Instructions to use daeunj/DeepSeek-V2-Lite-FP8-GPTQ with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use daeunj/DeepSeek-V2-Lite-FP8-GPTQ with Transformers:

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

pipe = pipeline("text-generation", model="daeunj/DeepSeek-V2-Lite-FP8-GPTQ", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("daeunj/DeepSeek-V2-Lite-FP8-GPTQ", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("daeunj/DeepSeek-V2-Lite-FP8-GPTQ", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use daeunj/DeepSeek-V2-Lite-FP8-GPTQ with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "daeunj/DeepSeek-V2-Lite-FP8-GPTQ"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "daeunj/DeepSeek-V2-Lite-FP8-GPTQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/daeunj/DeepSeek-V2-Lite-FP8-GPTQ

SGLang

How to use daeunj/DeepSeek-V2-Lite-FP8-GPTQ 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 "daeunj/DeepSeek-V2-Lite-FP8-GPTQ" \
    --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": "daeunj/DeepSeek-V2-Lite-FP8-GPTQ",
		"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 "daeunj/DeepSeek-V2-Lite-FP8-GPTQ" \
        --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": "daeunj/DeepSeek-V2-Lite-FP8-GPTQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use daeunj/DeepSeek-V2-Lite-FP8-GPTQ with Docker Model Runner:
```
docker model run hf.co/daeunj/DeepSeek-V2-Lite-FP8-GPTQ
```

DeepSeek V2 Lite FP8 GPTQ

This checkpoint was produced with block-wise GPTQ using FP8 E4M3 weights.

Typical pipeline:

bash scripts/download_model.sh --model_name deepseek-v2-lite
python tests/stage5_quantize_model.py --model_path models/DeepSeek-V2-Lite --quant_format fp8 --seq_len 4096
python tests/stage7_save_modelopt.py --model_path models/DeepSeek-V2-Lite-FP8 --output_dir models/DeepSeek-V2-Lite-FP8-modelopt --stage5_results results/stage5_DeepSeek-V2-Lite_fp8_quantize.json

Evaluate quality against the BF16 baseline before deployment:

python tests/stage4_baseline_perplexity.py --model_path models/DeepSeek-V2-Lite --seq_len 4096
python tests/stage6_eval_perplexity.py --model_path models/DeepSeek-V2-Lite-FP8 --quant_format fp8 --seq_len 4096

Downloads last month: 26

Safetensors

Model size

16B params

Tensor type

BF16

F8_E4M3

F32

Model tree for daeunj/DeepSeek-V2-Lite-FP8-GPTQ

Base model

deepseek-ai/DeepSeek-V2-Lite

Quantized

(25)

this model