LSTMBidirectional
4 eligible runs
model
KernelBench level3 problem 38: LSTMBidirectional. The computation is the reference PyTorch module's forward pass; the output shape follows the module (mirrored as each implementation's source).
Fastest reported
49.7ms±0.02 · mean of 100 · 1.09× faster than baseline
torch.compile (inductor)PyTorch · MIT · python
Reported evidence · last observed 2026-03-05. Reported by source; not independently reproduced.
Current records
HardwareBest knownImplementationRuns
NVIDIA H100 · env 149.7 mstorch.compile (inductor)2NVIDIA H100 · env 256.9 msPyTorch eager2Not measured on B200 for this workload. Challenges →
Source-native comparison · GPU NVIDIA H100 · Workload batch_size = 10 · input_size = 128 · hidden_size = 256 · sequence_length = 512 · fp32 · Protocol KernelBench timing scripts · mean · 2 results · last observed 2026-03-05Record history →
Estimated floor 856 ns · record 58068.85× above itestimate, not evidence ›
DRAM 856 ns · bandwidth-bound on H100 SXM
every declared tensor crosses HBM exactly once (3,350 GB/s, H100 SXM datasheet)
no arithmetic formula for this family: bandwidth floor only
headroom-v1: a lower bound from declared tensors and datasheet peaks. A kernel can sit well above it for good reasons.
#
Implementation
Latency
vs #1
Trust
Observed
1torch.compile (inductor)PyTorch49.7ms±0.021.00×Reported · MIT · source2026-03-05stale
1.09× faster than the baseline. Measured exactly what you asked. Reported by source; not independently reproduced.
source mirroredMITno install recipeView source →Run detail →
254.1ms±0.391.09×Reported · MIT · source2026-03-05stale
Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.
source mirroredMITno install recipeView source →Run detail →
Implementations
Implementation
Runtime
Best latency
Evidence
Availability
Semantics
Inputs and outputs
input_1fp32 [batch_size, sequence_length, input_size]
input_2fp32 [12, batch_size, hidden_size]
input_3fp32 [12, batch_size, hidden_size]
outfloat [out]
Axes and behavior
outvariable
batch_sizevariable
input_sizevariable
hidden_sizevariable
sequence_lengthvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
sha25625834ac44242…
Sources: KernelBench baseline timings (2026-03-05) · MITlast observed 2026-03-05How records are decidedJSON