Edit model card


MoMo-72B-lora-1.8.7-DPO is trained via Direct Preference Optimization(DPO) from MoMo-72B-LoRA-V1.4 as its base model, with several optimizations in hyperparameters.
MoMo-72B-LoRA-V1.4 is trained via Supervised Fine-Tuning (SFT) using LoRA, with the QWEN-72B model as its base-model.
Note that we did not exploit any form of weight merge.
For leaderboard submission, the trained weight is realigned for compatibility with llama.
MoMo-72B is trained using Moreh's MoAI platform, which simplifies the training of large-scale models, and AMD's MI250 GPU.


Used Librarys

  • torch
  • peft

Used Datasets

Model ARC MMLU TruthfulQA GSM8K
V1.8.7(result < 0.1, %) TBU TBU 0.44 0.47

Used Environments

How to use

# pip install transformers==4.35.2
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("moreh/MoMo-72B-lora-1.8.7-DPO")
model = AutoModelForCausalLM.from_pretrained(
Downloads last month
Model size
72.3B params
Tensor type

Spaces using moreh/MoMo-72B-lora-1.8.7-DPO 4