Quantization made by Richard Erkhov.

Llama-2-13b-hf-4bit-64rank - GGUF

Model creator: https://huggingface.co/LoftQ/
Original model: https://huggingface.co/LoftQ/Llama-2-13b-hf-4bit-64rank/

Name	Quant method	Size
Llama-2-13b-hf-4bit-64rank.Q2_K.gguf	Q2_K	4.52GB
Llama-2-13b-hf-4bit-64rank.IQ3_XS.gguf	IQ3_XS	4.99GB
Llama-2-13b-hf-4bit-64rank.IQ3_S.gguf	IQ3_S	5.27GB
Llama-2-13b-hf-4bit-64rank.Q3_K_S.gguf	Q3_K_S	5.27GB
Llama-2-13b-hf-4bit-64rank.IQ3_M.gguf	IQ3_M	5.57GB
Llama-2-13b-hf-4bit-64rank.Q3_K.gguf	Q3_K	5.9GB
Llama-2-13b-hf-4bit-64rank.Q3_K_M.gguf	Q3_K_M	5.9GB
Llama-2-13b-hf-4bit-64rank.Q3_K_L.gguf	Q3_K_L	6.45GB
Llama-2-13b-hf-4bit-64rank.IQ4_XS.gguf	IQ4_XS	6.54GB
Llama-2-13b-hf-4bit-64rank.Q4_0.gguf	Q4_0	6.86GB
Llama-2-13b-hf-4bit-64rank.IQ4_NL.gguf	IQ4_NL	6.9GB
Llama-2-13b-hf-4bit-64rank.Q4_K_S.gguf	Q4_K_S	6.91GB
Llama-2-13b-hf-4bit-64rank.Q4_K.gguf	Q4_K	7.33GB
Llama-2-13b-hf-4bit-64rank.Q4_K_M.gguf	Q4_K_M	7.33GB
Llama-2-13b-hf-4bit-64rank.Q4_1.gguf	Q4_1	7.61GB
Llama-2-13b-hf-4bit-64rank.Q5_0.gguf	Q5_0	8.36GB
Llama-2-13b-hf-4bit-64rank.Q5_K_S.gguf	Q5_K_S	8.36GB
Llama-2-13b-hf-4bit-64rank.Q5_K.gguf	Q5_K	8.6GB
Llama-2-13b-hf-4bit-64rank.Q5_K_M.gguf	Q5_K_M	8.6GB
Llama-2-13b-hf-4bit-64rank.Q5_1.gguf	Q5_1	9.1GB
Llama-2-13b-hf-4bit-64rank.Q6_K.gguf	Q6_K	9.95GB
Llama-2-13b-hf-4bit-64rank.Q8_0.gguf	Q8_0	12.88GB

Original model description:

license: mit language: - en pipeline_tag: text-generation tags: - 'quantization ' - lora

LoftQ Initialization

| Paper | Code | PEFT Example |

LoftQ (LoRA-fine-tuning-aware Quantization) provides a quantized backbone Q and LoRA adapters A and B, given a full-precision pre-trained weight W.

This model, Llama-2-13b-hf-4bit-64rank, is obtained from LLAMA-2-13b. The backbone is under LoftQ/Llama-2-13b-hf-4bit-64rank and LoRA adapters are under the subfolder='loftq_init'.

Model Info

Backbone

Stored format: torch.bfloat16
Size: ~ 26 GiB
Loaded format: bitsandbytes nf4
Size loaded on GPU: ~6.5 GiB

LoRA adapters

rank: 64
lora_alpha: 64
target_modules: ["down_proj", "up_proj", "q_proj", "k_proj", "v_proj", "o_proj", "gate_proj"]

Usage

Training Here's an example of loading this model and preparing for the LoRA fine-tuning.

import torch
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

MODEL_ID = "LoftQ/Llama-2-13b-hf-4bit-64rank"

base_model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID, 
    torch_dtype=torch.bfloat16,  # you may change it with different models
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=torch.bfloat16,  # bfloat16 is recommended
        bnb_4bit_use_double_quant=False,
        bnb_4bit_quant_type='nf4',
    ),
)
peft_model = PeftModel.from_pretrained(
    base_model,
    MODEL_ID,
    subfolder="loftq_init",
    is_trainable=True,
)

# Do training with peft_model ...

Experiment Results

We have conducted experiments on supervised fine-tuning of GSM8K and WikiText-2.

Model	Bits	Rank	LoRA Initial	GSM8K	WikiText-2
LLAMA-2-13b	16	64	Gaussian + 0	45.3	5.12
LLAMA-2-13b	4	64	Gaussian + 0 (QLoRA)	39.9	5.22
LLAMA-2-13b	4	64	LoftQ	45.0	5.16

Inference Here is an example code for inference after the model has been fine-tuned on GSM8K.

import torch
from transformers import AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel

MODEL_ID = "LoftQ/Llama-2-13b-hf-4bit-64rank"

base_model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID, 
    torch_dtype=torch.bfloat16,  # you may change it with different models
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=torch.bfloat16,  # bfloat16 is recommended
        bnb_4bit_use_double_quant=False,
        bnb_4bit_quant_type='nf4',
    ),
)
peft_model = PeftModel.from_pretrained(
    base_model,
    MODEL_ID,
    subfolder="gsm8k",
    is_trainable=True,
)

# Do inference with peft_model ...

See the full code at our Github Repo

Citation

@article{li2023loftq,
  title={Loftq: Lora-fine-tuning-aware quantization for large language models},
  author={Li, Yixiao and Yu, Yifan and Liang, Chen and He, Pengcheng and Karampatziakis, Nikos and Chen, Weizhu and Zhao, Tuo},
  journal={arXiv preprint arXiv:2310.08659},
  year={2023}
}