Public record
Software health reportschema 0.34.0 · metrics 2.10.0 · 2026-09-14 12:25 UTC

flagos-ai / FlagGems

FlagGems is an operator library for large language models implemented in the Triton Language.

PythonApache-2.0★ 1,120 stars⑂ 528 forkssince Mar 2024View on GitHub ↗
KindCommand-line toolLibraryhow this is determined
#pytorch#triton#triton-kernels

flagos-ai/FlagGems holds a health index of 95 out of 100, placing it in the Exceptional band. It scores highest on Vitality (100/100) and lowest on Security (64/100). It was last updated today. 8 contributors account for most of its recent work.

95
overall / 100
Exceptional

Software health index

Metrics are grouped into weighted categories on one standardized 1–100 scale. Overall starts as their weighted mean, calibrated against the distribution of the public record so bands carry percentile meaning; when public evidence triggers the High-Risk Jurisdiction Policy, the rating is adjusted and receives an At Risk ceiling of 34.

95
Exceptional93-100The record's top tier (≈ top 5%); essentially all checked criteria met
Excellent80-92Strong across the board; minor gaps
Good65-79Healthy; gaps are limited and manageable
Moderate50-64Acceptable with notable gaps; review recommended
Weak35-49Material weaknesses across several areas
At Risk20-34Significant weaknesses; adoption warrants caution
Critical1-19Severe problems (abandoned, single-maintainer, no hygiene)
VitalityCommunity &AdoptionSustainability &GovernanceEngineeringQualitySecurityAI Readiness

Score profile

Each axis is a category. The shape matters more than the average — a healthy subject fills the whole shape, while a spike-and-crater profile means strength in one dimension is masking risk in another.

The weighted overall 82 is calibrated to 95 on the published index scale (record calibration 2026-08-02).

Ownership

FlagOSOrganization
288 followers53 public repossince Nov 2025

This repository is backed by an organization — shared, accountable stewardship that can outlive any single maintainer.

Package ecosystems

RegistryPackageVersionDownloads / moVersionsLast publish
PyPIflag_gems5.0.21,5897143 days ago

Metrics by category

Vitality

Is the project alive — is code being written and are releases shipping?

100Exceptional · 21% of overall

Development activity

100Exceptional
How it's scored
36/36Push recency — last push 0 days ago
36/36Commit cadence — 52/52 weeks with commits
18/18Commit volume — 2,817 commits in the last year
0/10OpenSSF Scorecard: Maintained — no data
Inputs used
commits_last_year2,817
human_commit_share1
days_since_last_push0
active_weeks_last_year52

Release discipline

100Exceptional
How it's scored
27/27Ships releases — 6 releases published
36/36Release recency — latest release 82 days ago
27/27Release cadence — a release every ~34.1 days
0/10OpenSSF Scorecard: Signed-Releases — no data
Inputs used
releases_count6
latest_release_tagv5.3.0
releases_from_tagsno
days_since_latest_release82
mean_days_between_releases34.1

Community & Adoption

Does the project have users, downloads, attention, and a welcoming setup for contributors?

77Good · 17% of overall

Popularity & adoption

79Good
How it's scored
49.5/60Stars — 1,120 stars
22.7/25Forks — 528 forks
6.5/15Watchers — 16 watchers
Inputs used
forks528
stars1,120
watchers16
growth_stateunverified
growth_factor_pct100
growth_unverified_reasonno_history

Community health

92Excellent
How it's scored
22.5/22.5README
22.5/22.5License — recognized license (Apache-2.0)
18/18CONTRIBUTING guide
13.5/13.5Code of conduct
0/7.2Issue template
6.3/6.3PR template
Inputs used
has_readmeyes
has_licenseyes
readme_badges0
has_contributingyes
has_issue_templateno
has_code_of_conductyes
readme_badge_services—
has_pull_request_templateyes

Ecosystem adoption (downloads)

53Moderate
How it's scored
42.7/80Monthly downloads — 1,589 downloads/month across pypi
0/20Registry dependents — not reported by this ecosystem
Inputs used
packagesflag_gems
dependents—
ecosystemspypi
total_downloads—
monthly_downloads1,589
unverified_packages_excluded—
Excluded from scoring (no data or not applicable): Registry dependents. Remaining weights renormalized.

Sustainability & Governance

Will the project survive its people — bus factor, responsiveness, who backs it, and package upkeep?

