metadata
license: apache-2.0
datasets:
- google/docci
- google/imageinwords
- ProGamerGov/synthetic-dataset-1m-dalle3-high-quality-captions
language:
- en
library_name: transformers
pipeline_tag: image-text-to-text
tags:
- art
base_model: gokaygokay/Florence-2-SD3-Captioner
inference: false
Original model is here. Tagger for local environment is here.
# recipe
from transformers import AutoModelForCausalLM, AutoProcessor, BitsAndBytesConfig
import transformers
import torch
import json
model_id = 'gokaygokay/Florence-2-SD3-Captioner'
save_path = 'gokaygokay-Florence-2-SD3-Captioner-8bit'
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
model_id,
trust_remote_code=True,
torch_dtype=torch.float32,
low_cpu_mem_usage=True,
quantization_config=BitsAndBytesConfig(
load_in_8bit=True,
llm_int8_threshold=6.0,
llm_int8_enable_fp32_cpu_offload=True,
llm_int8_skip_modules=['lm_head'],
),
)
processor.save_pretrained(save_path)
model.save_pretrained(save_path, safe_serialization=True)
config = {}
with open(f'{save_path}/config.json') as f:
config = json.load(f)
config['vision_config']['model_type'] = 'davit'
with open(f'{save_path}/config.json', 'w') as f:
json.dump(config, f, indent=2)