Local AI Registry

Recipe

laguna-s-2-1-q4-k-m-dgx-spark-gb10-128gb-llama-cpp-tp1

laguna-s-2-1-q4-k-m-dgx-spark-gb10-128gb-llama-cpp-tp1
JSON APIFind compatible recipes

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

Record

Status
candidate
Source
localmaxxing
Engine
llama.cpp
Engine version
1 (04b2b72)
Accelerators
1
Tensor parallel
1
Context tokens
9,280
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

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

llama.cpp

Evidence only · candidate · reference

Candidate evidence — not a Run contract

Source
https://www.localmaxxing.com/en/runs/cmryoj31g0435o401cbxy7hen

Observed source tokens

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

  1. llama-server
  2. --model
  3. laguna-s-2.1-Q4_K_M.gguf
  4. --alias
  5. poolside/Laguna-S-2.1-GGUF
  6. --ctx-size
  7. 9280
  8. --parallel
  9. 1
  10. --batch-size
  11. 2048
  12. --ubatch-size
  13. 1024
  14. --n-gpu-layers
  15. auto
  16. --fit
  17. on
  18. --fit-target
  19. 24576
  20. --flash-attn
  21. on
  22. --cache-type-k
  23. q4_0
  24. --cache-type-v
  25. q4_0
  26. --cache-ram
  27. 0
  28. --cache-reuse
  29. 0
  30. --context-checkpoints
  31. 0
  32. --host
  33. 127.0.0.1
  34. --port
  35. 8000
  36. --metrics
FlagValue
--modellaguna-s-2.1-Q4_K_M.gguf
--aliaspoolside/Laguna-S-2.1-GGUF
--ctx-size9280
--parallel1
--batch-size2048
--ubatch-size1024
--n-gpu-layersauto
--fiton
--fit-target24576
--flash-attnon
--cache-type-kq4_0
--cache-type-vq4_0
--cache-ram0
--cache-reuse0
--context-checkpoints0
--host127.0.0.1
--port8000

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
19,280785.319.310,431.6observedlaguna-s-2-1-q4-k-m-dgx-spark-gb10-128gb-llama-cpp-tp1-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
laguna-s-2-1-q4-k-m-dgx-spark-gb10-128gb-llama-cpp-tp1
model instance id
poolside-laguna-s-2-1-gguf--q4-k-m
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
laguna-s-2-1-q4-k-m-dgx-spark-gb10-128gb-llama-cpp-tp1-sweep
status
candidate

capabilities

draft launch

accelerator backend
nvidia
arguments
-hf, poolside/Laguna-S-2.1-GGUF:Q4_K_M, --n-gpu-layers, 999, --host, 0.0.0.0, --port, 8080, -c, 9280
container port
8,080
environment · LLAMA CACHE
/root/.cache/huggingface
host port
8,080
image
ghcr.io/ggml-org/llama.cpp:server-cuda12-b10481@sha256:b2497f8834f5ecb4e38530f6bf2734b8e0be107f0f0857e259672d1cb85b71c2
ipc
host
kind
docker
shm size
16g
synthesized · generated at
2026-08-31T22:12:17Z
synthesized · image provenance
gemma-4-12b-q4-k-m-rtx-3060-12gb-llama-cpp-tp1
synthesized · template
llama-cpp-server-v1

mounts

read onlytarget
No/root/.cache/huggingface

engine

name
llama.cpp
version
1 (04b2b72)

serving

max concurrency
1
max context tokens
9,280
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
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

metadata

localmaxxing · backend
cuda
localmaxxing · hardware label
GB10 Grace Blackwell
localmaxxing · notes
Corrected ordinary-decoding GGUF result on one NVIDIA DGX Spark GB10. Seven distinct semantic prompts were measured once each with no warmup or prefix reuse. Six prompts contained 8,192 endpoint tokens and one contained 8,190; every request generated 1,024 tokens at concurrency one. Decode samples were 19.2, 19.3, 19.3, 19.3, 19.4, 19.4, and 19.4 tok/s; median 19.3. Fresh TTFT samples were 10,897.33, 10,419.64, 10,418.99, 10,462.31, 10,454.06, 10,376.58, and 10,431.60 ms; median 10,431.60 ms. Effective fresh-prefill samples were 751.7, 786.2, 786.3, 782.8, 783.6, 789.5, and 785.3 prompt tok/s; median 785.3. Poolside Laguna llama.cpp fork 04b2b72 served Q4_K_M revision e08e1fe855bb2d43f96ad78e24495283f3426c67, artifact SHA-256 7da520c5f44bc3c79d4eeebfd1151ba7114c5d7568e72a995638417093c5753f. Stable configuration: 9,280-token declared capacity, one slot, Q4_0 K/V cache, batch 2,048, microbatch 1,024, FlashAttention, all 49 layers offloaded, cache RAM disabled, prompt caching disabled, context checkpoints disabled, and no speculation. Minimum available RAM was 41.59 GiB and minimum free swap was 7.54 GiB; no guard event occurred. The old 18,155.28 ms result was real but used batch and microbatch 256, forcing 32 prompt-processing chunks, and left server prompt-cache/checkpoint behavior enabled. Prefill here is prompt tokens divided by fresh, client-observed TTFT, not kernel-only throughput. Rejected candidates and guard events are archived.
localmaxxing · observed command
llama-server --model laguna-s-2.1-Q4_K_M.gguf --alias poolside/Laguna-S-2.1-GGUF --ctx-size 9280 --parallel 1 --batch-size 2048 --ubatch-size 1024 --n-gpu-layers auto --fit on --fit-target 24576 --flash-attn on --cache-type-k q4_0 --cache-type-v q4_0 --cache-ram 0 --cache-reuse 0 --context-checkpoints 0 --host 127.0.0.1 --port 8000 --metrics
localmaxxing · run id
cmryoj31g0435o401cbxy7hen
localmaxxing · tokenized · arguments
llama-server, --model, laguna-s-2.1-Q4_K_M.gguf, --alias, poolside/Laguna-S-2.1-GGUF, --ctx-size, 9280, --parallel, 1, --batch-size, 2048, --ubatch-size, 1024, --n-gpu-layers, auto, --fit, on, --fit-target, 24576, --flash-attn, on, --cache-type-k, q4_0, --cache-type-v, q4_0, --cache-ram, 0, --cache-reuse, 0, --context-checkpoints, 0, --host, 127.0.0.1, --port, 8000, --metrics
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/cmryoj31g0435o401cbxy7hen ↗