rewirebio.iobenchmarks
Source

TorchX project site (torchx-cpl.github.io)

Project site as served on 2026-11-15 · Project source

Original source

What this source contributed

Records in this release that cite this source. A record can cite several sources.

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Model or configurationMetricValueDatasetEvaluation
TorchCraft on A100 with cuequivariance kernelsRuntime (wall-clock minutes per design at 100 residues, batch size 2 run divided by 2)3.92 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 100 residues, GPU w/ cuequivariance
TorchCraft on A100 without cuequivariance kernelsRuntime (wall-clock minutes per design at 100 residues, batch size 2 run divided by 2)4.76 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 100 residues, GPU w/o cuequivariance
TorchCraft on Ascend 910B with fused attentionRuntime (wall-clock minutes per design at 100 residues, batch size 2 run divided by 2)3.1 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 100 residues, NPU w/ FA
TorchCraft on Ascend 910B without fused attentionRuntime (wall-clock minutes per design at 100 residues, batch size 2 run divided by 2)3.0 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 100 residues, NPU w/o FA
TorchCraft on A100 with cuequivariance kernelsRuntime (wall-clock minutes per design at 200 residues, batch size 2 run divided by 2)4.27 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 200 residues, GPU w/ cuequivariance
TorchCraft on A100 without cuequivariance kernelsRuntime (wall-clock minutes per design at 200 residues, batch size 2 run divided by 2)7.96 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 200 residues, GPU w/o cuequivariance
TorchCraft on Ascend 910B with fused attentionRuntime (wall-clock minutes per design at 200 residues, batch size 2 run divided by 2)3.75 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 200 residues, NPU w/ FA
TorchCraft on Ascend 910B without fused attentionRuntime (wall-clock minutes per design at 200 residues, batch size 2 run divided by 2)4.13 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 200 residues, NPU w/o FA
TorchCraft on A100 with cuequivariance kernelsRuntime (wall-clock minutes per design at 300 residues, batch size 2 run divided by 2)4.54 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 300 residues, GPU w/ cuequivariance
TorchCraft on A100 without cuequivariance kernelsRuntime (wall-clock minutes per design at 300 residues, batch size 2 run divided by 2)13.08 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 300 residues, GPU w/o cuequivariance
TorchCraft on Ascend 910B with fused attentionRuntime (wall-clock minutes per design at 300 residues, batch size 2 run divided by 2)6.23 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 300 residues, NPU w/ FA
TorchCraft on Ascend 910B without fused attentionRuntime (wall-clock minutes per design at 300 residues, batch size 2 run divided by 2)8.63 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 300 residues, NPU w/o FA
TorchCraft on A100 with cuequivariance kernelsRuntime (wall-clock minutes per design at 400 residues, batch size 2 run divided by 2)5.85 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 400 residues, GPU w/ cuequivariance
TorchCraft on A100 without cuequivariance kernelsRuntime (wall-clock minutes per design at 400 residues, batch size 2 run divided by 2)21.65 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 400 residues, GPU w/o cuequivariance
TorchCraft on Ascend 910B with fused attentionRuntime (wall-clock minutes per design at 400 residues, batch size 2 run divided by 2)11.29 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 400 residues, NPU w/ FA
TorchCraft on Ascend 910B without fused attentionRuntime (wall-clock minutes per design at 400 residues, batch size 2 run divided by 2)14.64 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 400 residues, NPU w/o FA
TorchCraft on A100 with cuequivariance kernelsRuntime (wall-clock minutes per design at 500 residues, batch size 2 run divided by 2)7.59 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 500 residues, GPU w/ cuequivariance
TorchCraft on A100 without cuequivariance kernelsRuntime (wall-clock minutes per design at 500 residues, batch size 2 run divided by 2)33.3 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 500 residues, GPU w/o cuequivariance
TorchCraft on Ascend 910B with fused attentionRuntime (wall-clock minutes per design at 500 residues, batch size 2 run divided by 2)16.56 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 500 residues, NPU w/ FA
TorchCraft on Ascend 910B without fused attentionRuntime (wall-clock minutes per design at 500 residues, measured at batch size 1 because batch size 2 exceeded memory, not the batch-size-2 run divided by 2 that every other cell of the table reports)33.47 minuteTorchCraft runtime benchmark design runsTorchCraft runtime at 500 residues, NPU w/o FA

Claims

Statements this source supports, each with where in the source it was checked.

