metadata

license: cc-by-nc-4.0
tags:
  - DPO
model-index:
  - name: SJ-SOLAR-10.7b-DPO
    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: 68.26
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          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: 86.95
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          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: 66.73
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          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: 67.74
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          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: 84.21
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          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: 62.09
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=SJ-Donald/SJ-SOLAR-10.7b-DPO
          name: Open LLM Leaderboard

SJ-Donald/SJ-SOLAR-10.7b-DPO

SJ-Donald/SJ-SOLAR-10.7b-DPO is fine-tuned using DPO method.

Environment

Using Google CoLab A100

Base model

SJ-Donald/SOLAR-10.7B-slerp

Datasets

SJ-Donald/orca-dpo-pairs-ko

Benchmark

Open-LLM-Leaderboard(https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)

Average	ARC	HellaSwag	MMLU	TruthfulQA	Winogrande	GSM8K
72.67	68.26	86.95	66.73	67.74	84.21	62.03

open-ko-llm-leaderboard(https://huggingface.co/spaces/upstage/open-ko-llm-leaderboard)

Average	Ko-ARC	Ko-HellaSwag	Ko-MMLU	Ko-TruthfulQA	Ko-CommonGen V2
56.93	53.67	61.99	53.36	57.2	58.44

How to use

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = 'SJ-Donald/SJ-SOLAR-10.7b-DPO'

tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo,
    return_dict=True,
    torch_dtype=torch.float16,
    device_map='auto'
)

Chat Template

template = """### System:
{{system_content}}

### User:
{{question}}

### Assistant:
"""

GGUF Version

You can use gguf model file here! -> SJ-Donald/SJ-SOLAR-10.7b-DPO-GGUF

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	72.67
AI2 Reasoning Challenge (25-Shot)	68.26
HellaSwag (10-Shot)	86.95
MMLU (5-Shot)	66.73
TruthfulQA (0-shot)	67.74
Winogrande (5-shot)	84.21
GSM8k (5-shot)	62.09