README.md

metadata

language:
  - en
  - zh
license: apache-2.0
tags:
  - RLHF
  - PPO
  - custom-research
base_model:
  - sthenno/tempesthenno-14b-nuslerp-0111
  - sthenno/tempesthenno-hs2-rm
datasets:
  - nvidia/HelpSteer2
metrics:
  - accuracy
model-index:
  - name: tempesthenno-ppo-ckpt40
    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: 79.23
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          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: 50.57
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          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: 34.21
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          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: 17
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          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: 14.56
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          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: 47.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sthenno/tempesthenno-ppo-ckpt40
          name: Open LLM Leaderboard

tempesthenno--nuslerp (BASE MODEL)

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the NuSLERP merge method.

Models Merged

The following models were included in the merge:

• /Users/sthenno/models/tempesthenno--converge-dtask
• /Users/sthenno/models/tempesthenno--converge-breadcrumbs

Configuration

The following YAML configuration was used to produce this model:

name: tempesthenno--nuslerp
merge_method: nuslerp
tokenizer:
  source: /Users/sthenno/models/tempesthenno--converge-dtask
chat_template: "chatml"
dtype: float32
out_dtype: bfloat16
parameters:
  int8_mask: false
  normalize: true
  rescale: false
slices:
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [0, 8]
        parameters:
          weight: 0.65
          nuslerp_flatten: false
          nuslerp_row_wise: true
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [0, 8]
        parameters:
          weight: 0.35
          nuslerp_flatten: false
          nuslerp_row_wise: true
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [8, 16]
        parameters:
          weight: 0.60
          nuslerp_flatten: false
          nuslerp_row_wise: true
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [8, 16]
        parameters:
          weight: 0.40
          nuslerp_flatten: false
          nuslerp_row_wise: true
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [16, 24]
        parameters:
          weight: 0.55
          nuslerp_flatten: false
          nuslerp_row_wise: false
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [16, 24]
        parameters:
          weight: 0.45
          nuslerp_flatten: false
          nuslerp_row_wise: false
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [24, 32]
        parameters:
          weight: 0.50
          nuslerp_flatten: false
          nuslerp_row_wise: false
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [24, 32]
        parameters:
          weight: 0.50
          nuslerp_flatten: false
          nuslerp_row_wise: false
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [32, 40]
        parameters:
          weight: 0.45
          nuslerp_flatten: true
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [32, 40]
        parameters:
          weight: 0.55
          nuslerp_flatten: true
  - sources:
      - model: /Users/sthenno/models/tempesthenno--converge-dtask
        layer_range: [40, 48]
        parameters:
          weight: 0.40
          nuslerp_flatten: true
      - model: /Users/sthenno/models/tempesthenno--converge-breadcrumbs
        layer_range: [40, 48]
        parameters:
          weight: 0.60
          nuslerp_flatten: true

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	40.55
IFEval (0-Shot)	79.23
BBH (3-Shot)	50.57
MATH Lvl 5 (4-Shot)	34.21
GPQA (0-shot)	17.00
MuSR (0-shot)	14.56
MMLU-PRO (5-shot)	47.69