  • TorchCraft on Ascend 910B without fused attention: The 500-residue NPU run without fused attention used batch size 1: Measurement exception The 500-residue timing for the Ascend 910B without fused attention was measured at batch size 1 because batch size 2 exceeded memory. Every other cell of the table is a batch-size-2 run divided by 2. (Supplementary Table S9 footnote: '* Batch size 1; batch size 2 exceeded memory (OOM).')
  • TorchCraft runtime benchmark across sequence lengths: How the runtime figures were measured: Measurement setup Runtime is reported in minutes per design, obtained by dividing the total wall-clock time of a batch-size-2 run by 2, on an Ascend 910B with 64 GB with and without fused attention and on an A100 with 80 GB with and without cuequivariance kernels. Batch size 2 is the recommended setting and GPU timings are the mean of three independent runs. (Supplementary Table S9 caption)
  • TorchFold on the gated TorchFold antibody-distilled checkpoint: Developer-reported TorchFold benchmark figure in the README: Developer reported benchmark The repository README states that TorchFold reaches 70.1% ranked success on FoldBench-AB, compared with AlphaFold3, Protenix-v2 and OpenDDE, with MSAs for every method generated by the official AF3 Jackhmmer search, and that two-round distillation expanded unique antibody-antigen examples from 5.0k to 24.5k. (README sections 'TorchFold overview' and 'TorchFold Benchmark')
  • TorchX: The four TorchX modules: Suite modules TorchX has four modules: TorchFold for structure prediction, TorchScore for score-only evaluation of supplied coordinates, TorchCraft for binder design, and TorchFold Train for training and fine-tuning. The repository is split into a GPU tree on main and an Ascend NPU branch, which the README says share the same scientific modules. (README sections 'TorchX', 'Repository layout' and 'Other branches')

Models, benchmarks and other records

Methods (2)

Configurations (4)

Pipelines (1)

Benchmarks (1)

Datasets (1)

Evaluations (20)

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.

25 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance
Artifact SHA-256
fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.artifact_sha256

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Evidence concerns
1 values
  • applies to
    • torchcraft-20261115-claim-project-site-licence-statement
    artifact sha256
    fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c
    message
    The site's closing line says the GPU and NPU implementations, the TorchFold training code and the TorchFold-AbAg model weights are open under the MIT license. The repository's own LICENSE file and README at commit ebf9dc6676ecb1be505bc72f87de408995f0d8fd say Apache License 2.0, and the GitHub repository metadata reports apache-2.0. The gated Hugging Face dataset is tagged mit but conditions access on no redistribution, rehosting, selling or sharing without written permission. The MIT statement is therefore not relied on.
    review date
    2026-11-16
    source id
    torchcraft-20261115-source-project-site
    source locator
    Project site, closing line below the citations block
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.evidence_concerns

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Limitations
2 values
  • Mutable page; the hash describes one retrieval on 2026-11-15.
  • Its TorchFold figures belong to a separate TorchFold report that could not be retrieved; none of them is used here.
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.limitations

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Media type
text/html
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.media_type

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Publication status
official_project_source
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.publication_status

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Retrieved at
2026-11-15T23:36:25Z
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.retrieved_at

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Review: artifact sha256
fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.review.artifact_sha256

