VOOZH about

URL: https://huggingface.co/lamm-mit/LlaDA-8B-EditFlow-graph-v510

⇱ lamm-mit/LlaDA-8B-EditFlow-graph-v510 · Hugging Face


Edit Flow model: Graph-native reasoning

python3 -m venv ~/venvs/dllm
source ~/venvs/dllm/bin/activate
python -m pip install -U pip setuptools wheel

git clone https://github.com/lamm-mit/DiscoverydLLM.git
cd DiscoverydLLM
pip install -e .

export PYTHONPATH="$PYTHONPATH:/home/ubuntu/dllm"

Sampling

python examples/editflow/sample.py \
 --model_name_or_path lamm-mit/LlaDA-8B-EditFlow-graph-v510 \
 --tau 0.02 --mask_length 128 --seed 7070 \
 --prompt ‘Define materiomics.’ 

Advanced sampling

python examples/editflow/sample_advanced.py \
 --model_name_or_path "lamm-mit/LlaDA-8B-EditFlow-graph-v510" \
 --prompt "Define materiomics and relate with flowers." \
 --mask_length 128 \
 --max_refinement_iterations 3 \
 --confidence_method margin \
 --temperature_start 0.8 \
 --temperature_end 0.3  \
 --scheduler linear  \
 --verbose  --tau 0.002

With CFG:

python examples/editflow/sample_advanced.py \
 --model_name_or_path "lamm-mit/LlaDA-8B-EditFlow-graph-v510" \
 --prompt "Define materiomics." \
 --mask_length 128 \
 --cfg_scale 1.5 \
 --scheduler cosine
Downloads last month
21
Safetensors
Model size
9B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for lamm-mit/LlaDA-8B-EditFlow-graph-v510

Finetuned
(3)
this model

Dataset used to train lamm-mit/LlaDA-8B-EditFlow-graph-v510

Collection including lamm-mit/LlaDA-8B-EditFlow-graph-v510