Instructions to use Zigeng/DMax-Coder-16B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Zigeng/DMax-Coder-16B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Zigeng/DMax-Coder-16B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Zigeng/DMax-Coder-16B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Zigeng/DMax-Coder-16B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Zigeng/DMax-Coder-16B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Zigeng/DMax-Coder-16B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Zigeng/DMax-Coder-16B

SGLang

How to use Zigeng/DMax-Coder-16B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Zigeng/DMax-Coder-16B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Zigeng/DMax-Coder-16B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Zigeng/DMax-Coder-16B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Zigeng/DMax-Coder-16B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Zigeng/DMax-Coder-16B with Docker Model Runner:
```
docker model run hf.co/Zigeng/DMax-Coder-16B
```

DMax-Coder-16B / cmd.md

Zigeng

Upload folder using huggingface_hub

2c4aa79 verified 4 months ago

preview code

Raw

History Blame Contribute Delete

5 kB

	CUDA_VISIBLE_DEVICES=0 python demo.py

	PYTHONPATH=$(pwd)/VeOmni:$PYTHONPATH sh train.sh tasks/train_llada2_bd.py configs/sft/llada2_mini_bd_sft.yaml

	PYTHONPATH=$(pwd)/VeOmni:$PYTHONPATH sh train.sh tasks/train_llada2_bd_semi2.py configs/sft/llada2_mini_bd_sft_new.yaml

	PYTHONPATH=$(pwd)/VeOmni:$PYTHONPATH sh train.sh tasks/train_llada2_bd_hybrid.py configs/sft/llada2_mini_bd_sft_new.yaml

	sft2 batchsize=8
	sft3 batchsize=32
	sft4 batchsize=8

	python scripts/moe_convertor.py \
	--input-path /scratch/e0973935/model_weights/local_LLaDA2.1-mini \
	--output-path /scratch/e0973935/model_weights/local_LLaDA2.1-mini-merge \
	--mode merge

	python scripts/moe_convertor.py \
	--input-path /scratch/e0973935/model_weights/llada2.0_mini_sft_27 \
	--output-path /scratch/e0973935/model_weights/local_LLaDA2.0-mini-merge-cust \
	--mode merge

	python scripts/moe_convertor.py \
	--input-path /scratch/e0973935/dFactory/llada2_mini_bd_sft_outputs_mathabla/checkpoints/global_step_179430/hf_ckpt \
	--output-path /scratch/e0973935/model_weights/llada2.0_mini_abla \
	--mode split

	qsub -I \
	-P CFP03-SF-102 \
	-l select=1:ngpus=2 \
	-l walltime=1:40:00



	outputs3 online 0.6-1.0 lr=1e-6 bsz=8
	outputs4 online 0.6-1.0 lr=1e-5 bsz=64
	outputs5 online 0.4-0.8 lr=1e-6 bsz=8 allmath
	outputs6 online 0.4-0.8 lr=1e-6 bsz=8 allmath onpolicyremask
	outputs7 online 0.6-0.8 lr=1e-6 bsz=8 allmath
	outputs8 online 0.6-0.8 lr=2e-6 bsz=8 allmath
	outputs9 online 0.3-0.8 lr=1e-6 bsz=8 allmath ar-mask
	outputs10 online 0.0-1.0 lr=1e-6 bsz=8 allmath ar-mask
	outputs11 online 0.6-0.8 lr=1e-6 bsz=8 allmath+
	outputs12 online 0.6-0.8 lr=5e-7 bsz=8 allmath+
	outputs13 online 0.6-0.8 lr=1e-6 bsz=8 allmath+ block=64
	outputs14 online 0.6-0.8 lr=2e-6 bsz=8 allmath+
	outputs16 online 0.3-0.8 lr=1e-6 bsz=8 allmath+ ar-mask-8 label-mask
	outputs17 online 0.3-0.5 lr=1e-6 bsz=8 allmath ar-mask
	outputs18 online 0.6-0.8 lr=4e-6 bsz=8 allmath+
	outputs19 online 0.6-0.8 lr=1e-5 bsz=8 allmath+
	outputs20 online 0.6-0.8 lr=4e-6 bsz=8 allmath+ blockrand
	outputs21 online 0.7-0.7 lr=4e-6 bsz=8 allmath+

