README.md

metadata

language:
  - ko
license: cc-by-nc-sa-4.0
library_name: transformers
tags:
  - moe
  - merge
pipeline_tag: text-generation
model-index:
  - name: SOLARC-MOE-10.7Bx4
    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: 70.99
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          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: 88.43
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          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: 66.34
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          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: 71.91
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          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: 83.58
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          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: 64.37
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=DopeorNope/SOLARC-MOE-10.7Bx4
          name: Open LLM Leaderboard

The license is cc-by-nc-sa-4.0.

🐻‍❄️SOLARC-MOE-10.7Bx4🐻‍❄️

Model Details

Model Developers Seungyoo Lee(DopeorNope)

I am in charge of Large Language Models (LLMs) at Markr AI team in South Korea.

Input Models input text only.

Output Models generate text only.

Model Architecture
SOLARC-MOE-10.7Bx4 is an auto-regressive language model based on the SOLAR architecture.

---

Base Model

kyujinpy/Sakura-SOLAR-Instruct

Weyaxi/SauerkrautLM-UNA-SOLAR-Instruct

VAGOsolutions/SauerkrautLM-SOLAR-Instruct

fblgit/UNA-SOLAR-10.7B-Instruct-v1.0

Implemented Method

I have built a model using the Mixture of Experts (MOE) approach, utilizing each of these models as the base.

---

Implementation Code

Load model

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

repo = "DopeorNope/SOLARC-MOE-10.7Bx4"
OpenOrca = AutoModelForCausalLM.from_pretrained(
        repo,
        return_dict=True,
        torch_dtype=torch.float16,
        device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)

---

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	74.27
AI2 Reasoning Challenge (25-Shot)	70.99
HellaSwag (10-Shot)	88.43
MMLU (5-Shot)	66.34
TruthfulQA (0-shot)	71.91
Winogrande (5-shot)	83.58
GSM8k (5-shot)	64.37