README.md

metadata

license: apache-2.0
tags:
  - merge
  - mergekit
  - lazymergekit
  - mistralai/Mistral-7B-Instruct-v0.2
  - beowolx/CodeNinja-1.0-OpenChat-7B
base_model:
  - mistralai/Mistral-7B-Instruct-v0.2
  - beowolx/CodeNinja-1.0-OpenChat-7B
model-index:
  - name: Hugo-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: 64.51
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-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: 84.77
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-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: 62.54
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-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: 57.13
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-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: 80.03
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-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: 53.45
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=paulilioaica/Hugo-7B-slerp
          name: Open LLM Leaderboard

Hugo-7B-slerp

Hugo-7B-slerp is a successful merge of the following models using mergekit:

• mistralai/Mistral-7B-Instruct-v0.2
• beowolx/CodeNinja-1.0-OpenChat-7B

🧩 Configuration

slices:
  - sources:
      - model: mistralai/Mistral-7B-Instruct-v0.2
        layer_range: [0, 32]
      - model: beowolx/CodeNinja-1.0-OpenChat-7B
        layer_range: [0, 32]
merge_method: slerp
base_model: mistralai/Mistral-7B-Instruct-v0.2
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

📈 Performance

Model	Average	ARC	HellaSwag	MMLU	TruthfulQA	Winogrande	GSM8K
paulilioaica/Hugo-7B-slerp	67.07	64.51	84.77	62.54	57.13	80.03	53.45
mistralai/Mistral-7B-Instruct-v0.2	65.71	63.14	84.88	60.78	68.26	77.19	40.03
beowolx/CodeNinja-1.0-OpenChat-7B	67.4	63.48	83.65	63.77	47.16	79.79	66.57

With bold one can see the benchmarks where this merge overtakes the basemodel in performance.

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "paulilioaica/Hugo-7B-slerp"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "conversational",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(messages, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs)

🛈 More on megekit

mergekit

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	67.07
AI2 Reasoning Challenge (25-Shot)	64.51
HellaSwag (10-Shot)	84.77
MMLU (5-Shot)	62.54
TruthfulQA (0-shot)	57.13
Winogrande (5-shot)	80.03
GSM8k (5-shot)	53.45