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Review: method note
Re-retrieved and hashed by the reviewer. All seven pinned artifacts were re-retrieved by the reviewer on 2026-11-16T00:08Z. Six came back byte-identical to the hash on their source record. The Hugging Face page did not, and cannot: three consecutive fetches gave three different hashes. Normalising its per-request hf-sanitized CSS class hashes leaves one further per-request difference, window.requestId, and the page also prints a download counter; with those normalised the markup is identical. Its licence tag, gate text and dataset card matched the recorded quotations word for word, so the page is a per-retrieval stamp, not a changed page. The pdftotext -layout text layer of the refetched PDF hashes to 90f1da5f8a936766087beba0ade3f038f5b69504b9f581b42f2391f5640c5e6b, the value the extractor recorded.
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.review.method_note

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Review method
2 values
  • source-hash-verification
  • automated-source-review
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.review.method

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

Review: retrieval url
https://torchx-cpl.github.io/
Source metadata
TorchX project site (torchx-cpl.github.io)

Original source ↗

No field-specific location recorded

Version: Project site as served on 2026-11-15
Retrieved: 2026-11-15T23:36:25Z

catalogued

No individual claim review recorded

Source has a recorded evidence concern. Consult its source page before using the claim.

Audit details

Field: attributes.review.retrieval_url

Source artifact SHA-256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c

Hash scope: Hash scope not separately documented; inspect source record

How to read the evidence

Sources and history

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

The official TorchX project page, which summarises TorchFold and TorchCraft, reprints the runtime table and states a licence for the implementations and weights.

1 source record and release history

Read original source

Download this release (gzip)
Technical metadata and extraction receipts

Stable ID: torchcraft-20261115-source-project-site

areas
protein-structure; molecular-interactions
contexts
research
artifact sha256
fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c
evidence concerns
applies to: torchcraft-20261115-claim-project-site-licence-statement; artifact sha256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c; message: The site's closing line says the GPU and NPU implementations, the TorchFold training code and the TorchFold-AbAg model weights are open under the MIT license. The repository's own LICENSE file and README at commit ebf9dc6676ecb1be505bc72f87de408995f0d8fd say Apache License 2.0, and the GitHub repository metadata reports apache-2.0. The gated Hugging Face dataset is tagged mit but conditions access on no redistribution, rehosting, selling or sharing without written permission. The MIT statement is therefore not relied on.; review date: 2026-11-16; source id: torchcraft-20261115-source-project-site; source locator: Project site, closing line below the citations block
limitations
Mutable page; the hash describes one retrieval on 2026-11-15.; Its TorchFold figures belong to a separate TorchFold report that could not be retrieved; none of them is used here.
media type
text/html
publication status
official_project_source
retrieved at
2026-11-15T23:36:25Z
review
artifact sha256: fa95060dad51bd3e04278b48c7e3bc62c66f8f909db3871e654f79d5a98dcd3c; method: source-hash-verification; automated-source-review; method note: Re-retrieved and hashed by the reviewer. All seven pinned artifacts were re-retrieved by the reviewer on 2026-11-16T00:08Z. Six came back byte-identical to the hash on their source record. The Hugging Face page did not, and cannot: three consecutive fetches gave three different hashes. Normalising its per-request hf-sanitized CSS class hashes leaves one further per-request difference, window.requestId, and the page also prints a download counter; with those normalised the markup is identical. Its licence tag, gate text and dataset card matched the recorded quotations word for word, so the page is a per-retrieval stamp, not a changed page. The pdftotext -layout text layer of the refetched PDF hashes to 90f1da5f8a936766087beba0ade3f038f5b69504b9f581b42f2391f5640c5e6b, the value the extractor recorded.; retrieval url: https://torchx-cpl.github.io/; reviewed at: 2026-11-16T01:05:00Z; reviewer: claude; reviewer note: Claude (Opus 5) review agent, independent of the extractor; no human review claimed. Every pinned artifact was re-retrieved and hashed, and every stored value re-read against it. Nothing was reproduced: no design run, score, docking calculation or binding measurement was repeated, and no value was recomputed from source data.
source locator
Sections '1. TorchFold', '2. TorchCraft', 'Runtime on GPU and Ascend NPU' and the closing licence line
url
https://torchx-cpl.github.io/
version
Project site as served on 2026-11-15
Related records

Suggest a correction