language:
- en
license: llama2
tags:
- moe
model-index:
- name: orthorus-125b-moe
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: 67.66
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
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: 85.52
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
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: 68.94
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
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: 56.27
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
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: 82.32
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
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: 56.79
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=ibivibiv/orthorus-125b-moe
name: Open LLM Leaderboard
This is a test run for a future 70b parameter models moe model. I took WizardLM/WizardLM-70B-V1.0 and migtissera/Synthia-70B as two base models and created the discriminator prompts to push technical, logic, and math type questions to the Wizard side and then all creative or conversation questions to the Synthia side. Now that this is working for me I am going to move to fine tuning models for more specific tasks. This model takes about 240GB of VRAM for full resolution inference. As far as I know, it is the first 125B parameter moe model publicly available. I plan on making more and sharing of course.
Hopefully I can add more info on this model, it loads perfectly for me and responds nicely. It might take me a bit since I want to make "Cerberus" with the fine tuned models and get it released. But enjoy this one, llama2 model.
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 69.58 |
| AI2 Reasoning Challenge (25-Shot) | 67.66 |
| HellaSwag (10-Shot) | 85.52 |
| MMLU (5-Shot) | 68.94 |
| TruthfulQA (0-shot) | 56.27 |
| Winogrande (5-shot) | 82.32 |
| GSM8k (5-shot) | 56.79 |