Downloads · 30 days
4
8% of all-time downloads
t5-small_dff2048_dmodel32_token-pattern-predictor_mixtral8x7bInstructv0.1_xsum
marsggbo/t5-small_dff2048_dmodel32_token-pattern-predictor_mixtral8x7bInstructv0.1_xsum
t5-small_dff2048_dmodel32_token-pattern-predictor_mixtral8x7bInstructv0.1_xsum is a machine learning model from marsggbo. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
All-time downloads
49
Public
Repo size
30.4 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.bin15.2 MB · 100%
At a glance
- Model type
- t5
- Access
- Public
- Created
- Oct 5, 2024
- Updated
- Oct 5, 2024
- SHA
- 9c2cf628
Base models
Model card
From the Hugging Face model README
- Download the repo
import os
import torch
from glob import glob
from transformers import AutoModelForSeq2SeqLM, AutoConfig
model_name = 'marsggbo/t5-small_dff2048_dmodel32_token-pattern-predictor_mixtral8x7bInstructv0.1_xsum'
# ignore the mismatched size, because lm_head was modified
model = AutoModelForSeq2SeqLM.from_pretrained(
model_name, ignore_mismatched_sizes=True, use_safetensors=False
)
- Build the model
home_path = os.path.expanduser('~')
num_classes = 32*8 # 32 layers, each with 8 experts
ckpt_path = f"{home_path}/.cache/huggingface/hub/*{model_name.split('/')[-1]}/snapshots/*/*bin"
ckpt_path = glob(ckpt_path)[0]
model_config = AutoConfig.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_config(config=model_config)
model.lm_head = torch.nn.Linear(model.config.hidden_size, num_classes, bias=False)
model.load_state_dict(torch.load(ckpt_path, map_location='cpu'), strict=True)
- Type
- t5
- Created
- Oct 5, 2024
- Updated
- Oct 5, 2024