Quantization made by Richard Erkhov.

MoMo-72B-lora-1.8.7-DPO - GGUF

Model creator: https://huggingface.co/moreh/
Original model: https://huggingface.co/moreh/MoMo-72B-lora-1.8.7-DPO/

Name	Quant method	Size
MoMo-72B-lora-1.8.7-DPO.Q2_K.gguf	Q2_K	25.22GB
MoMo-72B-lora-1.8.7-DPO.Q3_K_S.gguf	Q3_K_S	29.4GB
MoMo-72B-lora-1.8.7-DPO.Q3_K.gguf	Q3_K	32.85GB
MoMo-72B-lora-1.8.7-DPO.Q3_K_M.gguf	Q3_K_M	22.65GB
MoMo-72B-lora-1.8.7-DPO.Q3_K_L.gguf	Q3_K_L	35.85GB
MoMo-72B-lora-1.8.7-DPO.IQ4_XS.gguf	IQ4_XS	12.38GB
MoMo-72B-lora-1.8.7-DPO.Q4_0.gguf	Q4_0	38.19GB
MoMo-72B-lora-1.8.7-DPO.IQ4_NL.gguf	IQ4_NL	22.78GB
MoMo-72B-lora-1.8.7-DPO.Q4_K_S.gguf	Q4_K_S	38.45GB
MoMo-72B-lora-1.8.7-DPO.Q4_K.gguf	Q4_K	40.77GB
MoMo-72B-lora-1.8.7-DPO.Q4_K_M.gguf	Q4_K_M	40.77GB
MoMo-72B-lora-1.8.7-DPO.Q4_1.gguf	Q4_1	42.32GB
MoMo-72B-lora-1.8.7-DPO.Q5_0.gguf	Q5_0	46.46GB
MoMo-72B-lora-1.8.7-DPO.Q5_K_S.gguf	Q5_K_S	46.46GB
MoMo-72B-lora-1.8.7-DPO.Q5_K.gguf	Q5_K	30.07GB
MoMo-72B-lora-1.8.7-DPO.Q5_K_M.gguf	Q5_K_M	44.32GB
MoMo-72B-lora-1.8.7-DPO.Q5_1.gguf	Q5_1	34.26GB
MoMo-72B-lora-1.8.7-DPO.Q6_K.gguf	Q6_K	55.24GB
MoMo-72B-lora-1.8.7-DPO.Q8_0.gguf	Q8_0	71.55GB

Original model description:

license: mit language: - en metrics: - accuracy library_name: transformers

24/04/05 update

We introduce Moreh AI Model Hub with AMD GPU, an ai model host platform powered by AMD MI250 GPUs. You can now test live-inference of this model at Moreh AI Model Hub.

Introduction

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.

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
V1.8.7(result < 0.1, %)	TBU	TBU	0.44	0.47

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-1.8.7-DPO")
model = AutoModelForCausalLM.from_pretrained(
    "moreh/MoMo-72B-lora-1.8.7-DPO"
)