Instructions to use BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B with Transformers:

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

pipe = pipeline("text-generation", model="BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B")
model = AutoModelForCausalLM.from_pretrained("BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B

SGLang

How to use BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B 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 "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B with Docker Model Runner:
```
docker model run hf.co/BlivionIaG/DeepSeek-V4-Flash-0731-Int4-FP8-REAP-216B
```

VLLM serving recipe.

by void009 - opened 13 days ago

Discussion

void009

13 days ago

Hi. Trying to get this running on 6x3090 with vllm, but apparently this particular tensor config used in this quant is not supported, at least that is what the ai agent found. You mention a simple vllm launch command. What vllm version can run this, or how did you test it. I dont want llamacpp since it is not good with concurrent requests afaik.

BlivionIaG

Owner 9 days ago

Hi @void009 , this was part of something i was testing as a vllm fork for rdna gpus. I paused the experiment since i was waiting for one part i needed for extending from 4 to 8 Radeon pro v620. I tested a working mxfp4 kernel for rdna2 in the mean time, for ampere i think you have to adapt something that would use marlin w4a16 or similar.
I will probably revisit but if you need something your setup must support the FP8 part that was rearranged in FP8 compressed and the scales changed from u8m0 to FP32.

void009

9 days ago

ds4 flash old news already. glm 5.3 flash new friend.

BlivionIaG

Owner 9 days ago

While its true, its demanding more resources to run. I have yet to see wether Qwen3.8 flash next is better or not or still v4 flash.

Currently focused on making AWQ INT4 fast on rdna2. Next target in parallel is a kernel for exl3 qtip 3bit.

This quant was an experiment for me to check many things, now i know i can quantize with only one gpu at 32GB its just slow.

BlivionIaG changed discussion status to closed 9 days ago

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment