vela3-fast

The default decision model. A CPU-friendly encoder with trained adapters for every built-in routing question, calibrated thresholds, and spans for PII and hallucination checks.

Property	Value
Repository	vllm-sr/vela3-fast, pinned at 7cbefd4b92b5
Family	Vela 3.0
Parameters	1.52B loaded
Backbone	UltraBERT
Precision (--dtype auto)	FP32 on every device
Weights	about 6.1 GB at FP32
Device memory	12 GiB or more
Reads	text

Question types

Type	Answers
choice	Yes
noul	Yes
score	No
set	Yes, asked as one noul per label
span	Yes, on the router span head

Built-in questions

The Vela 3.0 family was trained on the Router's built-in questions, so the Router asks each one in its trained form; the support matrix lists which.

A rule without a predicate takes this model's calibrated threshold:

Preset	Threshold
jailbreak (jailbreak)	0.88
domain (domain)	0.41
pii (pii)	0.18
fact check (fact_check)	0.9999
user feedback (feedback)	0.47
safety (harm)	0.52
modality (modality)	0.63

Hardware

Each time it loads, the model compares its answers with the reference answers recorded for its revision. They are recorded for:

• CPU (cpu)
• x86-64 CPU with AVX2 (BF16 emulated) (cpu:x86_64:avx2:bf16_emulated)

Where a hardware reference and its device class's are both recorded, the hardware one is checked.

On an AMD GPU or an NVIDIA GPU no reference is recorded, so the model serves there with its self-check reported as unverified.

Choose a model lists the validated hardware.

Quantization

The runtime loads no quantized packages of the Vela 3.0 family: it runs at its published precision.

Deploy it

Serve it on its own in Engine mode, to ask it directly:

CPU

vllm-sr serve vllm-sr/vela3-fast --engine --platform cpu

AMD GPU

vllm-sr serve vllm-sr/vela3-fast --engine --platform rocm

NVIDIA GPU

vllm-sr serve vllm-sr/vela3-fast --engine --platform cuda

Or make it the Router's decision model:

config.yaml

global:
  model_catalog:
    deployments:
      primary:
        provider: model_runtime
        artifact: vllm-sr/vela3-fast
        device: auto
    system:
      decision_model: {deployment: primary}

device: auto takes the first validated GPU with enough free memory, else the CPU; name rocm:0 or cuda:0 to pin a GPU.

To run it on another host and attach the Router to it:

vllm-srun serve vllm-sr/vela3-fast --device auto --host 0.0.0.0 --port 8100

Then set endpoint: http://<host>:8100 and model: vela3-fast on the deployment; see Attach to a runtime you run.

Learn more

• Model card on Hugging Face
• vela3-parity record
• vela3-router-signals record
• Vela 3.0: the family's sizes and how they differ