compressed-llm/vicuna-13b-v1.3-gptq

Compressed LLM Model Zone

The models are prepared by Visual Informatics Group @ University of Texas at Austin (VITA-group). Credits to Ajay Jaiswal, Zhenyu Zhang, Zhangheng Li, Lu Yin, Shiwei Liu and Junyuan Hong.

License: MIT License

Setup environment

pip install torch==2.0.0+cu117 torchvision==0.15.1+cu117 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu117
pip install transformers==4.31.0
pip install accelerate
pip install auto-gptq  # for gptq
pip install sentencepiece

How to use pruned models

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
base_model = 'llama-2-7b'
comp_method = 'magnitude_unstructured'
comp_degree = 0.2
model_path = f'vita-group/{base_model}_{comp_method}'
model = AutoModelForCausalLM.from_pretrained(
        model_path, 
        revision=f's{comp_degree}',
        torch_dtype=torch.float16, 
        low_cpu_mem_usage=True, 
        device_map="auto"
    )
tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-7b-hf')
input_ids = tokenizer('Hello! I am a VITA-compressed-LLM chatbot!', return_tensors='pt').input_ids.cuda()
outputs = model.generate(input_ids, max_new_tokens=128)
print(tokenizer.decode(outputs[0]))

How to use wanda+gptq models

from transformers import AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM
model_path = 'vita-group/llama-2-7b_wanda_2_4_gptq_4bit_128g'
tokenizer_path = 'meta-llama/Llama-2-7b-hf'
model = AutoGPTQForCausalLM.from_quantized(
        model_path,
        # inject_fused_attention=False, # or 
        disable_exllama=True,
        device_map='auto',
    )
tokenizer = AutoTokenizer.from_pretrained(tokenizer_path, trust_remote_code=True)
input_ids = tokenizer('Hello! I am a VITA-compressed-LLM chatbot!', return_tensors='pt').input_ids.to('cuda')
outputs = model.generate(input_ids=input_ids, max_length=128)
tokenizer.decode(outputs[0])

How to use gptq models

from transformers import AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM
# model_path = 'vita-group/llama-2-7b_wanda_2_4_gptq_4bit_128g'
# tokenizer_path = 'meta-llama/Llama-2-7b-hf'
model_path = 'vita-group/vicuna-7b-v1.3_gptq'
tokenizer_path = 'lmsys/vicuna-7b-v1.3'
model = AutoGPTQForCausalLM.from_quantized(
        model_path,
        # inject_fused_attention=False, # or 
        disable_exllama=True,
        device_map='auto',
        revision='2bit_128g',
    )
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(tokenizer_path, trust_remote_code=True)
input_ids = tokenizer('Hello! I am a VITA-compressed-LLM chatbot!', return_tensors='pt').input_ids.to('cuda')
outputs = model.generate(input_ids=input_ids, max_length=128)
tokenizer.decode(outputs[0])

	Base Model	Model Size	Compression Method	Compression Degree
0	Llama-2	7b	magnitude_unstructured	s0.1
1	Llama-2	7b	magnitude_unstructured	s0.2
2	Llama-2	7b	magnitude_unstructured	s0.3
3	Llama-2	7b	magnitude_unstructured	s0.5
4	Llama-2	7b	magnitude_unstructured	s0.6
5	Llama-2	7b	sparsegpt_unstructured	s0.1
6	Llama-2	7b	sparsegpt_unstructured	s0.2
7	Llama-2	7b	sparsegpt_unstructured	s0.3
8	Llama-2	7b	sparsegpt_unstructured	s0.5
9	Llama-2	7b	sparsegpt_unstructured	s0.6
10	Llama-2	7b	wanda_gptq	4bit_128g
11	Llama-2	7b	wanda_unstructured	s0.1
12	Llama-2	7b	wanda_unstructured	s0.2
13	Llama-2	7b	wanda_unstructured	s0.3
14	Llama-2	7b	wanda_unstructured	s0.5
15	Llama-2	7b	wanda_unstructured	s0.6
16	vicuna-v1.3	13b	gptq	10bit_128g
17	vicuna-v1.3	13b	gptq	12bit_128g
18	vicuna-v1.3	13b	gptq	14bit_128g
19	vicuna-v1.3	13b	gptq	2bit_128g
20	vicuna-v1.3	13b	gptq	3bit_128g
21	vicuna-v1.3	13b	gptq	4bit_128g
22	vicuna-v1.3	13b	gptq	6bit_128g
23	vicuna-v1.3	13b	gptq	8bit_128g
24	vicuna-v1.3	7b	gptq	10bit_128g
25	vicuna-v1.3	7b	gptq	12bit_128g
26	vicuna-v1.3	7b	gptq	14bit_128g
27	vicuna-v1.3	7b	gptq	2bit_128g
28	vicuna-v1.3	7b	gptq	3bit_128g
29	vicuna-v1.3	7b	gptq	4bit_128g
30	vicuna-v1.3	7b	gptq	6bit_128g
31	vicuna-v1.3	7b	gptq	8bit_128g