Skip to content

Supported Models

The vLLM Spyre plugin relies on model code implemented by the Foundation Model Stack.

Configurations

The following models have been verified to run on vLLM Spyre with the listed configurations.

Decoder Models

Static Batching:

Model AIUs Prompt Length New Tokens Batch Size
Granite-3.3-8b 4 7168 1024 4

Continuous Batching:

Model AIUs Context Length Batch Size
Granite-3.3-8b 1 3072 16
Granite-3.3-8b 4 32768 32
Granite-3.3-8b (FP8) 1 3072 16
Granite-3.3-8b (FP8) 4 32768 32

Encoder Models

Model AIUs Context Length Batch Size
Granite-Embedding-125m (English) 1 512 1
Granite-Embedding-125m (English) 1 512 64
Granite-Embedding-278m (Multilingual) 1 512 1
Granite-Embedding-278m (Multilingual) 1 512 64
BAAI/BGE-Reranker (v2-m3) 1 8192 1
BAAI/BGE-Reranker (Large) 1 512 1
BAAI/BGE-Reranker (Large) 1 512 64
Multilingual-E5-large 1 512 64

Model Configuration

The Spyre engine uses a model registry to manage model-specific configurations. Model configurations are defined in vllm_spyre/config/model_configs.yaml and include:

  • Architecture patterns for model matching
  • Device-specific configurations (environment variables, GPU block overrides)
  • Supported runtime configurations (static batching warmup shapes, continuous batching parameters)

When a model is loaded, the registry automatically matches it to the appropriate configuration and applies model-specific settings.

Supported Configurations

The following configurations are supported for each model:


Configuration Validation

By default, the Spyre engine will log warnings if a requested model or configuration is not found in the registry. To enforce strict validation and fail if an unknown configuration is requested, set the environment variable:

export VLLM_SPYRE_REQUIRE_KNOWN_CONFIG=1

When this flag is enabled, the engine will raise a RuntimeError if:

  • The model cannot be matched to a known configuration
  • The requested runtime parameters are not in the supported configurations list

See the Configuration Guide for more details on model configuration.