Top k sampling from probs v129280

Current records

No published measurement for the selected workload.

Implementations

Implementation

Runtime

Best latency

Evidence

Availability

claude-opus-4-1-20250805 / cuda29eefbclaude-opus-4-1-20250805

cuda

—

Apache-2.0 · source

claude-opus-4-1-20250805 / tritonafd42dclaude-opus-4-1-20250805

triton

—

Apache-2.0 · source

gemini-2.5-pro / cuda8cde37gemini-2.5-pro

cuda

—

Apache-2.0 · source

gemini-2.5-pro / triton2c9c7dgemini-2.5-pro

triton

—

Apache-2.0 · source

gpt-5-2025-08-07 / cudad0c81egpt-5-2025-08-07

cuda

—

Apache-2.0 · source

gpt-5-2025-08-07_triton_e65787gpt-5-2025-08-07

triton

—

Apache-2.0 · source

gpt-o3 / cudae3d1f4gpt-o3

cuda

—

Apache-2.0 · source

gpt-o3 / triton75f9e9gpt-o3

triton

—

Apache-2.0 · source

Show all 9 implementations ›

flashinfer_wrapper_4ec4ec35FlashInfer-Bench baselines

python

—

Apache-2.0 · source

Semantics

Inputs and outputs

probsfp32 [batch_size, vocab_size]

top_kint32 [batch_size]

samplesint64 [batch_size]

Axes and behavior

batch_sizevariable

vocab_sizeconstant = 129280

determinismunspecified

constraintsNo mutation or aliasing

Identity

aliastop_k_sampling_from_probs_v129280modeldeepseek-v3 modeldeepseek-r1sha256cf78171e8ec4…

No source imports for this operation yet.How records are decided JSON