78Good · 23% of overall

Maintainer resilience (bus factor)

93Exceptional
How it's scored
54/54Bus factor — 8 contributor(s) cover half of all commits
16.5/22.5Commit distribution — top contributor authored 27% of commits
13.5/13.5Contributor breadth — 99 contributors
0/10OpenSSF Scorecard: Contributors — no data
Inputs used
bus_factor8
contributors_sampled99
top_contributor_share0.268

Issue & PR responsiveness

58Moderate
How it's scored
20.2/42Issue resolution — 48% of issues closed
21.6/30PR acceptance — 3,460/4,811 decided PRs merged
0/13Newcomer PR acceptance — no first-time contributor's PR decided in 30d
0/15OpenSSF Scorecard: Code-Review — no data
Inputs used
merged_prs3,460
open_issues422
closed_issues392
prs_merged_7d46
prs_decided_7d58
prs_merged_30d46
prs_decided_30d58
issue_closed_ratio0.482
closed_unmerged_prs1,351
first_time_authors_30d0
first_time_prs_merged_30d0
first_time_prs_decided_30d0
Excluded from scoring (no data or not applicable): Newcomer PR acceptance. Remaining weights renormalized.

Ownership & stewardship

62Moderate
How it's scored
30/30Ownership backing — organization-owned
0/20Verified domain
17.7/25Owner reach — 288 followers of flagos-ai
14.3/25Track record — 53 public repos, account ~0 yr old
Inputs used
followers288
owner_typeOrganization
is_verifiedno
owner_loginflagos-ai
public_repos53
account_age_days310

Package maintenance

100Exceptional
How it's scored
25/25Published & resolvable — 1 package(s) on pypi
35/35Publish recency — latest publish 143 days ago
20/20Version history — 7 published versions
20/20Not deprecated — active, not deprecated or yanked
Inputs used
packagesflag_gems
ecosystemspypi
any_deprecatedno
min_days_since_publish143

Engineering Quality

Are baseline engineering and documentation practices in place?

91Excellent · 19% of overall

Engineering practices

92Excellent
How it's scored
24/24CI workflows — 19 workflow(s)
24/24Tests present
16/16Linter config
9.6/9.6Pre-commit hooks
0/6.4.editorconfig
0/20OpenSSF Scorecard: CI-Tests — no data
Inputs used
has_ciyes
has_testsyes
has_editorconfigno
has_linter_configyes
has_precommit_configyes

Documentation

90Excellent
How it's scored
30/30README
25/25Documentation directory
15/15Documentation / homepage site — https://flagos-ai.github.io/FlagGems
10/10Repository description
10/10Topics — 3 topics
0/10Wiki
Inputs used
topicspytorch, triton, triton-kernels
has_wikino
homepage—
docs_sitehttps://flagos-ai.github.io/FlagGems
has_readmeyes
has_docs_diryes
has_descriptionyes

Security

Are visible security and supply-chain practices strong, without unresolved high-risk jurisdiction exposure?

64Moderate · 16% of overall

Security posture

55Moderate
How it's scored
30/30Security policy (SECURITY.md)
25/25Dependabot config
0/25Dependency lockfiles
0/20CodeQL workflow
Inputs used
sourcefile_signals
lockfiles—
manifestscpp/pyproject.toml, pyproject.toml, setup.py
has_codeql_workflowno
has_security_policyyes
has_dependabot_configyes

Dependency advisories

100Exceptional
How it's scored
35/35Direct dependencies free of known advisories — no direct dependency carries a known advisory
25/25Indirect dependencies free of known advisories — no indirect dependency carries a known advisory
0/40No advisories left outstanding — no advisory carries a publication date
Inputs used
sourceosv
advisories0
affected_packages0
assessed_packages6
unassessed_packages0
affected_by_severitynone
direct_affected_packages0
Excluded from scoring (no data or not applicable): No advisories left outstanding. Remaining weights renormalized. Matched the pypi:flag_gems@5.0.2 runtime dependency closure — what installing the published package pulls in — 6 packages. Reachability is not analyzed.

AI Readiness

How well is the repo equipped to be developed and maintained with AI coding agents? Carries a deliberately small weight (4%): agent tooling is a real maintenance signal, but a repository with none can still reach 100/100.

