Vela 3.0
Vela 3.0 is a series of semantic routing models released through a collaboration between vLLM Semantic Router, KR Labs and AMD. Its core innovation combines routing decisions and span extraction into a single forward computation. The series provides open-weight models at six parameter scales ranging from 0.3B to 27B, intended for low-latency semantic dispatching at the request entry point.
Timeline (last 90 days)
Vela 3.0 was released by the vLLM Semantic Router, KR Labs, and the AMD team in collaboration.
Vela 3.0 merges routing decisions and segment extraction into a single call.
Vela 3.0 offers six open models with parameter counts ranging from 0.3B to 27B.
Completing routing and segment extraction in a single call can reduce the risk of inconsistent information transfer between two calls.
Models at the 0.3B level are suitable for resource-constrained edge environments or high-concurrency scenarios.
The official statement emphasizes more flexibility in what the router can detect (more flexibility in what the router can detect) and how small it can run (how small it can run).
If Vela 3.0 has been optimized for AMD platforms, that means this routing component can achieve usable inference performance in non-NVIDIA GPU environments.
Vela 3.0's official disclosures focus on two capability points, the merged call and the model lineage, and specific performance data and benchmark comparisons have not yet been elaborated.
All Facts (8)
Vela 3.0's official disclosures focus on two capability points, the merged call and the model lineage, and specific performance data and benchmark comparisons have not yet been elaborated.
50%UnverifiedVela 3.0 merges routing decisions and segment extraction into a single call.
50%UnverifiedVela 3.0 offers six open models with parameter counts ranging from 0.3B to 27B.
50%UnverifiedCompleting routing and segment extraction in a single call can reduce the risk of inconsistent information transfer between two calls.
50%UnverifiedModels at the 0.3B level are suitable for resource-constrained edge environments or high-concurrency scenarios.
50%UnverifiedThe official statement emphasizes more flexibility in what the router can detect (more flexibility in what the router can detect) and how small it can run (how small it can run).
50%UnverifiedIf Vela 3.0 has been optimized for AMD platforms, that means this routing component can achieve usable inference performance in non-NVIDIA GPU environments.
50%UnverifiedVela 3.0 was released by the vLLM Semantic Router, KR Labs, and the AMD team in collaboration.
50%