rewirebio.iobenchmarks
Result

Inference throughput of Evo 2 7B zero-shot likelihood score on Runpod H100 on BRCA1 saturation genome editing SNVs, all SNVs (evo2-cloud-operations W1)

2.91 Inference throughput

Methods

Tested configuration
Evo 2 7B zero-shot likelihood score on Runpod H100
Protocol
Evo 2 7B BRCA1 variant scoring throughput on one GPU (evo2-cloud-operations W1)
Dataset subset
BRCA1 saturation genome editing SNVs, all SNVs (evo2-cloud-operations W1)
Procedure
Workload W1, batch size 1
Evaluation
Evo 2 7B scoring throughput on Runpod H100
Coverage
Not reported scored / Not reported eligible
Uncertainty
Not reported by the source: The file gives no interval at this pointer.
Evidence
Rewire evaluation · source checkedevo2-cloud-operations recorded run b75c67aa: run receipt, Runpod H100 (P3, attempt-1) · evidence/runs/recorded-b75c67aa/platforms/P3/attempt-1/receipt-main.json, JSON pointer /results/w1/sequences_per_second

A source-checked result verifies the numerical transcription, not every model or protocol detail. Evaluation metadata: source checked. Source checked does not mean independently reproduced.

Reproduction

Split
none; every variant of the subset is scored
Adaptation
zero-shot inference
Scoring implementation
src/evo2ops/workloads.py; the recorded run used revision e93491a (evidence/runs/recorded-b75c67aa/provenance.json), an ancestor of the pinned commit

No execution recipe has been verified for this exact configuration and evaluation. A benchmark's general instructions may use different inputs, splits or model settings.

Reproducing this published result requires matching its model configuration, data, split and scorer. Source checking or a successful smoke test does not establish score reproduction.

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

1 evidence row matching the loaded filters

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance
Reported result
2.9075
Individual claims
evo2-cloud-operations recorded run b75c67aa: run receipt, Runpod H100 (P3, attempt-1)

Original source ↗

evidence/runs/recorded-b75c67aa/platforms/P3/attempt-1/receipt-main.json, JSON pointer /results/w1/sequences_per_second

Version: f8ae11c9a1c45505487c978b4cb7e064c4e59905
Retrieved: 2026-11-21T14:00:00Z

source checked

["execution-evidence-review","source-hash-verification"] · 2026-11-21

rewire run

Audit details

Read from the run receipt of this platform, which is the full W1 workload. One run; no interval. Source checked, not marked reproduced: no run recipe names the receipt.

Field: attributes.printed_value

Source artifact SHA-256: 4bfe70c67740043bef2f383beade8bdb09696f88a4c673c1194520b354e7ff30

Hash scope: complete bytes of the file at the pinned commit

Inspected artifact

Extraction artifact SHA-256: 4bfe70c67740043bef2f383beade8bdb09696f88a4c673c1194520b354e7ff30

Extraction artifact

Sources and history

Release 2026-11-21-c8bea25e5abb · Record review: source checked

1 source record and release historyDownload this release (gzip)
Technical metadata and extraction receipts

Stable ID: rewire-local-20261121-evo2ops-result-evo2-7b-runpod-h100-scoring-throughput

areas
dna-genomes
printed value
2.9075
numeric value
2.9075
metric
inference-throughput
metric direction
higher
unit
sample-per-second
unit detail
8,192 bp sequences per second
hardware
NVIDIA H100 80GB HBM3
limitations
One run on one host. Throughput depends on the host as well as the GPU model.
source locator
evidence/runs/recorded-b75c67aa/platforms/P3/attempt-1/receipt-main.json, JSON pointer /results/w1/sequences_per_second
review
method: execution-evidence-review; source-hash-verification; reviewer: claude; date: 2026-11-21; artifact sha256: 4bfe70c67740043bef2f383beade8bdb09696f88a4c673c1194520b354e7ff30; retrieval url: https://raw.githubusercontent.com/rewire-bio/evo2-cloud-operations/f8ae11c9a1c45505487c978b4cb7e064c4e59905/evidence/runs/recorded-b75c67aa/platforms/P3/attempt-1/receipt-main.json; method note: Value read by build_batch.py as the literal JSON number token at the pointer in the pinned file, then re-read from the same bytes and compared; file hash recomputed from a fresh download of the raw URL; reviewer note: Claude agent, automated. The same agent wrote the generator and ran its re-read check; no independent second pass and no human review; note: Read from the run receipt of this platform, which is the full W1 workload. One run; no interval. Source checked, not marked reproduced: no run recipe names the receipt.
missing metadata
uncertainty: reason: unreported; note: The file gives no interval at this pointer.
Related records

Suggest a correction