Pretergeek_-_OpenChat-3.5-0106_10.7B_48Layers-Interleaved-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

OpenChat-3.5-0106_10.7B_48Layers-Interleaved - GGUF

• Model creator: https://huggingface.co/Pretergeek/
• Original model: https://huggingface.co/Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved/

Original model description:

license: apache-2.0 library_name: transformers tags:

• mergekit
• merge base_model:
• openchat/openchat-3.5-0106 model-index:
• name: OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 59.61 name: strict accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 24.06 name: normalized accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 6.8 name: exact match source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 7.27 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 11.78 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved 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: 25.54 name: accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Pretergeek/OpenChat-3.5-0106_10.7B_48Layers-Interleaved name: Open LLM Leaderboard

---

<p align="center"> <a href="https://ko-fi.com/pretergeek">Buy me a Ko-Fi</a> • <a href="https://patreon.com/Pretergeek">Support my work using Patreon</a> </p>

OpenChat-3.5-0106_10.7B_48Layers-Interleaved

This is NOT your usual frankenmerge created using mergekit.

Merge Details

Merge Method

This model was merged using the passthrough merge method, but employing the Block Expansion method described in the paper LLaMA Pro: Progressive LLaMA with Block Expansion.

The authors of the paper added new layers interleaved in between the original layers of the model, setting the parameters of the o_proj and down_proj layers to zero. This effectively adds layers that will just output their input (as if they were "transparent") allowing the model to remain functional even without further training. These new layers can then be targeted during training or fine-tuning without risking catastrophic forgetting, if you follow the author's training method to freeze the original layers and only train the new layers.

This model has not yet received additional training, so it should perform close to the original model.

Models Merged

The following models were included in the merge:

• openchat/openchat-3.5-0106

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [0, 2]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [1, 2]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [2, 4]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [3, 4]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [4, 6]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [5, 6]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [6, 8]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [7, 8]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [8, 10]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [9, 10]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [10, 12]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [11, 12]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [12, 14]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [13, 14]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [14, 16]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [15, 16]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [16, 18]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [17, 18]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [18, 20]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [19, 20]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [20, 22]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [21, 22]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [22, 24]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [23, 24]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [24, 26]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [25, 26]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [26, 28]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [27, 28]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [28, 30]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [29, 30]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [30, 32]
  - sources:
    - model: openchat/openchat-3.5-0106
      layer_range: [31, 32]
      parameters:
        scale:
          - filter: o_proj
            value: 0.0
          - filter: down_proj
            value: 0.0
          - value: 1.0
merge_method: passthrough
dtype: bfloat16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Citation

@misc{wu2024llamaproprogressivellama,
      title={LLaMA Pro: Progressive LLaMA with Block Expansion}, 
      author={Chengyue Wu and Yukang Gan and Yixiao Ge and Zeyu Lu and Jiahao Wang and Ye Feng and Ying Shan and Ping Luo},
      year={2024},
      eprint={2401.02415},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2401.02415}, 
}