L3.1-70b-Inori

KaraKaraWitch/L3.1-70b-Inori

Inori is the second 70b for the weekend for me to play around.

Learning from the previous model, I yeeted Hermes into the atmosphere and used Glitz as a base.

Inori takes a different approach by using Model Stock.

• Dracarys (I just threw it in, but can be useful for code)
• Euryale (You all know it!)
• Cathallama (Athene + turboderp_cat)
• New Dawn (I heard people like it so I added it in)
• Celeste (RP)
• Japanese-Instruct (Enhancement of Japanese Language for the weebs out there.)

No Hermes was harmed in the making of this model stock merge.

L3.1-70b-Inori is a merge of the following models using LazyMergekit:

• abacusai/Dracarys-Llama-3.1-70B-Instruct
• Sao10K/L3-70B-Euryale-v2.1
• gbueno86/Cathallama-70B
• sophosympatheia/New-Dawn-Llama-3.1-70B-v1.1
• nothingiisreal/L3.1-70B-Celeste-V0.1-BF16
• cyberagent/Llama-3.1-70B-Japanese-Instruct-2407

General Thoughts

• This model has some weird censorship issues. Soketimes it triggers that it can't generate explicit text while sometimes it doesn't.
• For that reason, I don't recommend people to use this model.

Yap / Chat Format

L3 Instruct.

Quants & Hosts

🧩 Configuration

models:
  - model: Fizzarolli/L3.1-70b-glitz-v0.2
  - model: cyberagent/Llama-3.1-70B-Japanese-Instruct-2407
  - model: Sao10K/L3-70B-Euryale-v2.1
  - model: nothingiisreal/L3.1-70B-Celeste-V0.1-BF16
  - model: sophosympatheia/New-Dawn-Llama-3.1-70B-v1.1
  - model: gbueno86/Cathallama-70B
  - model: abacusai/Dracarys-Llama-3.1-70B-Instruct

merge_method: model_stock
base_model: Fizzarolli/L3.1-70b-glitz-v0.2
parameters:
  normalize: true
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "KaraKaraWitch/L3.1-70b-Inori"
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"])