66Good · 4% of overall

Agent context & guidance

40Weak
How it's scored
0/45Agent instructions — no CLAUDE.md / AGENTS.md / editor rules
0/15Machine-readable docs (llms.txt)
40/40Legible commit history — 89 of 100 human commits state their intent (structured subject or explanatory body)
Inputs used
has_llms_txtno
llms_txt_url—
legible_history_share0.89
agent_instruction_files—
agent_instruction_max_bytes—

Verify loop (build / test / typecheck)

78Good
How it's scored
12.6/18One-command bootstrap — docs/themes/hugo-book/go.mod (toolchain convention, no task runner)
22/22Automated tests
11/11Lint / format config
0/11Static type checking
10/10Reproducible environment — devcontainer, Dockerfile
10/10Demonstrated agent practice — 17 of the last 100 commits agent-authored or agent-credited
5/8Automated maintenance — dependency automation configured, none observed in the sampled commits
0/10OpenSSF Scorecard: Pinned-Dependencies — no data
Inputs used
has_nixno
has_testsyes
lockfiles—
has_dockerfileyes
typed_languageno
bootstrap_files—
has_devcontaineryes
has_linter_configyes
typecheck_configs—
agent_commit_share0.17
toolchain_manifestsdocs/themes/hugo-book/go.mod
dependency_bot_commit_share0

Code legibility for models

55Moderate
How it's scored
0/45Type-checkable code — Python without a type-check config
54.6/55Manageable file sizes — 38/4,752 source files over 60KB
Inputs used
primary_languagePython
largest_source_bytes2,707,645
source_files_sampled4,752
oversized_source_files38

Machine-readable interfaces

100Exceptional
How it's scored
0/40API schema (OpenAPI/GraphQL/proto) — not applicable to this kind of software
0/20MCP server — not applicable to this kind of software
40/40Runnable examples — example, examples
Inputs used
example_dirsexample, examples
has_mcp_signalno
api_schema_files—
interfaces_expected_of—
Excluded from scoring (no data or not applicable): API schema (OpenAPI/GraphQL/proto), MCP server. Remaining weights renormalized.

Key facts

1,120GitHub stars
99contributors
2,817commits, last 12 months
0days since last push
6releases
8bus factor
422open issues
PyPIpackage ecosystems

Data collection warnings

  • Star history unavailable: GitHub GraphQL error: Resource not accessible by personal access token
  • First-time contributor figures cover 12 of 30 authors (cap 12)
  • Could not fetch pypi package 'flag-gems-cpp-__VENDOR__' from its registry
  • OpenSSF Scorecard did not return a usable result (killed by SIGKILL — most likely the container's memory limit; 2026/09/14 12:24:32 Warning: PATs stored in env variables GITHUB_AUTH_TOKEN and GITHUB_TOKEN differ. Scorecard will use the former.); skipping Scorecard checks

More detail

Star and fork history 0 ★ / 528 ⇿
0Stars
528Forks
6Releases

When each star and fork was added, collected from GitHub and bucketed by day. Cumulative growth sits directly above the daily additions it is made of, so the two read against each other: steady organic accretion looks nothing like an abrupt, short-lived burst. Where that difference is measurable, it is reported as growth authenticity.

0100200300400500600502102024-062025-072026-09
Major 1Minor 2Patch 1

Each point covers 3 days.

Direct dependencies 5
RegistryPackageVersion constraintManifest
PyPIflag-gems—cpp/pyproject.toml
PyPIpackaging>=26.0pyproject.toml
PyPIPyYAML==6.0.1pyproject.toml
PyPIsqlalchemy==2.0.48pyproject.toml
PyPInumpy—pyproject.toml
All dependencies 59

Full resolved dependency set from the GitHub dependency graph: 4 direct and 55 indirect (transitive) packages. The transitive closure is complete when the repository commits a lockfile.

