Recipe

laguna-s-2-1-compressed-tensors-int4-group-32-w4a16-intel-arc-pro-b70-32gb-vllm-tp4

laguna-s-2-1-compressed-tensors-int4-group-32-w4a16-intel-arc-pro-b70-32gb-vllm-tp4

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

Record

Status
candidate
Source
localmaxxing
Engine
vllm
Engine version
0.1.dev1172+g4a6fd8747.xpu; local source ef334233deabeaeedb607056a2db1c90edb3887c; XPU kernels 4772f727590c51b72add79350b913d098cf67872
Accelerators
4
Tensor parallel
4
Context tokens
8,192
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/poolside/Laguna-S-2.1-INT4
Repository
poolside/Laguna-S-2.1-INT4
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/cmrzrd4tf001ipa013xpx4kid

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/laguna-s-2.1-xpu-b70/tools/run_laguna_m8_metadata_formal_crossover_leg.sh
  2. candidate
  3. B1
  4. RUN_DIR

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
18,19294.95,960.3observedlaguna-s-2-1-compressed-tensors-int4-group-32-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep

Remaining fields

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

hardware count
4
hardware id
intel-arc-pro-b70-32gb
id
laguna-s-2-1-compressed-tensors-int4-group-32-w4a16-intel-arc-pro-b70-32gb-vllm-tp4
model instance id
poolside-laguna-s-2-1-int4--compressed-tensors-int4-group-32-w4a16
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
laguna-s-2-1-compressed-tensors-int4-group-32-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-sweep
status
candidate

capabilities

engine

name
vllm
version
0.1.dev1172+g4a6fd8747.xpu; local source ef334233deabeaeedb607056a2db1c90edb3887c; XPU kernels 4772f727590c51b72add79350b913d098cf67872

serving

max concurrency
1
max context tokens
8,192
tensor parallel
4
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
xpu
localmaxxing · hardware label
Intel Arc Pro B70
localmaxxing · notes
Verified strict graph-vs-graph A1-B1-B2-A2 persistent exact-attention metadata record candidate. The submitted 94.92003934159611 tok/s is the lower of two exact fresh candidate starts (other 95.06654837534518), +2.991% over approved record cmrzjb7i906x4o401egrnm05m. Metadata-on beat both adjacent metadata-off graph controls, 13/13 rows in each pair, while saving 0.911 and 1.648 ms per aggregate DFlash target cycle. All 52 requests were bitwise token-identical to canonical q1 and cached_tokens=0; long-next and rollover passed on every leg. First lazy graph capture remained inside every leg's first measured cold request. No cache, prior-history acceleration, response reuse, warm-up generation, retry, qualification timing, or aggregate replica throughput.
localmaxxing · observed command
experiments/laguna-s-2.1-xpu-b70/tools/run_laguna_m8_metadata_formal_crossover_leg.sh candidate B1 RUN_DIR
localmaxxing · run id
cmrzrd4tf001ipa013xpx4kid
localmaxxing · tokenized · arguments
experiments/laguna-s-2.1-xpu-b70/tools/run_laguna_m8_metadata_formal_crossover_leg.sh, candidate, B1, RUN_DIR
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/cmrzrd4tf001ipa013xpx4kid