Recipe
qwen3-8-27b-nvfp4-rtx-5060-ti-16gb-vllm-tp2
qwen3-8-27b-nvfp4-rtx-5060-ti-16gb-vllm-tp2Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- vllm
- Accelerators
- 2
- Tensor parallel
- 2
- Context tokens
- 2,048
- Max concurrency
- 1
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/Qwen/Qwen3.8-27B- Repository
- Qwen/Qwen3.8-27B
- 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
- Endpoint
http://192.168.0.45:30000- Served model
gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090-LMHead4
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 2,048 | 395.5 | 63.2 | 189.6 | observed | qwen3-8-27b-nvfp4-rtx-5060-ti-16gb-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-5060-ti-16gb
- id
- qwen3-8-27b-nvfp4-rtx-5060-ti-16gb-vllm-tp2
- model instance id
- qwen-qwen3-8-27b--nvfp4
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- qwen3-8-27b-nvfp4-rtx-5060-ti-16gb-vllm-tp2-sweep
- status
- candidate
capabilities
draft launch
- accelerator backend
- nvidia
- arguments
- --model, Qwen/Qwen3.8-27B, --tensor-parallel-size, 2, --host, 0.0.0.0, --port, 8000, --max-model-len, 2048
- 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 only | target |
|---|---|
| No | /root/.cache/huggingface |
engine
- name
- vllm
serving
- max concurrency
- 1
- max context tokens
- 2,048
- 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 5060 Ti
- localmaxxing · notes
- vLLM V2 on dual RTX 5060 Ti 16GB (Unraid 7, 6.18.38). MTP speculative decoding: num_speculative_tokens=3, TP=2, NVFP4 quant, max-model-len 131072, KV-cache nvfp4 4.5GB budget, prefix-caching + chunked-prefill enabled, cudagraph_capture_sizes [1,2,4,8,16]. This is the warm-cache representative run (4 consecutive back-to-back benchmark iterations, prefix cache hot): 63.2-63.7 tok/s decode, ~186-195 ms TTFT. MTP draft acceptance during this window: avg 54.8% (41.4-71.1%, p50 52.6%, 18 samples) — well above the ~30% breakeven for 3 draft tokens, i.e. MTP spec-decode is net-positive here. For contrast, a cold-cache benchmark window earlier in the session measured 38.2 tok/s at ~35% acceptance; full-lifetime real-world Hermes traffic averages ~50.6% token-weighted acceptance (2.69M samples, per-position pos1=0.647 pos2=0.495 pos3=0.393). Engine: vLLM, quant NVFP4.
- localmaxxing · observed command
- # Remote endpoint: http://192.168.0.45:30000 servedModel: gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090-LMHead4
- localmaxxing · run id
- cmtdklchh005clm01b8ev1mrk
- localmaxxing · tokenized · endpoint
- 192.168.0.45 ↗
- localmaxxing · tokenized · fidelity
- faithful
- localmaxxing · tokenized · served model name
- gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090-LMHead4
provenance
- captured at
- 2026-08-30T09:10:02Z
sources
| captured at | kind | url |
|---|---|---|
| 2026-08-30T09:10:02Z | normalized-recipe | www.localmaxxing.com/en/runs/cmtdklchh005clm01b8ev1mrk ↗ |