README.md

metadata

language:
  - en
license: llama3
model-index:
  - name: L3-8B-Lunaris-v1
    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: 68.95
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          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: 32.11
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          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: 8.46
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          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: 6.82
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          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: 5.55
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          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: 30.97
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Sao10K/L3-8B-Lunaris-v1
          name: Open LLM Leaderboard

A generalist / roleplaying model merge based on Llama 3. Models are selected from my personal experience while using them.

I personally think this is an improvement over Stheno v3.2, considering the other models helped balance out its creativity and at the same time improving its logic.

Settings:

Instruct // Context Template: Llama-3-Instruct
Temperature: 1.4
min_p: 0.1

---

Merging seems to be a black box magic though? In my personal experience merging multiple models from different datasets / data works better than combining them all in one.

Values chosen are from long-running personal experimentation since Llama-2 Merging Era. I have tweaked them to fit this recipe.

Mergekit Config

models:
  - model: meta-llama/Meta-Llama-3-8B-Instruct
  - model: crestf411/L3-8B-sunfall-v0.1 # Another RP Model trained on... stuff
    parameters:
      density: 0.4
      weight: 0.25
  - model: Hastagaras/Jamet-8B-L3-MK1 - # Another RP / Storytelling Model
    parameters:
      density: 0.5
      weight: 0.3
  - model: maldv/badger-iota-llama-3-8b #Megamerge - Helps with General Knowledge
    parameters:
      density: 0.6
      weight: 0.35
  - model: Sao10K/Stheno-3.2-Beta # This is Stheno v3.2's Initial Name
    parameters:
      density: 0.7
      weight: 0.4
merge_method: ties
base_model: meta-llama/Meta-Llama-3-8B-Instruct
parameters:
  int8_mask: true
  rescale: true
  normalize: false
dtype: bfloat16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	25.48
IFEval (0-Shot)	68.95
BBH (3-Shot)	32.11
MATH Lvl 5 (4-Shot)	8.46
GPQA (0-shot)	6.82
MuSR (0-shot)	5.55
MMLU-PRO (5-shot)	30.97