README.md

metadata

tags:
  - merge
  - mergekit
  - lazymergekit
  - SciPhi/SciPhi-Mistral-7B-32k
  - SciPhi/SciPhi-Mistral-7B-32k
base_model:
  - SciPhi/SciPhi-Mistral-7B-32k
  - SciPhi/SciPhi-Mistral-7B-32k

SciPhi-Mistral-7B-32k-sliced

This is purely an experiment in sliced layer extraction to find active layers.

SciPhi-Mistral-7B-32k-sliced is a merge of the following models using LazyMergekit:

• SciPhi/SciPhi-Mistral-7B-32k
• SciPhi/SciPhi-Mistral-7B-32k

🧩 Configuration

slices:
  - sources:
      - model: SciPhi/SciPhi-Mistral-7B-32k
        layer_range: [0, 6]
      - model: SciPhi/SciPhi-Mistral-7B-32k
        layer_range: [26, 32]

merge_method: slerp
base_model: SciPhi/SciPhi-Mistral-7B-32k

parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
tokenizer_source: union

dtype: float16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "jtatman/SciPhi-Mistral-7B-32k-sliced"
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"])