|
--- |
|
library_name: transformers |
|
language: |
|
- en |
|
base_model: |
|
- RozGrov/NemoDori-v0.2-12B-MN-BT |
|
datasets: |
|
- Inv/c2-logs-cleaned-deslopped |
|
tags: |
|
- unsloth |
|
- trl |
|
- sft |
|
- merge |
|
- mergekit |
|
- lazymergekit |
|
- RozGrov/NemoDori-v0.2-12B-MN-BT |
|
--- |
|
|
|
# NemoDori-v0.2-Frankend.2-v1-16.6B |
|
|
|
_Experimental!_ |
|
|
|
A more upscaled version of [**NemoDori-v0.2-12B-MN-BT**](https://huggingface.co/RozGrov/NemoDori-v0.2-12B-MN-BT), now at **16.6B**. |
|
<br> |
|
This is also my first successful(?) fine-tuned model using **500 random rows** from dataset |
|
[Inv/c2-logs-cleaned-deslopped](https://huggingface.co/datasets/Inv/c2-logs-cleaned-deslopped) in 70 steps. |
|
The reason I used that dataset is... just for testing. What I thought is, if I can replace/fill up those duplicated layers by training it, maybe that makes it better. |
|
|
|
NemoDori v0.2 is my best merge model so far, but I'm afraid it's still 12B, not much to improve after merging all kinds of models. |
|
<br> |
|
Again, I'm just interested to play with these LLM stuff for awhile. Maybe more version of this will come out later. |
|
|
|
|
|
As far from my short testing, this model has become a little more strict than the parent model (v0.2).I haven't notice anything major yet. |
|
<br> |
|
You can use ST with this preset [here](https://huggingface.co/RozGrov/NemoDori-v0.2-Frankend.2-v1-16.6B/resolve/main/NemoDori-v0.2-Frankend.2-v1-16.6B%20-%20ST%20Preset.json). |
|
Unfortunately, you can't go wild with this model (from my short tests), sometimes it makes little senses, and sometimes... you will get a reddit link (i'm not kidding). |
|
|
|
I didn't have enough time to test it, because it's more pricey without quantization. |
|
<br> |
|
I trust @mradermacher to make the quants version of this model. (Thank you so much for making those GGUF on my models ^_^) |
|
<br> |
|
And yeah... Your feedbacks are always welcome, and let me know what's your experience using this model, that would be appreciated. |
|
|
|
Take care everyone. |
|
|
|
### Merge Method |
|
|
|
This model was merged from the following models using the `passthrough` merge method: |
|
* [RozGrov/NemoDori-v0.2-12B-MN-BT](https://huggingface.co/RozGrov/NemoDori-v0.2-12B-MN-BT) |
|
|
|
## 🧩 Configuration |
|
|
|
```yaml |
|
slices: |
|
- sources: |
|
- model: RozGrov/NemoDori-v0.2-12B-MN-BT |
|
layer_range: [0, 8] |
|
- sources: |
|
- model: RozGrov/NemoDori-v0.2-12B-MN-BT |
|
layer_range: [8, 24] |
|
parameters: |
|
scale: |
|
- filter: q_proj |
|
value: 0.919 |
|
- filter: k_proj |
|
value: 0.919 |
|
- value: 1.0 |
|
- sources: |
|
- model: RozGrov/NemoDori-v0.2-12B-MN-BT |
|
layer_range: [16, 24] |
|
parameters: |
|
scale: |
|
- filter: q_proj |
|
value: 0.7 |
|
- filter: k_proj |
|
value: 0.7 |
|
- filter: o_proj |
|
value: 0.0 |
|
- filter: down_proj |
|
value: 0.0 |
|
- value: 1.0 |
|
- sources: |
|
- model: RozGrov/NemoDori-v0.2-12B-MN-BT |
|
layer_range: [16, 32] |
|
parameters: |
|
scale: |
|
- filter: q_proj |
|
value: 0.919 |
|
- filter: k_proj |
|
value: 0.919 |
|
- value: 1.0 |
|
- sources: |
|
- model: RozGrov/NemoDori-v0.2-12B-MN-BT |
|
layer_range: [32, 40] |
|
merge_method: passthrough |
|
dtype: bfloat16 |
|
``` |
|
|
|
## 💻 Usage |
|
|
|
```python |
|
!pip install -qU transformers accelerate |
|
|
|
from transformers import AutoTokenizer |
|
import transformers |
|
import torch |
|
|
|
model = "RozGrov/NemoDori-v0.2-Frankend.2-pre" |
|
messages = [{"role": "user", "content": "What is a large language model?"}] |
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model) |
|
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) |
|
pipeline = transformers.pipeline( |
|
"text-generation", |
|
model=model, |
|
torch_dtype=torch.float16, |
|
device_map="auto", |
|
) |
|
|
|
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95) |
|
print(outputs[0]["generated_text"]) |
|
``` |