Neo_7b-merge18

Neo_7b-merge18 is a merge of the following models using LazyMergekit:

• DewEfresh/neo_7b
• m-a-p/neo_7b

🧩 Configuration

# Define the slices for the model merging process
slices:
  - sources:
      # First part: merge layer 0 with layer 3
      - model: DewEfresh/neo_7b
        layer_range: [0, 1]
      - model: m-a-p/neo_7b
        layer_range: [3, 4]
  - sources:
      # Second part: merge layer 1 with layer 3
      - model: DewEfresh/neo_7b
        layer_range: [1, 2]
      - model: m-a-p/neo_7b
        layer_range: [3, 4]
  - sources:
      # Third part: merge layer 2 with layer 3
      - model: DewEfresh/neo_7b
        layer_range: [2, 3]
      - model: m-a-p/neo_7b
        layer_range: [3, 4]
  - sources:
      # Fourth part: merge layer 4 with layer 7
      - model: DewEfresh/neo_7b
        layer_range: [4, 5]
      - model: m-a-p/neo_7b
        layer_range: [7, 8]
  - sources:
      # Fifth part: merge layer 5 with layer 7
      - model: DewEfresh/neo_7b
        layer_range: [5, 6]
      - model: m-a-p/neo_7b
        layer_range: [7, 8]
  - sources:
      # Sixth part: merge layer 6 with layer 7
      - model: DewEfresh/neo_7b
        layer_range: [6, 7]
      - model: m-a-p/neo_7b
        layer_range: [7, 8]
  - sources:
      # Seventh part: merge layer 8 with layer 11
      - model: DewEfresh/neo_7b
        layer_range: [8, 9]
      - model: m-a-p/neo_7b
        layer_range: [11, 12]
  - sources:
      # Eighth part: merge layer 9 with layer 11
      - model: DewEfresh/neo_7b
        layer_range: [9, 10]
      - model: m-a-p/neo_7b
        layer_range: [11, 12]
  - sources:
      # Ninth part: merge layer 10 with layer 11
      - model: DewEfresh/neo_7b
        layer_range: [10, 11]
      - model: m-a-p/neo_7b
        layer_range: [11, 12]
  - sources:
      # Tenth part: merge layer 12 with layer 15
      - model: DewEfresh/neo_7b
        layer_range: [12, 13]
      - model: m-a-p/neo_7b
        layer_range: [15, 16]
  - sources:
      # Eleventh part: merge layer 13 with layer 15
      - model: DewEfresh/neo_7b
        layer_range: [13, 14]
      - model: m-a-p/neo_7b
        layer_range: [15, 16]
  - sources:
      # Twelfth part: merge layer 14 with layer 15
      - model: DewEfresh/neo_7b
        layer_range: [14, 15]
      - model: m-a-p/neo_7b
        layer_range: [15, 16]
  - sources:
      # Thirteenth part: merge layer 16 with layer 19
      - model: DewEfresh/neo_7b
        layer_range: [16, 17]
      - model: m-a-p/neo_7b
        layer_range: [19, 20]
  - sources:
      # Fourteenth part: merge layer 17 with layer 19
      - model: DewEfresh/neo_7b
        layer_range: [17, 18]
      - model: m-a-p/neo_7b
        layer_range: [19, 20]
  - sources:
      # Fifteenth part: merge layer 18 with layer 19
      - model: DewEfresh/neo_7b
        layer_range: [18, 19]
      - model: m-a-p/neo_7b
        layer_range: [19, 20]
  - sources:
      # Sixteenth part: merge layer 20 with layer 23
      - model: DewEfresh/neo_7b
        layer_range: [20, 21]
      - model: m-a-p/neo_7b
        layer_range: [23, 24]
  - sources:
      # Seventeenth part: merge layer 21 with layer 23
      - model: DewEfresh/neo_7b
        layer_range: [21, 22]
      - model: m-a-p/neo_7b
        layer_range: [23, 24]
  - sources:
      # Eighteenth part: merge layer 22 with layer 23
      - model: DewEfresh/neo_7b
        layer_range: [22, 23]
      - model: m-a-p/neo_7b
        layer_range: [23, 24]
  - sources:
      # Nineteenth part: merge layer 24 with layer 27
      - model: DewEfresh/neo_7b
        layer_range: [24, 25]
      - model: m-a-p/neo_7b
        layer_range: [26, 27]
  - sources:
      # Twentieth part: merge layer 25 with layer 27
      - model: DewEfresh/neo_7b
        layer_range: [25, 26]
      - model: m-a-p/neo_7b
        layer_range: [26, 27]
  - sources:
      # Twenty-first part: merge layer 26 with layer 27
      - model: DewEfresh/neo_7b
        layer_range: [26, 27]
      - model: m-a-p/neo_7b
        layer_range: [26, 27]
# Specify the merging method for the slices
merge_method: slerp
base_model: DewEfresh/neo_7b
parameters:
  t: 0.2 # Set global interpolation value to 33.33%
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "DewEfresh/Neo_7b-merge18"
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"])