djinn

djinn is a merge of the following models using LazyMergekit:

• CultriX/NeuralTrix-7B-dpo
• paulml/DPOB-INMTOB-7B

🧩 Configuration

merge_method: linear
parameters:
  weight: 1.0 
slices:
  - sources:
      - model: CultriX/NeuralTrix-7B-dpo # embed_tokens comes along with the ride with whatever is the first layer
        layer_range: [0, 1]
      - model: paulml/DPOB-INMTOB-7B # add dummy second model with 0 weight so tokenizer-based merge routine is invoked for embed_tokens
        layer_range: [0, 1]
        parameters:
          weight: 0
  - sources:
      - model: cognitivecomputations/dolphin-2.1-mistral-7b
        layer_range: [0, 8]
  - sources:
      - model: bardsai/jaskier-7b-dpo-v5.6
        layer_range: [8, 16]
  - sources:
      - model: paulml/OGNO-7B
        layer_range: [16, 24]
  - sources:
      - model: argilla/distilabeled-OpenHermes-2.5-Mistral-7B
        layer_range: [24, 31]
  - sources: # same as above, but for lm_head with the last layer
      - model: CultriX/NeuralTrix-7B-dpo
        layer_range: [31, 32]
      - model: paulml/DPOB-INMTOB-7B
        layer_range: [31, 32]
        parameters:
          weight: 0
dtype: float16
tokenizer_source: model:cognitivecomputations/dolphin-2.1-mistral-7b

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mayacinka/djinn"
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"])