Current repository readingScanned Jul 24, 2026, 5:23 AM UTC

MTPLX Codebase Valuation

3x decode TPS increase On Qwen 3.6 27B @ temp 0.6 | Native MTP Speculative Decoding On Apple Silicon With No External Drafter.

View youssofal/MTPLX on GitHub
Open share card
CodeValuation.com
RepoWorth
github.com/youssofal

MTPLX

3x decode TPS increase On Qwen 3.6 27B @ temp 0.6 | Native MTP Speculative Decoding On Apple Silicon With No External Drafter.

RepoWorth$37,482Current measured development value · 22.0 active build days
Rebuild complexity90.448/100Profile: Complex
2.8M code tokens600 source filesPython primary language
Scanned Jul 24, 2026Public GitHub repositorycodevaluation.com/repoworth/mtplx
Authored code2.8M tokens · 287,212 source lines
Complexity90.448/100 · Complex
Primary stackPython
Scan coverage100% · Content scan
RepoWorth development valuation

One valuation, with the evidence behind it.

RepoWorth models the current development value represented by the visible repository evidence. It does not include ideation, intellectual property, business value, product discovery, or sale price.

RepoWorthCurrent measured development value
$37,482

Modeled range $29,986 to $44,979

22.0 active build days176.3 engineering hours2.2M implementation tokens
Repository evidenceMeasured at the scanned commit
2.8M tokens

287,212 source lines across 600 authored files

127 test files107 dependencies100% scan coverage
Measured RepoWorth history

The current RepoWorth starts the measured history.

This current scan is the first measured point. Later scans add comparable history when the repository changes. RepoWorth is not back-cast onto commits that were never scanned.

Recorded RepoWorth$37,482
Current measured reading
Measured · Jul 24, 2026Data through · Jul 24, 2026
Measurement basisCurrent readingCurrent published repository scan
Evidence coverage100%current scan
Complexity90/100Measured from code scale and structural signals
Maintainability70/100Current structure, tests, and generated-code balance

This current scan establishes the RepoWorth baseline. Comparable value changes appear only after another completed scan under the same measurement basis.

RepoWorth history compares completed repository scans only when their measurement basis is compatible. Commit activity is shown separately so delivery volume is not presented as code value.

Repository contributors

Top contributors.

Ranked from the public GitHub contributor evidence available at the latest scan. This table describes measured repository activity, not ownership or original authorship.

RankContributorContributionsShare of top groupLatest measured activity
#1youssofal@youssofal21188.7%Current GitHub contributor snapshot
#2daniel-farina@daniel-farina93.8%Current GitHub contributor snapshot
#3mmmugh@mmmugh83.4%Current GitHub contributor snapshot
#4PhilipJohnBasile@PhilipJohnBasile20.8%Current GitHub contributor snapshot
#5titan550@titan55020.8%Current GitHub contributor snapshot
#6SuperMarioYL@SuperMarioYL20.8%Current GitHub contributor snapshot
#7ashalliants@ashalliants10.4%Current GitHub contributor snapshot
#8Jonathangadeaharder@Jonathangadeaharder10.4%Current GitHub contributor snapshot
#9shiftedx@shiftedx10.4%Current GitHub contributor snapshot
#10jedisct1@jedisct110.4%Current GitHub contributor snapshot
About MTPLX

What the public repository appears to contain.

3x decode TPS increase On Qwen 3.6 27B @ temp 0.6 | Native MTP Speculative Decoding On Apple Silicon With No External Drafter.

anthropic-compatibleapple-siliconinference-enginelocal-aimetalmlxmtpmtplxnative-mtpopenai-compatibleqwenqwen3-next
License
Apache-2.0
Default branch
main
GitHub stars
1,088
Forks
75
Last public push
Jul 21, 2026
Repository state
Active public repository
Detected language profile11.9MB authored source
Python60.87%
Swift23.38%
JavaScript12.42%
TypeScript1.35%
C++0.78%
Shell0.74%
CSS0.39%
C/C++0.06%
Source files600
Test files127
Manifests3
Dependencies107
Measured technical evidence

A fuller view of what the scanner found.

These values come from the same immutable repository scan as the headline RepoWorth. They stay tied to the scanned commit so the reading can be inspected against one exact source state.

Maintainability signal70/100

Established based on deterministic repository structure and complexity signals.

Production source2.2M tokens

First-party implementation tokens eligible for the modeled code-scale input.

Test evidence606.3K tokens

127 detected test files contribute separately from production code.

Configuration and infrastructure1.5K tokens

3 manifests and 0 workspace boundaries were detected.

Documentation44.5K tokens

52 documentation files remain visible as a distinct enablement signal.

Integration surface4,015 signals

107 dependency declarations and 532 public-interface signals were measured.

Repository structure7 levels deep

0 large files and 50,012 branch signals add structural context.

Excluded from authored evidence0 files

3 generated files were identified. Ignored and generated material does not inflate RepoWorth.

Measured repository distinctions

Repository evidence highlights.

Each highlight records a measured characteristic of the same repository scan used for this RepoWorth report.

architecture

Interface builder

At least 500 deterministic public interface signals were measured in source.

architecture

Integration surface

At least 1,000 deterministic integration signals were measured in source.

quality

Verification depth

At least 20 percent of measured source tokens are in test paths.

Technical effort profile

What makes this repository harder to rebuild.

Code scale, branching, dependencies, languages, structure, and visible test evidence shape the implementation and verification effort represented by this repository.

Code scaleHigh

Measured from the current source archive.

ComplexityHigh

Derived from branch density, dependencies, languages, and code scale.

VerificationHigh

Measures visible test code relative to the source corpus.

Development effort composition

Source volume alone does not explain the work.

This view separates authored evidence from the implementation, integration, verification, and engineering oversight represented by the current repository.

Reading boundary: These are modeled development inputs for the visible code. They do not include product discovery, company value, private systems, or commercial outcomes.

Weighted authored evidence2.2MProduction, tests, configuration, and documentation
Estimated implementation output2.2MImplementation, retries, integration, and verification
Engineering oversight176.3 hoursOrientation, integration, tests, and review
Active build time22.0 daysModeled effort represented by the current source state
Evidence boundary

What this valuation includes.

CodeValuation.com counted supported authored source on commit c1300f17c66e5fef7810efa80f2a53489b8d001d, measured deterministic repository signals, then estimated implementation, rework, testing, integration, and engineering oversight.

Included

Supported first-party source, tests, configuration, manifests, repository structure, dependency signals, and framework boundaries.

Excluded

Common generated and vendored files, build output, dependencies, lockfiles, minified assets, binaries, media, and private systems outside this repository.

Not claimed

Ideation, intellectual-property rights, security posture, production reliability, company value, sale price, revenue, user value, proprietary data, or the cost of discovering the original product.

This is a modeled current development valuation for the repository's visible code at a specific commit. It reflects code volume, implementation output, complexity, and engineering oversight. It is not a business valuation, investment appraisal, security review, or estimate of sale price, ideation, intellectual property, revenue, users, brand, proprietary data, undocumented domain knowledge, or private infrastructure.

Check another repository

Run RepoWorth on your own codebase.

Enter a public GitHub URL to create a current repository reading.

No sign-up for public repositories. Private repositories require a free trial. Evaluated in real time directly from GitHub. Code is never stored. Derived measurements and valuation history are saved.