README.md

metadata

language:
  - en
license: cc-by-nc-4.0
tags:
  - mistral
datasets:
  - pankajmathur/orca_mini_v1_dataset
  - openai/summarize_from_feedback
  - PygmalionAI/PIPPA
  - chargoddard/rpguild
  - lemonilia/LimaRP
model-index:
  - name: loyal-piano-m7
    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: 66.72
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          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: 85.03
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          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: 64.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          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: 60.03
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          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: 79.08
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          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: 56.71
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chargoddard/loyal-piano-m7
          name: Open LLM Leaderboard

Experimenting with dataset ratios. Intended to be a roleplay-focused model with some smarts and good long-context recall.

Not sure if I've succeeded on the roleplay front, but something sure went right! Currently the #4 7B model on the leaderboard as of 11/30/2023. Going to riff on this and see where it goes.

model	Average	ARC	HellaSwag	MMLU	TruthfulQA	Winogrande	GSM8K	DROP
fblgit/juanako-7b-UNA	59.91	68.17	85.34	62.47	65.13	78.85	20.7	38.74
Intel/neural-chat-7b-v3-1	59.06	66.21	83.64	62.37	59.65	78.14	19.56	43.84
Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-1-7B	58.6	66.55	84.47	63.34	61.22	78.37	23.58	32.66
chargoddard/loyal-piano-m7	58.42	66.72	85.03	64.43	60.03	79.08	25.7	27.92
Gryphe/MythoMist7b	58.26	65.87	83.55	62.32	59.98	78.06	20.24	37.82

Dataset composition:

dataset	rows used	percent of total
PIPPA	14.6k	43%
summarize_from_feedback	9k	26%
orca_mini_v1_dataset	5.6k	17%
rpguild	2.86k	8%
LimaRP	2k	6%

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	68.67
AI2 Reasoning Challenge (25-Shot)	66.72
HellaSwag (10-Shot)	85.03
MMLU (5-Shot)	64.43
TruthfulQA (0-shot)	60.03
Winogrande (5-shot)	79.08
GSM8k (5-shot)	56.71