szl-holdings/szl-serve

Canonical SZL governed serve recipe: llama.cpp/Ollama CPU honest path for Khipu GGUF, schema validation outside the weights, energy MEASURED or UNAVAILABLE, GPU ROADMAP. Not a vLLM/TGI/TRT-LLM copy. Doctrine v11, Apache-2.0.

0

stars

8

commits

Python

primary language

Sep 8, 2026

updated

apache-2-0
governed-ai
llama-cpp
receipts

README

szl-serve

One OpenAI-shaped chat call. One pinned weight. One receipt.

KERNEL-owned canonical governed-inference recipe + validator + energy receipt for SZL Holdings.

This is not a vendor engine. It is the silhouette of 2026 serving (vLLM / SGLang / TGI / TensorRT-LLM / llama.cpp / Dynamo) — without copying those engines, and without fabricating tokens/s or joules.

Homepage: a-11-oy.com · Owner: Stephen P. Lutar Jr. / SZL Holdings · Doctrine v11 LOCKED 749/14/163 · Λ = Conjecture 1 (advisory, never a theorem) · Apache-2.0


What this is NOT

  • Not vLLM. Not TGI. Not TensorRT-LLM. Not SGLang. Not Dynamo. Those names are the 2026 silhouette. This repo does not vendor-copy them. GPU path is ROADMAP until MEASURED on named hardware.
  • Not a tokens/s leaderboard. A MEASURED sample is raw completion tokens + elapsed_ms. No derived tok/s rating.
  • Not proven trust. A hash-chained receipt proves integrity of that record. It does not prove the model, the lab, or Λ.
  • Not a takeover of the Hugging Face Space. FORGE owns szl-holdings/szl-forge / SZLHOLDINGS/szl-model-inference-lab. This KERNEL repo pins that MEASURED lab as the live CPU path. It does not invent a second runtime.
  • Not szl-forge-lab. That Space is SNAPSHOT / READ-ONLY, not a trainer, not a serve runtime. Do not point szl-serve at forge-lab.
  • Not Ask & Act as a live control plane. The a11oy operator tab does not gate, launch, or control this recipe.
  • Not energy-attested-runs as MEASURED serve energy. That corpus is 8/8 SIMULATED, not an NVML joule delta and not a live serve path.

Honesty plate: CARD.md.


ATELIER lock

Serve studio = SZLHOLDINGS/szl-model-inference-lab only. GPU remains ROADMAP.

SurfaceClass
SZLHOLDINGS/szl-model-inference-labLIVE / MEASURED serve studio (this pin)
SZLHOLDINGS/szl-forge-labSNAPSHOT, not a trainer, not a serve target
GPU / vLLM / TGI / TensorRT-LLM / SGLang / DynamoROADMAP
SZLHOLDINGS/energy-attested-runs8/8 SIMULATED — not this recipe's CPU energy evidence
Ask & Actnot a live control plane

LIVE CPU (MEASURED) — pin this lab, do not invent a second runtime

ATELIER serve studio. Not szl-forge-lab.

Surfacehttps://szlholdings-szl-model-inference-lab.hf.space
Spacehttps://huggingface.co/spaces/SZLHOLDINGS/szl-model-inference-lab
FORGE sourcehttps://github.com/szl-holdings/szl-forge/tree/main/spaces/szl-model-inference-lab
APIOpenAI subset POST /v1/chat/completions
ModelSZLHOLDINGS/SZL-Khipu-1.5B-GGUF@67d60ec577730747055491640cfb91fc4a4b5d25
GGUFSZL-Khipu-1.5B-Q4_K_M.gguf · SHA-256 13c1a1993063e1dff92f7413ccf48eaca6d48efc8801ae9af35961ae3396623a · 986047904 bytes
Sample 2026-08-2821 completion tokens · elapsed_ms 2053 · execution_record UNSIGNED
Energyjoules=null · label UNAVAILABLE (CPU lab; MEASURED only from a real NVML delta)
SLABest-effort public demo. No provider SLA. Dummy bearer not-a-secret only — never a real token.
curl https://szlholdings-szl-model-inference-lab.hf.space/v1/chat/completions \
  -H "content-type: application/json" \
  -H "authorization: Bearer not-a-secret" \
  -d '{"model":"SZLHOLDINGS/SZL-Khipu-1.5B-GGUF@67d60ec577730747055491640cfb91fc4a4b5d25","messages":[{"role":"user","content":"Explain one limit of cryptographic receipts."}],"max_tokens":24,"stream":false}'