RegistryPackageVersionRelation
PyPInumpy2.2.6direct
PyPIpackaging—direct
PyPIpyyaml6.0.1direct
PyPIsqlalchemy2.0.48direct
PyPIapex0.1indirect
PyPIattrs24.2.0indirect
PyPIbenchflow1.0.0indirect
PyPIcambricon-dali0.13.0indirect
PyPIcmake—indirect
PyPIcolorama0.4.6indirect
PyPIcoverage7.13.5indirect
PyPIdatasets—indirect
PyPIdecorator5.1.1indirect
PyPIdistro1.9.0indirect
PyPIenflame-modelopt3.6.20260615+torch.2.11.0indirect
PyPIflash-attn2.4.2+7e2dd4dindirect
PyPIflash-linear-attention0.5.0+metax3.8.1.0torch2.10indirect
PyPIflash-mla1.0.1+metax3.8.1.0torch2.10indirect
PyPIflashinfer0.2.6+metax3.8.1.0torch2.10indirect
PyPIhydrax0.1.0indirect
PyPIhyperparameter0.5.6indirect
PyPImkl2024.0.0indirect
PyPIninja—indirect
PyPIopenpyxl3.1.5indirect
PyPIpandas—indirect
PyPIpandas2.3.3indirect
PyPIpsutil6.0.0indirect
PyPIpybind11—indirect
PyPIpybind113.0.3indirect
PyPIpyefml1.9.10indirect
PyPIpytest9.0.3indirect
PyPIpytest-md-report0.8.0indirect
PyPIregex2026.4.4indirect
PyPIscikit-build-core—indirect
PyPIscikit-build-core0.12.2indirect
PyPIscipy—indirect
PyPIsetuptools—indirect
PyPIsetuptools-scm—indirect
PyPItops-extension3.6.20260529+torch.2.11.0indirect
PyPItopstx1.10.6indirect
PyPItorch2.7.0+cpuindirect
PyPItorch-gcu2.10.0+3.7.20260408indirect
PyPItorch-mlu1.29.2+torch2.7.1indirect
PyPItorch-mlu-ops1.8.0+torch2.7.1indirect
PyPItorch-musa2.9.0indirect
PyPItorch-npu2.9.0indirect
PyPItorch-plugin0.1.0indirect
PyPItorch-ptpu0.2.3+torch2.11indirect
PyPItorch-txda0.1.0+20260615.37ba6bbdindirect
PyPItorch-xray2.0.4indirect
PyPItorchaudio2.4.1+metax3.8.1.0indirect
PyPItorchcodec0.6.0+metax3.8.1.0indirect
PyPItorchvision0.22.0+cpuindirect
PyPItransformers—indirect
PyPItxops0.1.0+20260508.60287151indirect
PyPIvllm—indirect
PyPIwheel0.46.2indirect
PyPIxformers0.0.29+1e7a8ec.d20260114indirect
PyPIxmlir1.0.0.1indirect
Dependency advisories 0

Installing pypi:flag_gems@5.0.2 pulls in 6 packages, direct and transitive: 0 carry known advisories, of which 0 are direct dependencies.

No known advisories affect the assessed dependencies.

An advisory means the version recorded in the dependency graph falls inside an advisory’s affected range. Reachability is not analysed, and the graph includes development and test pins — a finding may concern tooling rather than shipped software.

Raw JSON report machine-readable

Feedback

Spotted something off in this report, or have thoughts to share? Wrong measurements, missed tooling, ideas, questions — anything is welcome. Every message is read and gets a response.

The message is kept through sign-in.

Related records

Inspected repositories sharing catalogue tags or ecosystems with flagos-ai/FlagGems. Read them side by side →

98linkedin/Liger-KernelEfficient Triton Kernels for LLM TrainingExceptional91meta-pytorch/tritonparseTritonParse: A Compiler Tracer, Visualizer, and Reproducer for Triton KernelsExcellent77ahb-sjsu/turboquant-proConsumer-aware compression for embedding indexes and LLM KV caches — compress by the metric the downstream consumer actually uses. PCA-Matryoshka + TurboQuant (27x @ 99.8% recall@10), asymmetric K/V, CUDA/Triton kernels, vLLM plugin, replayable CI-gated claims. MIT.Good71pjordanandrsn/experts4bit-qloraTrain and serve MoE models that do not fit in VRAM: fused 4-bit experts, QLoRA, CPU/NVMe offload, and fast inference on consumer NVIDIA GPUs.Good22nactttch/g2nA torch.compile backend that pays attention: custom FX fusion passes, a Triton LayerNorm kernel, and a persistent compile cache. Apache-2.0 free core of the g2n platform.At Risk100huggingface/transformers🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training. Exceptional

Scores are signals, not warranties. They reflect publicly visible practices on GitHub — not a code audit, and not a security guarantee.

Missing data is excluded and weights renormalized, never scored as zero. Methodology is versioned and open: metrics v2.10.0, schema v0.34.0 — full methodology · metrics wiki.

How one result sits in the wider record: aggregate statistics — PyPI.