DeepSeek-V4.1-Flash

FP8 · vllm · 512K context · 4 × RTX PRO 6000 Blackwell in one machine · 384 GB

DeepSeek's V4.1 Flash, a large mixture-of-experts model.

150 tok/sdecode

– tok/sprefill

512Kcontext window

Earlier checkSep 26, 2026

Run it

Follow the steps below to download and run this recipe. Omarchy Local AI lists the recipes it can launch on your hardware. The server then answers on http://localhost:8000/v1.

1 Download the weights

hf download deepseek-ai/DeepSeek-V4.1-Flash \
  --revision fb2764a5cf321eaa5070ca8f9e892818f477c16d \
  --local-dir ~/models/DeepSeek-V4.1-Flash-fb2764a5

2 Start the server

docker run --rm \
  --gpus all \
  -p 8000:8000 \
  --ipc host \
  --security-opt seccomp=seccomp-default-plus-io_uring.json \
  --ulimit memlock=-1 \
  --ulimit stack=67108864 \
  --shm-size 32g \
  -e BLOCK_SIZE=256 \
  -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
  -e DCP_SIZE=1 \
  -e DSPARK_ADAPTIVE_VERIFICATION=1 \
  -e ENGRAM_TABLE_MEMORY=disk \
  -e GPU_MEMORY_UTILIZATION=0.95 \
  -e HF_HUB_OFFLINE=1 \
  -e HOST=0.0.0.0 \
  -e MAX_CUDAGRAPH_CAPTURE_SIZE=192 \
  -e MAX_MODEL_LEN=524288 \
  -e MAX_NUM_BATCHED_TOKENS=4096 \
  -e MAX_NUM_SEQS=32 \
  -e MAX_PARALLEL_PREFILLS=1 \
  -e MODEL_PATH=/models \
  -e OMP_NUM_THREADS=8 \
  -e PORT=8000 \
  -e PREFILL_COMPUTE_SHARE=0.4 \
  -e SERVED_MODEL_NAME=deepseek-v4.1-flash \
  -e SWA_BLOCK_SIZE=128 \
  -e TP_SIZE=4 \
  -e TRANSFORMERS_OFFLINE=1 \
  -e VLLM_USE_BREAKABLE_CUDAGRAPH=1 \
  -v ~/models/DeepSeek-V4.1-Flash-fb2764a5:/models:ro \
  --entrypoint /usr/local/bin/serve-ds41-jovian.sh \
  localinferencelab/vllm@sha256:f41ca8bb10bb3a125a50340d70d39ad4b7f5605f3fcb661bc992ed0bc4701a00 \
  --served-model-name deepseek-v4.1-flash \
  --kv-cache-memory-bytes 11274289152 \
  --speculative-config '{"method": "dspark", "num_speculative_tokens": 5, "draft_tensor_parallel_size": 4, "attention_backend": "B12X", "draft_sample_method": "greedy", "rejection_sample_method": "standard", "enable_adaptive_verification": true}' \
  --enable-prompt-tokens-details

What it passed

✓ loadLoads and serves within an hour

✓ chatAnswers a plain question and stops on its own

✓ reasoningThinks separately and gets 17 × 23 right

✓ toolsCalls a tool with the right arguments and uses the result

– contextRecalls a code buried in a prompt that fills 85% of the window

– speedDecodes at 15 tok/s or more

Passed the older check (loads and chats); a full six-check run is pending.

Details

Weights: deepseek-ai/DeepSeek-V4.1-Flash @ fb2764a5cf ›
Image: localinferencelab/vllm@sha256:f41ca8bb10bb3a125a50340d70d39ad4b7f5605f3fcb661bc992ed0bc4701a00
Launch: vllm-deepseek-v4.1-flash-fp8-512k-tp4 ›
Recipe file: registry/recipes/nvidia/rtx-pro-6000-blackwell-96gb/deepseek-v4.1-flash.vllm.512k.4x.json ›
Model released: Sep 10, 2026
Tested: earlier acceptance