LLM inference engines: star history compared

Running large language models yourself went from research stunt to commodity in about two years, and these five projects did most of the commoditizing. llama.cpp proved frontier-class models could run on a laptop, Ollama wrapped that power in a one-line install, and vLLM and SGLang turned GPU serving throughput into a public leaderboard of its own. The star histories here are effectively a timeline of the local-AI movement.

Ranked by current GitHub stars · growth = stars gained in the trailing 90 days · data updated Jul 2026

LLM inference engines — star history

FromTo

ollama/ollama
ggml-org/llama.cpp
vllm-project/vllm
sgl-project/sglang
huggingface/text-generation-inference

Ranking

5 repositories

#	Repository	Stars	+90d	First star	Health
1	ollama/ollama	177,460	~357	2023	debt charts
2	ggml-org/llama.cpp	122,290	~781	2023	debt charts
3	vllm-project/vllm	87,826	~388	2023	debt charts
4	sgl-project/sglang	31,027	~353	2024	debt charts
5	huggingface/text-generation-inference	10,887	~22	2022	debt charts

Decision matrix

Repository health compared

Live Postgres analysis · unavailable data stays labeled

Popularity is only one axis. Compare technical scale, maintenance cadence, ownership resilience, and change pressure using the same completed repository analysis behind each report.

:: Reach

Popularity and real-world adoption

Signal	ollama/ollama	ggml-org/llama.cpp	vllm-project/vllm	sgl-project/sglang	huggingface/text-generation-inference
GitHub starsCurrent public GitHub metadata total.	◆177,460	122,290	87,826	31,027	10,887
Stars gained · trailing 90dChange in the newest 90 days covered by each cached series.	+357	◆+781	+388	+353	+22
ForksCurrent public GitHub metadata total.	Not available	Not available	Not available	Not available	Not available
Largest package audienceLargest resolved registry total; sources are shown explicitly.	No resolved package	No resolved package	No resolved package	No resolved package	No resolved package
Repository ageElapsed years since the GitHub creation timestamp.	3.1 years	3.4 years	3.5 years	2.6 years	3.8 years

:: Codebase

Size and technical center of gravity

Signal	ollama/ollama	ggml-org/llama.cpp	vllm-project/vllm	sgl-project/sglang	huggingface/text-generation-inference
Dominant languageLargest share of analyzed code lines at current HEAD.	Not available	Not available	Not available	Not available	Not available
Code linesAnalyzed source lines at current HEAD, excluding blanks and comments.	Not available	Not available	Not available	Not available	Not available
Languages representedLanguages with non-zero files or line counts.	Not available	Not available	Not available	Not available	Not available

:: Maintenance

Commit volume and recent operating cadence

Signal	ollama/ollama	ggml-org/llama.cpp	vllm-project/vllm	sgl-project/sglang	huggingface/text-generation-inference
Repository commitsFull repository commit total reported by the analysis.	Not available	Not available	Not available	Not available	Not available
Commits · trailing 90dCommit count in the newest 90 days covered by the repository analysis.	Not available	Not available	Not available	Not available	Not available

:: Ownership

How broadly maintenance responsibility is distributed

Signal	ollama/ollama	ggml-org/llama.cpp	vllm-project/vllm	sgl-project/sglang	huggingface/text-generation-inference
Bus factorFewest analyzed contributors carrying at least half the commits.	Not available	Not available	Not available	Not available	Not available
Top contributor concentrationShare of attributed commits carried by the most active contributor.	Not available	Not available	Not available	Not available	Not available

:: Change signals

Debt markers and fix pressure in frequently changed files

Signal	ollama/ollama	ggml-org/llama.cpp	vllm-project/vllm	sgl-project/sglang	huggingface/text-generation-inference
Fix-labelled share · frequently changed filesFix-labelled commits divided by commits across the analyzed high-frequency file set.	Not available	Not available	Not available	Not available	Not available
Recent TODO/FIXME movementLatest cumulative marker count and its change over the trailing 90 days.	Not available	Not available	Not available	Not available	Not available

Head to head

open in the builder

llama.cpp vs ollama ollama vs vllm llama.cpp vs vllm

Frequently asked

Which of these llm inference engines has the most GitHub stars?

ollama/ollama is the most-starred of the llm inference engines gitdebt compares, with 177,460 GitHub stars as of Jul 2026. It earned its first star in 2023.

Which of these llm inference engines is growing the fastest?

ggml-org/llama.cpp added the most stars over the last 90 days of its cached history — about 781 new stars, compared with about 357 for the category leader ollama/ollama. 90-day figures are approximate for older repos, whose history is downsampled.

How is this comparison computed?

The table ranks public repository metadata by current total stars. When historical stargazer timestamps are available, gitdebt plots their cumulative series on shared axes and estimates trailing 90-day growth. Repo-health charts come independently from each repository's commit history.

Related categories

all categories

Fastest-growing repos across all of gitdebt