RichardErkhov/moreh_-_MoMo-72B-lora-1.8.6-DPO-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

MoMo-72B-lora-1.8.6-DPO - GGUF

• Model creator: https://huggingface.co/moreh/
• Original model: https://huggingface.co/moreh/MoMo-72B-lora-1.8.6-DPO/

Name	Quant method	Size
MoMo-72B-lora-1.8.6-DPO.Q2_K.gguf	Q2_K	25.22GB
MoMo-72B-lora-1.8.6-DPO.IQ3_XS.gguf	IQ3_XS	27.88GB
MoMo-72B-lora-1.8.6-DPO.IQ3_S.gguf	IQ3_S	29.4GB
MoMo-72B-lora-1.8.6-DPO.Q3_K_S.gguf	Q3_K_S	29.4GB
MoMo-72B-lora-1.8.6-DPO.IQ3_M.gguf	IQ3_M	30.98GB
MoMo-72B-lora-1.8.6-DPO.Q3_K.gguf	Q3_K	32.85GB
MoMo-72B-lora-1.8.6-DPO.Q3_K_M.gguf	Q3_K_M	32.85GB
MoMo-72B-lora-1.8.6-DPO.Q3_K_L.gguf	Q3_K_L	35.85GB
MoMo-72B-lora-1.8.6-DPO.IQ4_XS.gguf	IQ4_XS	36.41GB
MoMo-72B-lora-1.8.6-DPO.Q4_0.gguf	Q4_0	38.19GB
MoMo-72B-lora-1.8.6-DPO.IQ4_NL.gguf	IQ4_NL	38.42GB
MoMo-72B-lora-1.8.6-DPO.Q4_K_S.gguf	Q4_K_S	38.45GB
MoMo-72B-lora-1.8.6-DPO.Q4_K.gguf	Q4_K	40.77GB
MoMo-72B-lora-1.8.6-DPO.Q4_K_M.gguf	Q4_K_M	40.77GB
MoMo-72B-lora-1.8.6-DPO.Q4_1.gguf	Q4_1	42.32GB
MoMo-72B-lora-1.8.6-DPO.Q5_0.gguf	Q5_0	46.46GB
MoMo-72B-lora-1.8.6-DPO.Q5_K_S.gguf	Q5_K_S	46.46GB
MoMo-72B-lora-1.8.6-DPO.Q5_K.gguf	Q5_K	47.79GB
MoMo-72B-lora-1.8.6-DPO.Q5_K_M.gguf	Q5_K_M	47.79GB
MoMo-72B-lora-1.8.6-DPO.Q5_1.gguf	Q5_1	50.59GB
MoMo-72B-lora-1.8.6-DPO.Q6_K.gguf	Q6_K	55.24GB
MoMo-72B-lora-1.8.6-DPO.Q8_0.gguf	Q8_0	71.55GB

Original model description:

license: mit language: - en

Introduction

MoMo-72B-lora-1.8.6-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.

Details

Used Librarys

• torch
• peft

Used Datasets

• slimorca
• truthy
• orca_dpo_pairs
• No other dataset was used
• No benchmark test set or the training set are used
  • data contamination check result

Model	ARC	MMLU	TruthfulQA	GSM8K
V1.8.6(result < 0.1, %)	TBU	TBU	0.73	TBU

Used Environments

• AMD MI250 & MoAI platform
• Please visit https://moreh.io/product for more information about MoAI platform
• Or, contact us directly contact@moreh.io

How to use

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

tokenizer = AutoTokenizer.from_pretrained("moreh/MoMo-72B-lora-1.8.6-DPO")
model = AutoModelForCausalLM.from_pretrained(
    "moreh/MoMo-72B-lora-1.8.6-DPO"
)