README.md

metadata

license: other
tags:
  - mergekit
  - merge
base_model:
  - Qwen/Qwen2.5-3B
  - Qwen/Qwen2.5-3B-Instruct
  - arcee-ai/raspberry-3B
license_name: qwen-research
license_link: https://huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/main/LICENSE
model-index:
  - name: Rombos-LLM-V2.5.1-Qwen-3b
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 25.95
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 14.88
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 8.31
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 3.24
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 7.82
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 19.1
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=rombodawg/Rombos-LLM-V2.5.1-Qwen-3b
          name: Open LLM Leaderboard

Rombos-LLM-V2.5.1-Qwen-3b

A little experiment I threw together to take a really high quality LLM I found (arcee-ai/raspberry-3B) and merge it using the last step of my Continuous Finetuning method outlines in the paper linked bellow.

https://docs.google.com/document/d/1OjbjU5AOz4Ftn9xHQrX3oFQGhQ6RDUuXQipnQ9gn6tU/edit?usp=sharing

Mergekit.yaml file is as follows:

models:
  - model: Qwen2.5-3B-Instruct
    parameters:
      weight: 1
      density: 1
  - model: raspberry-3B
    parameters:
      weight: 1
      density: 1
merge_method: ties
base_model: Qwen2.5-3B
parameters:
  weight: 1
  density: 1
  normalize: true
  int8_mask: true
dtype: bfloat16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	13.22
IFEval (0-Shot)	25.95
BBH (3-Shot)	14.88
MATH Lvl 5 (4-Shot)	8.31
GPQA (0-shot)	3.24
MuSR (0-shot)	7.82
MMLU-PRO (5-shot)	19.10