3 |
+
|
| 4 |
+
base_model: HuggingFaceTB/SmolLM-135M
|
| 5 |
+
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
# QuantFactory/Biggie-SmoLlm-0.15B-Base-GGUF
|
| 11 |
+
This is quantized version of [nisten/Biggie-SmoLlm-0.15B-Base](https://huggingface.co/nisten/Biggie-SmoLlm-0.15B-Base) created using llama.cpp
|
| 12 |
+
|
| 13 |
+
# Original Model Card
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
###EVEN SMALLER Frankenstein of smolLm-0.13b upped to 0.15b
|
| 17 |
+
Use this frankenbase for training.
|
| 18 |
+
|
| 19 |
+
Done via semi-automated continuous merging to figure out the recipe.
|
| 20 |
+
Model is more coherent.
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
```bash
|
| 25 |
+
wget https://huggingface.co/nisten/Biggie-SmoLlm-0.15B-Base/resolve/main/Biggie_SmolLM_0.15B_Base_bf16.gguf
|
| 26 |
+
```
|
| 27 |
+
```verilog
|
| 28 |
+
llama-cli -ngl 99 -co --temp 0 -p "How to build a city on Mars via calculating Aldrin-Cycler orbits?" -m Biggie_SmolLM_0.15B
|
| 29 |
+
_Base_bf16.gguf
|
| 30 |
+
```
|
| 31 |
+
The temperature settings and min p etc need to be adjusted but even at default temp0 it was coherent for first 100 tokens.
|
| 32 |
+
Amazing option for further training. And this is a merge of the base, not the instruct!
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
I don't understand how the f a 150mb file can talk but it can
|
| 37 |
+
|