Contract: /.well-known/szl-inference-contract.json. Identity: /api/v1/identity.


Airgap twin — same pin, local llama.cpp / Ollama

Local llama.cpp and Ollama run the same Q4_K_M bytes. They are the airgap twin of the Space, not a clone of vLLM.

llama-cli    -hf SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M
llama-server -hf SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M
ollama run hf.co/SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M

Print the recipe (no GGUF download):

python -m szl_serve recipe

Controller — schema outside the weights

Model output is proposal-only. Copy of khipu.schema.json is pinned from the GGUF repo (SHA-256 b95f9927366dae7c5d36cfb7de6e229eb605524318ab642a6aa2292a212170d0).

  • Hallucinated citedNodeIdsREJECT. Never silently repaired into a green plan.
  • brainBinding.status stays NOT_RESOLVED.
  • Real model output provenance is MODEL_PROPOSED, not SYNTHETIC.
  • Invalid JSON / cross-field contract failure → fail closed (REJECT).
python -m szl_serve validate tests/fixtures/plan_model_proposed.json
python -m szl_serve validate tests/fixtures/plan_hallucinated_cited.json   # exits 1

Energy + receipt

Each wrap uses szl-energy-attest semantics (import if available, else a tiny honest local meter):

EnergyWhen
MEASUREDReal NVML energy-counter delta. joules is that number.
UNAVAILABLENo GPU / no fresh NVML delta (the CPU lab and this CI). joules=null.

Never invent a joule. Hash-chain a serve receipt: request hash, output hash, model pin, energy label. Receipt is UNSIGNED unless a key is actually present. Do not fake DSSE.


MEASURED vs ROADMAP vs UNAVAILABLE

SurfaceLabel
FORGE Space CPU OpenAI subset + GGUF pin + 2026-08-28 sampleMEASURED
llama.cpp / Ollama commands for the same pinLIVE recipe (airgap twin)
Energy on CPUUNAVAILABLE
GPU / vLLM / TGI / TensorRT-LLM / SGLang / DynamoROADMAP
GPU attentionROADMAP — no fused/paged-attention LIVE claim; no tokens/s invented
SZLHOLDINGS/szl-forge-labSNAPSHOT — not a trainer; do not point this recipe at it
SZLHOLDINGS/energy-attested-runs8/8 SIMULATED — not MEASURED NVML; not a live serve path
Ask & Actnot a live control plane
Chaski, ReceiptAgent weightsROADMAP stubs on this config surface — not fake LIVE

Λ is Conjecture 1. It does not become a theorem because a chat call returned 21 tokens.


Tests

Unit tests use fixture plan JSON. No network. Do not download the ~1 GB GGUF in CI.

pip install -e ".[test]"
pytest -q

Layout

LICENSE                 Apache-2.0, Copyright 2026 SZL Holdings
README.md               this Series A face
CARD.md                 honesty plate
schemas/khipu.schema.json
szl_serve/{__init__,schema,energy,recipe,cli}.py
tests/fixtures/*.json

Governed AI you can pin — not a leaderboard you can fake.

a-11-oy.com · Khipu GGUF · MEASURED lab (FORGE)

SLSA: L1 honest · L2 attested · L3 roadmap. Λ = Conjecture 1. Trust ceiling 0.97 — never 100%.

