README.md

metadata

license: apache-2.0
base_model:
  - OpenPipe/mistral-ft-optimized-1218
  - mlabonne/NeuralHermes-2.5-Mistral-7B
tags:
  - merge
  - mergekit
model-index:
  - name: NeuralPipe-7B-slerp
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 67.75
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 86.15
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 63.94
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 59.8
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 79.64
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 69.75
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/NeuralPipe-7B-slerp
          name: Open LLM Leaderboard

NeuralPipe-7B

This model is a merge of the following models made with mergekit:

• OpenPipe/mistral-ft-optimized-1218
• mlabonne/NeuralHermes-2.5-Mistral-7B

⚡ Quantized models

Thanks to TheBloke and ZeroWw for the quantized models:

• GGUF: https://huggingface.co/TheBloke/NeuralPipe-7B-slerp-GGUF
• GGUF f16.qX: https://huggingface.co/ZeroWw/NeuralPipe-7B-slerp-GGUF
• AWQ: https://huggingface.co/TheBloke/NeuralPipe-7B-slerp-AWQ
• GPTQ: https://huggingface.co/TheBloke/NeuralPipe-7B-slerp-GPTQ

🧩 Configuration

slices:
  - sources:
      - model: OpenPipe/mistral-ft-optimized-1218
        layer_range: [0, 32]
      - model: mlabonne/NeuralHermes-2.5-Mistral-7B
        layer_range: [0, 32]
merge_method: slerp
base_model: OpenPipe/mistral-ft-optimized-1218
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
dtype: bfloat16

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "mlabonne/NeuralPipe-7B-slerp"
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"])

Output:

A large language model is an AI system that uses deep learning techniques to process and understand vast amounts of natural language data. It is designed to generate human-like text, perform complex language tasks, and understand the context, nuance, and meaning of textual data. These models are trained on large datasets, often including billions of words, to learn the patterns and relationships in language. As a result, they can generate coherent and contextually relevant text, answer questions, and perform a variety of other language-related tasks. Some well-known large language models include OpenAI's GPT-3, Google's BERT, and Facebook's RoBERTa.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	71.17
AI2 Reasoning Challenge (25-Shot)	67.75
HellaSwag (10-Shot)	86.15
MMLU (5-Shot)	63.94
TruthfulQA (0-shot)	59.80
Winogrande (5-shot)	79.64
GSM8k (5-shot)	69.75