	outputs23 online 0.3-0.8 lr=2e-6 bsz=8 allmath+ ar-mask
	outputs24 online 0.3-0.8 lr=2e-6 bsz=8 allmath+ gumblemask
	outputs25 online 0.6-0.8 lr=2e-6 bsz=8 allmath+ gumblemask

	outputs26 online 0.6-0.8 lr=2e-6 bsz=8 allmath++
	outputs27 online 0.75 lr=2e-6 bsz=8 allmath++
	outputs28 online 0.6-0.8 lr=2e-6 bsz=8 allmath++ label-mask
	outputs29 online 0.75 lr=2e-6 bsz=8 allmath++ gumblemask thresh=0.5
	outputs30 online 0.75 lr=2e-6 bsz=8 allmath++ gumblemask thresh=0.3
	outputs31 online 0.5-0.8 lr=2e-6 bsz=8 allmath++ gumblemask thresh=0.3
	outputs32 online 0.75 lr=2e-6 bsz=8 allmath+ rkd
	outputs33 online 0.6-1.0 lr=2e-6 bsz=8 allmath+ rkd
	outputs34 online 0.75 lr=2e-6 bsz=8 allmath+ rkd w0.25
	outputs36 online 0.75 lr=2e-6 bsz=8 allmath+ ar-attention
	outputs37 online 0.75 lr=2e-6 bsz=8 allmath+ ar-attention-no-uni

	outputs38 online 0.75 lr=2e-6 bsz=8 allmath+ cont k=3
	outputs39 online 0.6-0.8 lr=2e-6 bsz=8 allmath+ cont k=3
	outputs40 online 0.75 lr=2e-6 bsz=8 allmath+ cont k=1
	outputs41 online 0.6-1.0 lr=2e-6 bsz=8 allmath+ cont k=1
	outputs42 online 0.75 lr=2e-6 bsz=8 allcode+

	outputs43 online 0.75 lr=2e-6 bsz=8 allmath+ cont-norm k=1
	outputs44 online 0.6-1.0 lr=2e-6 bsz=8 allmath+ cont-norm k=1
	outputs45 online 0.75 lr=2e-6 bsz=8 allmath+ cont-norm k=3
	outputs47 online 0.75 lr=2e-6 bsz=8 allmath+ cont-norm nomask k=3

	outputs48 online 0.6-0.9 lr=2e-6 bsz=8 allmath+
	outputs49 online 0.7-0.9 lr=2e-6 bsz=8 allcode+-
	outputs50 online 0.75 lr=2e-6 bsz=8 allcode+-
	outputs51 online 0.6-0.8 lr=2e-6 bsz=8 allcode+
	outputs52 online 0.75 lr=2e-6 bsz=8 allmath++ 27+epoch2


	outputs61 online 0.8 lr=2e-6 bsz=4 codefinal epoch=1

	export PYTHONPATH="/scratch/e0973935/dInfer/python:${PYTHONPATH}"
	python -c "import dinfer; print(dinfer.__file__)"


	amgr login

	hpc project

	CUDA_VISIBLE_DEVICES=0,1,2,3 python load.py

	deepspeed --include localhost:0 train_compress_ed2.py

	deepspeed --num_nodes=1 --num_gpus=8 train_compress3.py


	MAX_JOBS=4 pip install flash-attn --no-build-isolation

	MAX_JOBS=64 pip install flash_attn==2.8.3 --no-build-isolation

	pip install https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp310-cp310-linux_x86_64.whl

	scp -r /home/svu/e0973935/CompThinker /scratch/e0973935

	scp -r /scratch/e0973935/model_weights/custom_Qwen3-1.7B /scratch/e0950166

	scp -r /Users/yuruonan/Downloads/VITON_traindata/* yuruonan@deep40:/scratch/e0973935/model_weights/custom_Qwen3-1.7B

	scp -r e0973935@hopper.nus.edu.sg:/scratch/e0973935/model_weights/custom_Qwen3-1.7B /Users/zigeng/Downloads/nips26/models

	/Project_Storage/CFP-03/CFP03-SF-102

	scp -r /scratch/e0973935/model_weights/llada2.0_mini_sft_70 /Project_Storage/CFP-03/CFP03-SF-102

	scp -r /Project_Storage/CFP-03/CFP03-SF-102/llada2.0_mini_sft_70_5 /scratch/e0973935/model_weights/