mstar.engine.resources.attn.ragged.flashinfer

Cacheless varlen attention through FlashInfer’s ragged prefill wrapper.

Classes

FlashInferRaggedBlockCausalManager(device, ...)	Block-causal self-attention: every label in the step is its own layout, each segment attending block-causally within itself.
FlashInferRaggedCrossManager(device, dtype, ...)
FlashInferRaggedManager(device, dtype, config)

class mstar.engine.resources.attn.ragged.flashinfer.FlashInferRaggedBlockCausalManager(device, dtype, config, block_size)

Bases: _FlashInferRaggedManagerBase, RaggedBlockCausalAttnManager

Block-causal self-attention: every label in the step is its own layout, each segment attending block-causally within itself.

Parameters:

• device (device)

• dtype (dtype)

• config (RaggedAttentionConfig)

• block_size (int)

plan(step, ctx)

ret is immutable and opaque to runner; only gives to ctx.plan_results

Parameters:

• step (AttentionStep)

• ctx (StepContext)

wrapper_class

alias of RaggedBlockCausalWrapper

class mstar.engine.resources.attn.ragged.flashinfer.FlashInferRaggedCrossManager(device, dtype, config)

Bases: _FlashInferRaggedManagerBase, _RaggedCrossManagerBase

Parameters:

• device (device)

• dtype (dtype)

• config (RaggedAttentionConfig)

wrapper_class

alias of FlashInferRaggedCrossPrefillWrapper

class mstar.engine.resources.attn.ragged.flashinfer.FlashInferRaggedManager(device, dtype, config)

Bases: _FlashInferRaggedManagerBase, _RaggedSelfManagerBase

Parameters:

• device (device)

• dtype (dtype)

• config (RaggedAttentionConfig)

wrapper_class

alias of FlashInferRaggedPrefillWrapper