Instructions to use 0xSero/DeepSeek-V4-Flash-162B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use 0xSero/DeepSeek-V4-Flash-162B with Transformers:

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

pipe = pipeline("text-generation", model="0xSero/DeepSeek-V4-Flash-162B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("0xSero/DeepSeek-V4-Flash-162B")
model = AutoModelForCausalLM.from_pretrained("0xSero/DeepSeek-V4-Flash-162B")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use 0xSero/DeepSeek-V4-Flash-162B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "0xSero/DeepSeek-V4-Flash-162B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "0xSero/DeepSeek-V4-Flash-162B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/0xSero/DeepSeek-V4-Flash-162B

SGLang

How to use 0xSero/DeepSeek-V4-Flash-162B 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 "0xSero/DeepSeek-V4-Flash-162B" \
    --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": "0xSero/DeepSeek-V4-Flash-162B",
		"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 "0xSero/DeepSeek-V4-Flash-162B" \
        --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": "0xSero/DeepSeek-V4-Flash-162B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use 0xSero/DeepSeek-V4-Flash-162B with Docker Model Runner:
```
docker model run hf.co/0xSero/DeepSeek-V4-Flash-162B
```

0xSero commited on May 27

Commit

7f2e7a3

verified ·

1 Parent(s): 818e1ab

Update DGX Spark 200K serving recipe

Browse files

Files changed (1) hide show

README.md +8 -8

README.md CHANGED Viewed

@@ -15,11 +15,12 @@ tags:
 base_model: deepseek-ai/DeepSeek-V4-Flash
 ---
-# Deepseek-V4-Flash-162B-REAP
-This is the 162B / K144 REAP-pruned DeepSeek V4 Flash model. The validated single-DGX Spark serving recipe is maintained here:
-- GitHub: https://github.com/0xSero/deepseek-v4-flash-spark-200k
 - Docker registry target: `ghcr.io/0xsero/deepseek-v4-flash-spark-vllm:cutlass451-g27`
 - Validated local Docker image: `vllm-node-dsv4-cutlass451:latest` / `sha256:5df60ebb9c10dfb86d5946cae8244adfe65a7fd405401bd542ecf22d5c497a4a`
 - Model repo used by the recipe: `0xSero/DeepSeek-V4-Flash-162B-codex-K144-REAP`
@@ -30,7 +31,7 @@ This is the 162B / K144 REAP-pruned DeepSeek V4 Flash model. The validated singl
 Run this on the DGX Spark. `HF_TOKEN` is only required if the model repo is private or not already cached on the machine.
 ```bash
-HF_TOKEN=... bash -lc 'set -euo pipefail; cd /home/sero/spark; rm -rf deepseek-v4-flash-spark-200k; git clone https://github.com/0xSero/deepseek-v4-flash-spark-200k.git; cd deepseek-v4-flash-spark-200k; ./install.sh --profile k144-nospec-200k --launch'
 ```
 Do not commit tokens into the repo or a model card. Pass them only through the environment for the one command above.
@@ -42,7 +43,7 @@ The profile lives at `configs/k144-nospec-200k.env` in the GitHub repo.
 ```bash
 MODEL_REPO=0xSero/DeepSeek-V4-Flash-162B-codex-K144-REAP
 MODEL_REVISION=d663e8fb16809f6619000648b187b257249ed824
-SERVED_MODEL_NAME=deepseek-v4-flash-k144-g27-cutlass451
 CONTEXT_LENGTH=200000
 KV_CACHE_MEMORY_BYTES=14G
 MAX_NUM_BATCHED_TOKENS=8192
@@ -50,14 +51,13 @@ MAX_NUM_SEQS=1
 GPU_MEMORY_UTILIZATION=0.88
 WATCHDOG_MIN_AVAILABLE_KB=8388608
 KV_CACHE_DTYPE=fp8
-ENFORCE_EAGER=0
-THINKING=false
 SPECULATIVE_CONFIG=
 VLLM_ENABLE_DEEPSEEK_V4_SPARSE_MLA_WARMUP=0
 VLLM_TRITON_MLA_SPARSE_ALLOW_CUDAGRAPH=1
 ```
-The launcher enables DeepSeek V4 tokenizer, reasoning parser, tool-call parser, prefix caching, FP8 KV, and CUDA graphs. Do not add `--enforce-eager`; this profile was validated with CUDA graph capture enabled.
 ## Docker runtime

 base_model: deepseek-ai/DeepSeek-V4-Flash
 ---
+# DeepSeek-V4-Flash-Spark-Mini
+This is the 162B / K144 REAP-pruned DeepSeek V4 Flash model, served as `DeepSeek-V4-Flash-Spark-Mini`. The validated single-DGX Spark serving recipe is maintained here:
+- One-command Spark wrapper: https://github.com/0xSero/deepseek-spark
+- Runtime module: https://github.com/0xSero/deepseek-v4-flash-spark-200k
 - Docker registry target: `ghcr.io/0xsero/deepseek-v4-flash-spark-vllm:cutlass451-g27`
 - Validated local Docker image: `vllm-node-dsv4-cutlass451:latest` / `sha256:5df60ebb9c10dfb86d5946cae8244adfe65a7fd405401bd542ecf22d5c497a4a`
 - Model repo used by the recipe: `0xSero/DeepSeek-V4-Flash-162B-codex-K144-REAP`
 Run this on the DGX Spark. `HF_TOKEN` is only required if the model repo is private or not already cached on the machine.
 ```bash
+HF_TOKEN=... bash -lc 'set -euo pipefail; cd /home/sero/spark; rm -rf deepseek-spark; git clone https://github.com/0xSero/deepseek-spark.git; cd deepseek-spark; ./setup.sh full k144'
 ```
 Do not commit tokens into the repo or a model card. Pass them only through the environment for the one command above.
 ```bash
 MODEL_REPO=0xSero/DeepSeek-V4-Flash-162B-codex-K144-REAP
 MODEL_REVISION=d663e8fb16809f6619000648b187b257249ed824
+SERVED_MODEL_NAME=DeepSeek-V4-Flash-Spark-Mini
 CONTEXT_LENGTH=200000
 KV_CACHE_MEMORY_BYTES=14G
 MAX_NUM_BATCHED_TOKENS=8192
 GPU_MEMORY_UTILIZATION=0.88
 WATCHDOG_MIN_AVAILABLE_KB=8388608
 KV_CACHE_DTYPE=fp8
+THINKING=true
 SPECULATIVE_CONFIG=
 VLLM_ENABLE_DEEPSEEK_V4_SPARSE_MLA_WARMUP=0
 VLLM_TRITON_MLA_SPARSE_ALLOW_CUDAGRAPH=1
 ```
+The launcher enables DeepSeek V4 tokenizer, reasoning parser, tool-call parser, prefix caching, FP8 KV, and CUDA graph capture.
 ## Docker runtime