Contributors

szl-holdings/szl-serve

Canonical SZL governed serve recipe: llama.cpp/Ollama CPU honest path for Khipu GGUF, schema validation outside the weights, energy MEASURED or UNAVAILABLE, GPU ROADMAP. Not a vLLM/TGI/TRT-LLM copy. Doctrine v11, Apache-2.0.

0

stars

8

commits

Python

primary language

Sep 8, 2026

updated

apache-2-0
governed-ai
llama-cpp
receipts

README

szl-serve

One OpenAI-shaped chat call. One pinned weight. One receipt.

KERNEL-owned canonical governed-inference recipe + validator + energy receipt for SZL Holdings.

This is not a vendor engine. It is the silhouette of 2026 serving (vLLM / SGLang / TGI / TensorRT-LLM / llama.cpp / Dynamo) — without copying those engines, and without fabricating tokens/s or joules.

Homepage: a-11-oy.com · Owner: Stephen P. Lutar Jr. / SZL Holdings · Doctrine v11 LOCKED 749/14/163 · Λ = Conjecture 1 (advisory, never a theorem) · Apache-2.0


What this is NOT

  • Not vLLM. Not TGI. Not TensorRT-LLM. Not SGLang. Not Dynamo. Those names are the 2026 silhouette. This repo does not vendor-copy them. GPU path is ROADMAP until MEASURED on named hardware.
  • Not a tokens/s leaderboard. A MEASURED sample is raw completion tokens + elapsed_ms. No derived tok/s rating.
  • Not proven trust. A hash-chained receipt proves integrity of that record. It does not prove the model, the lab, or Λ.
  • Not a takeover of the Hugging Face Space. FORGE owns szl-holdings/szl-forge / SZLHOLDINGS/szl-model-inference-lab. This KERNEL repo pins that MEASURED lab as the live CPU path. It does not invent a second runtime.
  • Not szl-forge-lab. That Space is SNAPSHOT / READ-ONLY, not a trainer, not a serve runtime. Do not point szl-serve at forge-lab.
  • Not Ask & Act as a live control plane. The a11oy operator tab does not gate, launch, or control this recipe.
  • Not energy-attested-runs as MEASURED serve energy. That corpus is 8/8 SIMULATED, not an NVML joule delta and not a live serve path.

Honesty plate: CARD.md.


ATELIER lock

Serve studio = SZLHOLDINGS/szl-model-inference-lab only. GPU remains ROADMAP.

SurfaceClass
SZLHOLDINGS/szl-model-inference-labLIVE / MEASURED serve studio (this pin)
SZLHOLDINGS/szl-forge-labSNAPSHOT, not a trainer, not a serve target
GPU / vLLM / TGI / TensorRT-LLM / SGLang / DynamoROADMAP
SZLHOLDINGS/energy-attested-runs8/8 SIMULATED — not this recipe's CPU energy evidence
Ask & Actnot a live control plane

LIVE CPU (MEASURED) — pin this lab, do not invent a second runtime

ATELIER serve studio. Not szl-forge-lab.

Surfacehttps://szlholdings-szl-model-inference-lab.hf.space
Spacehttps://huggingface.co/spaces/SZLHOLDINGS/szl-model-inference-lab
FORGE sourcehttps://github.com/szl-holdings/szl-forge/tree/main/spaces/szl-model-inference-lab
APIOpenAI subset POST /v1/chat/completions
ModelSZLHOLDINGS/SZL-Khipu-1.5B-GGUF@67d60ec577730747055491640cfb91fc4a4b5d25
GGUFSZL-Khipu-1.5B-Q4_K_M.gguf · SHA-256 13c1a1993063e1dff92f7413ccf48eaca6d48efc8801ae9af35961ae3396623a · 986047904 bytes
Sample 2026-08-2821 completion tokens · elapsed_ms 2053 · execution_record UNSIGNED
Energyjoules=null · label UNAVAILABLE (CPU lab; MEASURED only from a real NVML delta)
SLABest-effort public demo. No provider SLA. Dummy bearer not-a-secret only — never a real token.
curl https://szlholdings-szl-model-inference-lab.hf.space/v1/chat/completions \
  -H "content-type: application/json" \
  -H "authorization: Bearer not-a-secret" \
  -d '{"model":"SZLHOLDINGS/SZL-Khipu-1.5B-GGUF@67d60ec577730747055491640cfb91fc4a4b5d25","messages":[{"role":"user","content":"Explain one limit of cryptographic receipts."}],"max_tokens":24,"stream":false}'

