README.md

metadata

license: mit
model-index:
  - name: blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
    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: 34.9
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          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: 63.11
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          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: 26.75
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          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: 37.33
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          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: 57.14
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          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: 0.76
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=alnrg2arg/blockchainlabs_tinyllama_fusion_LHK_yunkong_v2
          name: Open LLM Leaderboard

This model is based on the fusion strategy offered by Fanqi Wan(https://github.com/fanqiwan/FuseLLM).

Three models are fused together. 10epochs

Base model: TinyLlama/TinyLlama-1.1B-Chat-v1.0

Blending model 1: HanNayeoniee/LHK_DPO_v1

Blending model 2: yunconglong/Truthful_DPO_TomGrc_FusionNet_7Bx2_MoE_13B

This model will be optimized by Laser and DPO later.

This project is to make the on-device sLM. We are doing experiments on the models.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	36.67
AI2 Reasoning Challenge (25-Shot)	34.90
HellaSwag (10-Shot)	63.11
MMLU (5-Shot)	26.75
TruthfulQA (0-shot)	37.33
Winogrande (5-shot)	57.14
GSM8k (5-shot)	0.76