aipib_-_karasu-moe2-4bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

karasu-moe2 - bnb 4bits

• Model creator: https://huggingface.co/aipib/
• Original model: https://huggingface.co/aipib/karasu-moe2/

Original model description:

license: apache-2.0 tags:

• moe
• frankenmoe
• merge
• mergekit
• lazymergekit
• yuiseki/karasu-sake-qa-v0.1
• lightblue/karasu-1.1B base_model:
• yuiseki/karasu-sake-qa-v0.1
• lightblue/karasu-1.1B

---

karasu-moe2

karasu-moe2 is a Mixture of Experts (MoE) made with the following models using LazyMergekit:

• yuiseki/karasu-sake-qa-v0.1
• lightblue/karasu-1.1B

🧩 Configuration

base_model: lightblue/karasu-1.1B
experts:
  - source_model: yuiseki/karasu-sake-qa-v0.1
    positive_prompts:
    - "chat"
    - "assistant"
    - "instruct"
    - "explain"
  - source_model: lightblue/karasu-1.1B
    positive_prompts:
    - "storywriting"
    - "write"
    - "reason"
    - "count"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "aipib/karasu-moe2"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
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"])