Local AI Registry

Recipe

deepseek-v4-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp2

deepseek-v4-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp2
JSON APIFind compatible recipes

Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.

Record

Status
candidate
Source
localmaxxing
Engine
vllm
Engine version
0.11.2.dev279-b12x
Accelerators
2
Tensor parallel
2
Context tokens
262,000
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash
Repository
deepseek-ai/DeepSeek-V4-Flash
Status
known
Link type
Exact Hub repository

Public Hugging Face repository confirmed by the Hub API.

Candidate: useful compatibility or speed evidence. The registry does not offer Run until promotion requirements are met.

Observed configuration

vllm

Evidence only · candidate · reference

Candidate evidence — not a Run contract

Source
https://www.localmaxxing.com/en/runs/cmr0cglf10090ld01banhfvi0

Observed source tokens

Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.

  1. vllm
  2. serve
  3. deepseek-ai/DeepSeek-V4-Flash-DSpark
  4. --served-model-name
  5. deepseek/v4flash
  6. --kv-cache-dtype
  7. fp8
  8. --block-size
  9. 256
  10. --tensor-parallel-size
  11. 2
  12. --moe-backend
  13. b12x
  14. --linear-backend
  15. b12x
  16. --gpu-memory-utilization
  17. 0.93
  18. --max-model-len
  19. 262000
  20. --max-num-seqs
  21. 4
  22. --async-scheduling
  23. --no-scheduler-reserve-full-isl
  24. --max-num-batched-tokens
  25. 4096
  26. --max-cudagraph-capture-size
  27. 256
  28. --attention-backend
  29. B12X_MLA_SPARSE
  30. --enable-chunked-prefill
  31. --enable-prefix-caching
  32. --compilation-config
  33. {cudagraph_mode:FULL_AND_PIECEWISE,custom_ops:[all]}
  34. --speculative-config
  35. {method:dspark,num_speculative_tokens:5}
FlagValue
--served-model-namedeepseek/v4flash
--kv-cache-dtypefp8
--block-size256
--tensor-parallel-size2
--moe-backendb12x
--linear-backendb12x
--gpu-memory-utilization0.93
--max-model-len262000
--max-num-seqs4
--max-num-batched-tokens4096
--max-cudagraph-capture-size256
--attention-backendB12X_MLA_SPARSE
--compilation-config{cudagraph_mode:FULL_AND_PIECEWISE,custom_ops:[all]}
--speculative-config{method:dspark,num_speculative_tokens:5}

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
1262,0006,058261.84,830observeddeepseek-v4-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp2-sweep

Remaining fields

Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.

hardware count
2
hardware id
rtx-pro-6000-blackwell-96gb
id
deepseek-v4-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp2
model instance id
deepseek-ai-deepseek-v4-flash--fp8
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
deepseek-v4-flash-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp2-sweep
status
candidate

capabilities

draft launch

accelerator backend
nvidia
arguments
--model, deepseek-ai/DeepSeek-V4-Flash, --tensor-parallel-size, 2, --host, 0.0.0.0, --port, 8000, --max-model-len, 262000
container port
8,000
host port
8,000
image
vllm/vllm-openai@sha256:0a51ea5b4ae2dc5d81890e5173f54203d2a3ae0cfffe51b8fd2afd4391bfd967
ipc
host
kind
docker
shm size
16g
synthesized · generated at
2026-08-31T22:12:17Z
synthesized · image provenance
deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1
synthesized · template
vllm-openai-v1

mounts

read onlytarget
No/root/.cache/huggingface

engine

name
vllm
version
0.11.2.dev279-b12x

serving

max concurrency
1
max context tokens
262,000
tensor parallel
2
Provenance & metadata (3)

facts

capabilities.chat · provenance · captured at
2026-08-30T09:10:02Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
capabilities.reasoning · provenance · captured at
2026-08-30T09:10:02Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
capabilities.tools · provenance · captured at
2026-08-30T09:10:02Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
capabilities.vision · provenance · captured at
2026-08-30T09:10:02Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
engine.graph mode · provenance · captured at
2026-08-30T09:10:02Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
serving.kv cache tokens · provenance · captured at
2026-08-30T09:10:02Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown

metadata

localmaxxing · backend
cuda
localmaxxing · hardware label
RTX PRO 6000 Blackwell
localmaxxing · notes
DSpark spec decode (num_spec=5, accept_len 4.48) on the b12x SM120 stack accelerating DeepSeek-V4-Flash, 2x RTX PRO 6000 TP2. Code gen, single stream, greedy, 262k config. Checkpoint: DeepSeek-V4-Flash-DSpark. Beats the 181 MTP run. Repo: github.com/jarrelscy/dspark-b12x
localmaxxing · observed command
vllm serve deepseek-ai/DeepSeek-V4-Flash-DSpark --served-model-name deepseek/v4flash --kv-cache-dtype fp8 --block-size 256 --tensor-parallel-size 2 --moe-backend b12x --linear-backend b12x --gpu-memory-utilization 0.93 --max-model-len 262000 --max-num-seqs 4 --async-scheduling --no-scheduler-reserve-full-isl --max-num-batched-tokens 4096 --max-cudagraph-capture-size 256 --attention-backend B12X_MLA_SPARSE --enable-chunked-prefill --enable-prefix-caching --compilation-config {cudagraph_mode:FULL_AND_PIECEWISE,custom_ops:[all]} --speculative-config {method:dspark,num_speculative_tokens:5}
localmaxxing · run id
cmr0cglf10090ld01banhfvi0
localmaxxing · tokenized · arguments
vllm, serve, deepseek-ai/DeepSeek-V4-Flash-DSpark, --served-model-name, deepseek/v4flash, --kv-cache-dtype, fp8, --block-size, 256, --tensor-parallel-size, 2, --moe-backend, b12x, --linear-backend, b12x, --gpu-memory-utilization, 0.93, --max-model-len, 262000, --max-num-seqs, 4, --async-scheduling, --no-scheduler-reserve-full-isl, --max-num-batched-tokens, 4096, --max-cudagraph-capture-size, 256, --attention-backend, B12X_MLA_SPARSE, --enable-chunked-prefill, --enable-prefix-caching, --compilation-config, {cudagraph_mode:FULL_AND_PIECEWISE,custom_ops:[all]}, --speculative-config, {method:dspark,num_speculative_tokens:5}
localmaxxing · tokenized · fidelity
faithful

provenance

captured at
2026-08-30T09:10:02Z

sources

captured atkindurl
2026-08-30T09:10:02Znormalized-recipewww.localmaxxing.com/en/runs/cmr0cglf10090ld01banhfvi0 ↗