Recipe

qwen3-6-27b-int4-autoround-w4a16-int8-head-int4-draft-intel-arc-pro-b70-32gb-vllm-tp1-u760xh0t

qwen3-6-27b-int4-autoround-w4a16-int8-head-int4-draft-intel-arc-pro-b70-32gb-vllm-tp1-u760xh0t

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

Record

Status
candidate
Source
localmaxxing
Engine
vllm
Engine version
0.20.2rc1.dev13+g9557d9108.d20260620 local XPU patch stack
Graph
piecewise
Accelerators
1
Tensor parallel
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/webhie/Qwen3.6-27B-int4-AutoRound
Repository
webhie/Qwen3.6-27B-int4-AutoRound
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/cmr9atqb800msqr01u760xh0t

Observed source tokens

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

  1. experiments/qwen36-27b-autoround-int4-b70/scripts/run-vllm-candidate.sh

Environment

VariableValue
COMPILATION_CONFIG{"cudagraph_mode":"PIECEWISE","max_cudagraph_capture_size":8}
GPU_INDEX0
MAX_MODEL_LEN2048
MAX_NUM_BATCHED_TOKENS1024
MODEL_DIR/mnt/fast-ai/llm-cache/hf/hub/models--webhie--Qwen3.6-27B-int4-AutoRound/snapshots/f5750c90b3776db658594df5fe8051098226dd8e
NUM_SPECULATIVE_TOKENS3
PORT19642
QWEN36_27B_ENABLE_MTP1
QWEN36_27B_ENABLE_XPU_GRAPH1
VLLM_XPU_DRAFT_LM_HEAD_INT41
VLLM_XPU_DRAFT_LM_HEAD_INT4_GROUP_SIZE128
VLLM_XPU_DRAFT_LM_HEAD_INT4_SCALE_DTYPEbf16
VLLM_XPU_GDN_NONSPEC_POSTPROCESS_ACCEPTED_STATE0
VLLM_XPU_GDN_PROMOTE_ACCEPTED_SPEC_STATE1
VLLM_XPU_GDN_REPLAYSSM_COMMIT_IN_FORWARD1
VLLM_XPU_GDN_REPLAYSSM_SLOT_MGMT_TORCH_FALLBACK1
VLLM_XPU_GDN_REPLAYSSM_SPEC1
VLLM_XPU_GDN_REPLAYSSM_SPEC_CACHE_LEN8
VLLM_XPU_GDN_REPLAYSSM_STAGE_CONV_TORCH_FALLBACK0
VLLM_XPU_GDN_REPLAYSSM_TORCH_FALLBACK0
VLLM_XPU_LM_HEAD_INT81
VLLM_XPU_LM_HEAD_INT8_SCALE_DTYPEbf16

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
2,04868.2479.1observedqwen3-6-27b-int4-autoround-w4a16-int8-head-int4-draft-intel-arc-pro-b70-32gb-vllm-tp1-u760xh0t-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
intel-arc-pro-b70-32gb
id
qwen3-6-27b-int4-autoround-w4a16-int8-head-int4-draft-intel-arc-pro-b70-32gb-vllm-tp1-u760xh0t
model instance id
webhie-qwen3-6-27b-int4-autoround--int4-autoround-w4a16-int8-head-int4-draft
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-6-27b-int4-autoround-w4a16-int8-head-int4-draft-intel-arc-pro-b70-32gb-vllm-tp1-u760xh0t-sweep
status
candidate

capabilities

engine

graph mode
piecewise
name
vllm
version
0.20.2rc1.dev13+g9557d9108.d20260620 local XPU patch stack

serving

tensor parallel
1
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
2026-09-01T01:32:29Z
engine.graph mode · reason explicit-cudagraph-mode-in-observed-command
engine.graph mode · state known
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
serving.max concurrency · provenance · captured at
2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown
serving.max context tokens · provenance · captured at
2026-09-01T10:20:39Z
serving.max context tokens · reason context-limit-not-evidenced
serving.max context tokens · state unknown