Contract: /.well-known/szl-inference-contract.json. Identity: /api/v1/identity.


Airgap twin — same pin, local llama.cpp / Ollama

Local llama.cpp and Ollama run the same Q4_K_M bytes. They are the airgap twin of the Space, not a clone of vLLM.

llama-cli    -hf SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M
llama-server -hf SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M
ollama run hf.co/SZLHOLDINGS/SZL-Khipu-1.5B-GGUF:Q4_K_M

Print the recipe (no GGUF download):

python -m szl_serve recipe

Controller — schema outside the weights

Model output is proposal-only. Copy of khipu.schema.json is pinned from the GGUF repo (SHA-256 b95f9927366dae7c5d36cfb7de6e229eb605524318ab642a6aa2292a212170d0).

  • Hallucinated citedNodeIdsREJECT. Never silently repaired into a green plan.
  • brainBinding.status stays NOT_RESOLVED.
  • Real model output provenance is MODEL_PROPOSED, not SYNTHETIC.
  • Invalid JSON / cross-field contract failure → fail closed (REJECT).
python -m szl_serve validate tests/fixtures/plan_model_proposed.json
python -m szl_serve validate tests/fixtures/plan_hallucinated_cited.json   # exits 1

Energy + receipt

Each wrap uses szl-energy-attest semantics (import if available, else a tiny honest local meter):

EnergyWhen
MEASUREDReal NVML energy-counter delta. joules is that number.
UNAVAILABLENo GPU / no fresh NVML delta (the CPU lab and this CI). joules=null.

Never invent a joule. Hash-chain a serve receipt: request hash, output hash, model pin, energy label. Receipt is UNSIGNED unless a key is actually present. Do not fake DSSE.


MEASURED vs ROADMAP vs UNAVAILABLE

SurfaceLabel
FORGE Space CPU OpenAI subset + GGUF pin + 2026-08-28 sampleMEASURED
llama.cpp / Ollama commands for the same pinLIVE recipe (airgap twin)
Energy on CPUUNAVAILABLE
GPU / vLLM / TGI / TensorRT-LLM / SGLang / DynamoROADMAP
GPU attentionROADMAP — no fused/paged-attention LIVE claim; no tokens/s invented
SZLHOLDINGS/szl-forge-labSNAPSHOT — not a trainer; do not point this recipe at it
SZLHOLDINGS/energy-attested-runs8/8 SIMULATED — not MEASURED NVML; not a live serve path
Ask & Actnot a live control plane
Chaski, ReceiptAgent weightsROADMAP stubs on this config surface — not fake LIVE

Λ is Conjecture 1. It does not become a theorem because a chat call returned 21 tokens.


Tests

Unit tests use fixture plan JSON. No network. Do not download the ~1 GB GGUF in CI.

pip install -e ".[test]"
pytest -q

Layout

LICENSE                 Apache-2.0, Copyright 2026 SZL Holdings
README.md               this Series A face
CARD.md                 honesty plate
schemas/khipu.schema.json
szl_serve/{__init__,schema,energy,recipe,cli}.py
tests/fixtures/*.json

Governed AI you can pin — not a leaderboard you can fake.

a-11-oy.com · Khipu GGUF · MEASURED lab (FORGE)

SLSA: L1 honest · L2 attested · L3 roadmap. Λ = Conjecture 1. Trust ceiling 0.97 — never 100%.

Contributors

Languages

Python

100.0%