Instructions to use abanerjee10/DeepSeek-V2-Lite-INT8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use abanerjee10/DeepSeek-V2-Lite-INT8 with Transformers:

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

pipe = pipeline("text-generation", model="abanerjee10/DeepSeek-V2-Lite-INT8", 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("abanerjee10/DeepSeek-V2-Lite-INT8", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("abanerjee10/DeepSeek-V2-Lite-INT8", 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 abanerjee10/DeepSeek-V2-Lite-INT8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "abanerjee10/DeepSeek-V2-Lite-INT8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "abanerjee10/DeepSeek-V2-Lite-INT8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/abanerjee10/DeepSeek-V2-Lite-INT8

SGLang

How to use abanerjee10/DeepSeek-V2-Lite-INT8 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 "abanerjee10/DeepSeek-V2-Lite-INT8" \
    --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": "abanerjee10/DeepSeek-V2-Lite-INT8",
		"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 "abanerjee10/DeepSeek-V2-Lite-INT8" \
        --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": "abanerjee10/DeepSeek-V2-Lite-INT8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use abanerjee10/DeepSeek-V2-Lite-INT8 with Docker Model Runner:
```
docker model run hf.co/abanerjee10/DeepSeek-V2-Lite-INT8
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

DeepSeek-V2-Lite-INT8 (block-wise GPTQ, modelopt)

INT8 post-training quantization of deepseek-ai/DeepSeek-V2-Lite using the Runaraai block-wise GPTQ pipeline.

This checkpoint is Stage 7 modelopt format — packed int8 weights + per-channel float32 scales, with quantization_config for vLLM / SGLang / TensorRT-LLM. (Stage 5 alone stores GPTQ-tuned values as BF16 on disk.)

Quantization

Setting	Value
Method	GPTQ, block-wise
Storage	INT8 symmetric + per-channel `weight_scale` (modelopt)
Calibration	C4, 512 samples × 4096 tokens
Parallel Hessian	ON
Finished pack	2026-06-30 UTC

Quality (Δppl vs BF16 baseline, Stage 5/6 eval)

Dataset	BF16 ppl	INT8 ppl	Δppl
WikiText-2	7.224	7.225	+0.001
C4	11.409	11.403	−0.006

Usage (vLLM)

from vllm import LLM
llm = LLM("abanerjee10/DeepSeek-V2-Lite-INT8", trust_remote_code=True)

Weights on disk are int8, not BF16 — expect ~half the weight memory vs the BF16 source.

Provenance

Quantized by: Aranya @ Runara
Pipeline: Stages 5 (GPTQ) + 7 (modelopt pack)

Downloads last month: 6

Safetensors

Model size

16B params

Tensor type

BF16

F32

Model tree for abanerjee10/DeepSeek-V2-Lite-INT8

Base model

deepseek-ai/DeepSeek-V2-Lite

Quantized

(25)

this model