metadata

localmaxxing · backend
xpu
localmaxxing · batch size
1
localmaxxing · hardware label
Intel Arc Pro B70
localmaxxing · notes
Current best valid Qwen3.6 27B INT4 AutoRound fresh-response row on one Intel Arc Pro B70 for the webhie AutoRound + runtime INT8 target LM-head BF16-scale + runtime INT4 draft LM-head ReplaySSM recipe. Same recipe as the approved 67.519 tok/s row, now fully quality-confirmed at 68.236 tok/s; treat as a small variance-sensitive current-confirm record, not a new mechanism. Fixed realistic suite, each prompt once cold, cached_tokens=0 on every request, no prompt/KV/cache/history/ngram/response reuse, repeat64 quality matched baseline.
localmaxxing · observed command
MODEL_DIR=/mnt/fast-ai/llm-cache/hf/hub/models--webhie--Qwen3.6-27B-int4-AutoRound/snapshots/f5750c90b3776db658594df5fe8051098226dd8e GPU_INDEX=0 PORT=19642 MAX_MODEL_LEN=2048 MAX_NUM_BATCHED_TOKENS=1024 QWEN36_27B_ENABLE_MTP=1 NUM_SPECULATIVE_TOKENS=3 QWEN36_27B_ENABLE_XPU_GRAPH=1 VLLM_XPU_GDN_PROMOTE_ACCEPTED_SPEC_STATE=1 VLLM_XPU_GDN_NONSPEC_POSTPROCESS_ACCEPTED_STATE=0 VLLM_XPU_GDN_REPLAYSSM_SPEC=1 VLLM_XPU_GDN_REPLAYSSM_SPEC_CACHE_LEN=8 VLLM_XPU_GDN_REPLAYSSM_TORCH_FALLBACK=0 VLLM_XPU_GDN_REPLAYSSM_STAGE_CONV_TORCH_FALLBACK=0 VLLM_XPU_GDN_REPLAYSSM_COMMIT_IN_FORWARD=1 VLLM_XPU_GDN_REPLAYSSM_SLOT_MGMT_TORCH_FALLBACK=1 VLLM_XPU_LM_HEAD_INT8=1 VLLM_XPU_LM_HEAD_INT8_SCALE_DTYPE=bf16 VLLM_XPU_DRAFT_LM_HEAD_INT4=1 VLLM_XPU_DRAFT_LM_HEAD_INT4_GROUP_SIZE=128 VLLM_XPU_DRAFT_LM_HEAD_INT4_SCALE_DTYPE=bf16 COMPILATION_CONFIG='{"cudagraph_mode":"PIECEWISE","max_cudagraph_capture_size":8}' experiments/qwen36-27b-autoround-int4-b70/scripts/run-vllm-candidate.sh
localmaxxing · run id
cmr9atqb800msqr01u760xh0t
localmaxxing · tokenized · arguments
experiments/qwen36-27b-autoround-int4-b70/scripts/run-vllm-candidate.sh
localmaxxing · tokenized · environment · COMPILATION CONFIG
{"cudagraph_mode":"PIECEWISE","max_cudagraph_capture_size":8}
localmaxxing · tokenized · environment · GPU INDEX
0
localmaxxing · tokenized · environment · MAX MODEL LEN
2048
localmaxxing · tokenized · environment · MAX NUM BATCHED TOKENS
1024
localmaxxing · tokenized · environment · NUM SPECULATIVE TOKENS
3
localmaxxing · tokenized · environment · PORT
19642
localmaxxing · tokenized · environment · QWEN36 27B ENABLE MTP
1
localmaxxing · tokenized · environment · QWEN36 27B ENABLE XPU GRAPH
1
localmaxxing · tokenized · environment · VLLM XPU DRAFT LM HEAD INT4
1
localmaxxing · tokenized · environment · VLLM XPU DRAFT LM HEAD INT4 GROUP SIZE
128
localmaxxing · tokenized · environment · VLLM XPU DRAFT LM HEAD INT4 SCALE DTYPE
bf16
localmaxxing · tokenized · environment · VLLM XPU GDN NONSPEC POSTPROCESS ACCEPTED STATE
0
localmaxxing · tokenized · environment · VLLM XPU GDN PROMOTE ACCEPTED SPEC STATE
1
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM COMMIT IN FORWARD
1
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM SLOT MGMT TORCH FALLBACK
1
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM SPEC
1
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM SPEC CACHE LEN
8
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM STAGE CONV TORCH FALLBACK
0
localmaxxing · tokenized · environment · VLLM XPU GDN REPLAYSSM TORCH FALLBACK
0
localmaxxing · tokenized · environment · VLLM XPU LM HEAD INT8
1
localmaxxing · tokenized · environment · VLLM XPU LM HEAD INT8 SCALE DTYPE
bf16
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/cmr9atqb800msqr01u760xh0t