README.md

metadata

license: apache-2.0
tags:
  - merge
  - mergekit
  - '#dpo'
  - MaximeLabonne
  - '#mergeofmerge'
base_model:
  - CultriX/NeuralTrix-7B-dpo
  - paulml/OmniBeagleSquaredMBX-v3-7B-v2
model-index:
  - name: NeuTrixOmniBe-DPO
    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: 72.78
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          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: 89.03
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          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: 64.28
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          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: 77.21
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          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: 85.16
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          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: 68.54
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Kukedlc/NeuTrixOmniBe-DPO
          name: Open LLM Leaderboard

WARNING: Not for Use - Bug INSTINST in response.

This model was merged, trained, and so on, thanks to the knowledge I gained from reading Maxime Labonne's course. Special thanks to him!

Labonne LLM Course

NeuTrixOmniBe-DPO

NeuTrixOmniBe-DPO is a merge of the following models using LazyMergekit:

🧩 Configuration

MODEL_NAME = "NeuTrixOmniBe-DPO"
yaml_config = """
slices:
  - sources:
      - model: CultriX/NeuralTrix-7B-dpo
        layer_range: [0, 32]
      - model: paulml/OmniBeagleSquaredMBX-v3-7B-v2
        layer_range: [0, 32]
merge_method: slerp
base_model: CultriX/NeuralTrix-7B-dpo
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
"""

It was then trained with DPO using:

• Intel/orca_dpo_pairs

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Kukedlc/NeuTrixOmniBe-DPO"
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=128, do_sample=True, temperature=0.5, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	76.17
AI2 Reasoning Challenge (25-Shot)	72.78
HellaSwag (10-Shot)	89.03
MMLU (5-Shot)	64.28
TruthfulQA (0-shot)	77.21
Winogrande (5-shot)	85.16
GSM8k (5-shot)	68.54