|
--- |
|
base_model: |
|
- crestf411/nemo-sunfall-v0.6.1 |
|
- unsloth/Mistral-Nemo-Instruct-2407 |
|
- RozGrov/NemoDori-v0.1-12B-MS |
|
tags: |
|
- merge |
|
- mergekit |
|
- lazymergekit |
|
- crestf411/nemo-sunfall-v0.6.1 |
|
- unsloth/Mistral-Nemo-Instruct-2407 |
|
- RozGrov/NemoDori-v0.1-12B-MS |
|
library_name: transformers |
|
pipeline_tag: text-generation |
|
--- |
|
|
|
# NemoDori-v0.2-12B-MN-BT |
|
|
|
NemoDori-v0.2-12B-MN-BT is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing): |
|
* [crestf411/nemo-sunfall-v0.6.1](https://huggingface.co/crestf411/nemo-sunfall-v0.6.1) |
|
* [unsloth/Mistral-Nemo-Instruct-2407](https://huggingface.co/unsloth/Mistral-Nemo-Instruct-2407) |
|
* [RozGrov/NemoDori-v0.1-12B-MS](https://huggingface.co/RozGrov/NemoDori-v0.1-12B-MS) |
|
|
|
Experimental 'sequel' of NemoDori ([v0.1](https://huggingface.co/RozGrov/NemoDori-v0.1-12B-MS)), an ERP-focused model, just for testing purpose. I still don't know what I've done... |
|
|
|
My short experience using this: |
|
- When you instruct it to roleplay, it generates short chat-like response and to-the-point. (sometimes really short) |
|
- It advances the story slowly (even slower than v0.1 i think), responding to the last roleplay message quite nicely. |
|
- Creativity is *maybe* good(?). |
|
- Can follow instructions quite well (even on depth-0). |
|
|
|
<u>Update 1</u>: |
|
<br> |
|
I was playing around with [**Talemate**](https://github.com/vegu-ai/talemate) and... this model is pretty good. It was able to follow Talemate's instructions well, which Talemate then parsed it into it's format. |
|
<br> |
|
It can become all of Talemate agents (mostly maybe). So far i've tested, it is best when having a Conversation, capable of generating character's attributes (Creator) and the world state. When it's used for narrating (Narrator), sometimes it speaks for you and includes the Talemate's conversation (for this one, maybe I just didn't have a good instruction for it). |
|
<br><br> |
|
Those are my experiences using default presets from Talemate. I did tweaked them and played a bit near the end, and it does affect the results in a good way. |
|
<br> |
|
I'll try to test it again some more later. |
|
|
|
## 🧩 Configuration |
|
|
|
```yaml |
|
models: |
|
- model: crestf411/nemo-sunfall-v0.6.1 |
|
parameters: |
|
weight: 0.5 |
|
- model: unsloth/Mistral-Nemo-Instruct-2407 |
|
parameters: |
|
weight: 0.3 |
|
- model: RozGrov/NemoDori-v0.1-12B-MS |
|
parameters: |
|
weight: 1.0 |
|
merge_method: breadcrumbs_ties |
|
base_model: RozGrov/NemoDori-v0.1-12B-MS |
|
parameters: |
|
density: 0.95 |
|
gamma: 0.01 |
|
dtype: float16 |
|
``` |
|
|
|
## 💻 Usage |
|
|
|
```python |
|
!pip install -qU transformers accelerate |
|
|
|
from transformers import AutoTokenizer |
|
import transformers |
|
import torch |
|
|
|
model = "RozGrov/NemoDori-v0.2-12B-MN-BT" |
|
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"]) |
|
``` |