Skip to main content

Buildability / Receipt

Ragged Paged Attention: A High-Performance and Flexible LLM Inference Kernel for TPU

This public receipt window renders only fields present in the canonical receipt object, deterministic fixture receipt, or canonical evidence receipt. Missing compute, demo, hash, signature, approval, telemetry, and adoption fields stay explicit.

Public buildability page receipt window

Watch and verify: Ragged Paged Attention: A High-Performance and Flexible LLM Inference Kernel for TPU

/buildability/ragged-paged-attention-a-high-performance-and-flexible-llm-inference-kernel-for-tpu

Watchwatch

Subject: Ragged Paged Attention: A High-Performance and Flexible LLM Inference Kernel for TPU

Verdict

Watch

Verdict is Watch because viability or proof quality is intermediate and should be re-evaluated before execution.

Time to first demo

Insufficient data

No first-demo timestamp, owner estimate, or elapsed demo receipt is attached to this surface.

Compute envelope

Data

Flexible LLM Inference Kernel for TPU Jevin Jiang jevinjiang@google.com Ying Chen chenyin@google.com Blake Hechtman blakehechtman@google.com Fenghui Zhang fhzhang@google

Compute

operations, which enable the overlapping of TensorCore computation with DMAs. This overlap is the primary mechanism for extracting hardware parallelism on TPUs. While Pallas provides high-level abstractions—such as block specifications—to automatically generate multi-buffered pipelines

Inference

operations, which enable the overlapping of TensorCore computation with DMAs. This overlap is the primary mechanism for extracting hardware parallelism on TPUs. While Pallas provides high-level abstractions—such as block specifications—to automatically generate multi-buffered pipelines

Hardware

operations, which enable the overlapping of TensorCore computation with DMAs. This overlap is the primary mechanism for extracting hardware parallelism on TPUs. While Pallas provides high-level abstractions—such as block specifications—to automatically generate multi-buffered pipelines

Evidence ids

Receipt path

/buildability/ragged-paged-attention-a-high-performance-and-flexible-llm-inference-kernel-for-tpu

Paper ref

ragged-paged-attention-a-high-performance-and-flexible-llm-inference-kernel-for-tpu

arXiv id

2604.15464

Freshness

Generated at

2026-04-20T20:24:20.774Z

Evidence freshness

stale

Last verification

2026-04-20T20:24:20.774Z

Sources

3

References

0

Coverage

50%

Hash state

Lineage hash

23143088a8871265dd37ba91ed0f53c0a6ae42f9d27ad87877d20eda9689ea9c

Canonical opportunity-kernel lineage hash.

Signature state

External signature

unsigned_external

No founder, registry, pilot, or production-adoption signature is attached to this receipt.

Verification

not_verified

Verification is blocked until an external signature is provided.

Blockers

  • Missing: repo_url
  • Missing: references
  • Missing: proof_status
  • Unknown: proof verification has not been recorded yet

Some score or evidence fields are outside the preferred freshness window.

repo_url

references

Missing proof, requirement, signature, approval, adoption, or telemetry fields are blockers and must not be inferred.

Truth Boundary

External gate remains unresolved for live deployment claims.

Buildability surfaces only report computed viability and proof receipts. They do not claim live production usage, pilot outcomes, founder sign-off, public Brier calibration, judge divergence, or external adoption unless explicitly sourced.