Local AI Registry

Recipe

qwen3-6-35b-a3b-nvfp4-dgx-spark-gb10-128gb-vllm-tp1-m6i8dge6

qwen3-6-35b-a3b-nvfp4-dgx-spark-gb10-128gb-vllm-tp1-m6i8dge6
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
0.19.2rc1.dev213+g9558f4390.d20260426
Accelerators
1
Tensor parallel
1
Context tokens
262,144
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/Qwen/Qwen3.6-35B-A3B
Repository
Qwen/Qwen3.6-35B-A3B
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/cmon8glp0000el204m6i8dge6

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. /models/Qwen3.6-35B-A3B-NVFP4
  4. --served-model-name
  5. qwen36-35b-a3b
  6. --host
  7. 0.0.0.0
  8. --port
  9. 8000
  10. --tensor-parallel-size
  11. 1
  12. --gpu-memory-utilization
  13. 0.50
  14. --max-model-len
  15. 262144
  16. --max-num-batched-tokens
  17. 4096
  18. --max-num-seqs
  19. 1
  20. --trust-remote-code
  21. --load-format
  22. fastsafetensors
  23. --attention-backend
  24. flash_attn
  25. --enable-prefix-caching
  26. --speculative-config
  27. {"method":"dflash","num_speculative_tokens":15,"model":"/models/Qwen3.6-35B-A3B-DFlash"}
  28. --seed
  29. 0
FlagValue
--served-model-nameqwen36-35b-a3b
--host0.0.0.0
--port8000
--tensor-parallel-size1
--gpu-memory-utilization0.50
--max-model-len262144
--max-num-batched-tokens4096
--max-num-seqs1
--load-formatfastsafetensors
--attention-backendflash_attn
--speculative-config{"method":"dflash","num_speculative_tokens":15,"model":"/models/Qwen3.6-35B-A3B-DFlash"}
--seed0

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
1262,144—102.1153observedqwen3-6-35b-a3b-nvfp4-dgx-spark-gb10-128gb-vllm-tp1-m6i8dge6-sweep

Remaining fields

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

hardware count
1
hardware id
dgx-spark-gb10-128gb
id
qwen3-6-35b-a3b-nvfp4-dgx-spark-gb10-128gb-vllm-tp1-m6i8dge6
model instance id
qwen-qwen3-6-35b-a3b--nvfp4
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-6-35b-a3b-nvfp4-dgx-spark-gb10-128gb-vllm-tp1-m6i8dge6-sweep
status
candidate

capabilities

draft launch

accelerator backend
nvidia
arguments
serve, --model, Qwen/Qwen3.6-35B-A3B, --tensor-parallel-size, 1, --host, 0.0.0.0, --port, 8000, --max-model-len, 262144
container port
8,000
entrypoint
vllm
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
0.19.2rc1.dev213+g9558f4390.d20260426

serving

max concurrency
1
max context tokens
262,144
tensor parallel
1
Provenance & metadata (3)

facts

capabilities.chat · provenance · captured at
2026-08-30T09:26:09Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
capabilities.reasoning · provenance · captured at
2026-08-30T09:26:09Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
capabilities.tools · provenance · captured at
2026-08-30T09:26:09Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
capabilities.vision · provenance · captured at
2026-08-30T09:26:09Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
draft launch.entrypoint · provenance · captured at
2026-09-01T01:41:26Z
draft launch.entrypoint · reason linux-amd64-container-config-entrypoint
draft launch.entrypoint · state known
engine.graph mode · provenance · captured at
2026-08-30T09:26:09Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
serving.kv cache tokens · provenance · captured at
2026-08-30T09:26:09Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown
serving.max concurrency · provenance · captured at
2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-derived-from-source-evidence
serving.max concurrency · state known

metadata

localmaxxing · backend
cuda
localmaxxing · batch size
1
localmaxxing · hardware label
GB10 Grace Blackwell
localmaxxing · notes
Companion submission to cmoi65drc000bl104fehrcah8 (RedHatAI/Qwen3.6-35B-A3B-NVFP4). Same measurement, posted against root model Qwen/Qwen3.6-35B-A3B with quantization=NVFP4 explicit so it surfaces on the canonical model page. Qwen3.6-35B-A3B-NVFP4 + DFlash speculative decoding on a single DGX Spark (GB10 Blackwell, 128 GiB unified). vLLM 0.19.2rc1 + the DFlash off-by-one patch (without it, drafter reads pre-embedding state and acceptance collapses ~80%→3%). Sherlock prose, pp=128, tg=128, c=1, n=29 warm. tokSOut = 102.05 t/s MEDIAN (std 48.74 — DFlash acceptance varies per token). 2.37× over NVFP4 AR baseline (42.98 t/s). peakVramGb=87 at gpu-memory-utilization=0.50 (right-sized from 118 GB at 0.92, no decode regression). KV pool 294K tokens covers the 262K context with 1.12× headroom. MoE-A3B is bound by expert weights not KV — FP8 KV + alt attention backends gave <5% AR gain. Recipe + Dockerfile + raw JSONs: github.com/my-other-github-account/spark-bench-reproducers/tree/main/qwen36-35b-a3b-dflash-spark (commit b8776d3, reproduces clean-from-scratch). By @banana_baeee.
localmaxxing · observed command
vllm serve /models/Qwen3.6-35B-A3B-NVFP4 --served-model-name qwen36-35b-a3b --host 0.0.0.0 --port 8000 --tensor-parallel-size 1 --gpu-memory-utilization 0.50 --max-model-len 262144 --max-num-batched-tokens 4096 --max-num-seqs 1 --trust-remote-code --load-format fastsafetensors --attention-backend flash_attn --enable-prefix-caching --speculative-config '{"method":"dflash","num_speculative_tokens":15,"model":"/models/Qwen3.6-35B-A3B-DFlash"}' --seed 0
localmaxxing · run id
cmon8glp0000el204m6i8dge6
localmaxxing · tokenized · arguments
vllm, serve, /models/Qwen3.6-35B-A3B-NVFP4, --served-model-name, qwen36-35b-a3b, --host, 0.0.0.0, --port, 8000, --tensor-parallel-size, 1, --gpu-memory-utilization, 0.50, --max-model-len, 262144, --max-num-batched-tokens, 4096, --max-num-seqs, 1, --trust-remote-code, --load-format, fastsafetensors, --attention-backend, flash_attn, --enable-prefix-caching, --speculative-config, {"method":"dflash","num_speculative_tokens":15,"model":"/models/Qwen3.6-35B-A3B-DFlash"}, --seed, 0
localmaxxing · tokenized · fidelity
faithful

provenance

captured at
2026-08-30T09:26:09Z

sources

captured atkindurl
2026-08-30T09:26:09Znormalized-recipewww.localmaxxing.com/en/runs/cmon8glp0000el204m6i8dge6 ↗