?Neo-Models | ?Neo-Datasets | Github Neo is a completely ope source large laguage model, icludig code, all model weights, datasets used for traiig, ad traiig details.NEO
Model
Model
Describe
Dowload
eo_7b
This repository cotais the base model of eo_7b
• ? Huggig Face
eo7bitermediate
This repo cotais ormal pre-traiig itermediate ckpts. A total of 3.7T tokes were leared at this phase.
• ? Huggig Face
eo7bdecay
This repo cotais itermediate ckpts durig the decay phase. A total of 720B tokes were leared at this phase.
• ? Huggig Face
eoscaliglaw980M
This repo cotais ckpts related to scaliglaw experimets
• ? Huggig Face
eoscaliglaw460M
This repo cotais ckpts related to scaliglaw experimets
• ? Huggig Face
eoscaliglaw250M
This repo cotais ckpts related to scaliglaw experimets
• ? Huggig Face
eo2bgeeral
This repo cotais ckpts of 2b model traied usig commo domai kowledge
• ? Huggig Face
Usage
from trasformers import AutoModelForCausalLM, AutoTokeizer
model_path = '<your-hf-model-path-with-tokeizer>'
tokeizer = AutoTokeizer.from_pretraied(model_path, use_fast=False, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretraied(
model_path,
device_map="auto",
torch_dtype='auto'
).eval()
iput_text = "A log, log time ago,"
iput_ids = tokeizer(iput_text, add_geeratio_prompt=True, retur_tesors='pt').to(model.device)
output_ids = model.geerate(**iput_ids, max_ew_tokes=20)
respose = tokeizer.decode(output_ids[0], skip_special_tokes=True)
prit(respose)
点击空白处退出提示
评论