ggml-org/llama.cpp

ggml-org/llama.cpp — High-Cadence Infra Bet on LLM Inference

llama.cpp remains the de facto lightweight C/C++ inference layer with extreme release velocity, but star growth is decelerating and the momentum score rests on a single measurable signal family.

Key numbers

• GitHub stars: 120,327 [Mgh_stars]
• Stars 30d velocity: +78 [Mgh_stars]
• Stars 7d velocity: +4 [Mgh_stars]
• Forks: 20,572 [Mgh_forks]
• Open issues: 1,844 [Mghopenissues]
• Momentum score: 97.1 [S2026-07-14]
• Momentum z (code family): 3.517 [S2026-07-14]

Bull case

• Momentum score of 97.1 places the repo at the extreme top of its datacentres cohort, driven by a strong code-release z-score of 3.517. [S2026-07-14]
• Release cadence is exceptionally high — over a dozen tagged releases (b9982 through b10001) in a two-day window, signaling active maintenance and rapid iteration. [E4780 E4781 E4776 E675 E676 E5 E6 E7 E8 E9 E10 E11 E12 E13 E14]
• Large existing installed base (120k+ stars, 20k+ forks, 779 subscribers) provides a durable distribution moat for a C/C++ inference runtime. [Mghstars Mghforks Mgh_subscribers]
• Open prediction market gives 60% odds of top-decile 90-day star growth within its cohort, consistent with current momentum readings. [P1 S2026-07-14]

Bear case

• Star growth has slowed sharply in absolute terms — only +4 stars in 7 days and +78 in 30 days against a base of 120k, implying a growth rate under 0.1% monthly, far below what the 97.1 momentum score might suggest to a casual reader. [Mgh_stars]
• The momentum score is built on a single present family ('code'/gh_stars) while five other families — models, apps, hiring, sdk, attention — are entirely missing, meaning the score reflects one narrow signal rather than broad-based demand or commercial traction. [S2026-07-14]
• Forks, open issues, and subscriber counts show zero 30-day velocity, suggesting community engagement growth (beyond stars) has stalled even as release volume stays high. [Mghforks Mghopenissues Mghsubscribers]
• High release frequency (15+ tagged builds in ~2 days) could reflect churn/instability or minor incremental builds rather than substantive feature velocity — the events pack offers no differentiation on release significance. [E4780 E4781 E4776 E675 E676 E5]
• No revenue, funding, hiring, or downstream adoption signals are present in this pack, leaving the EV case entirely dependent on open-source mindshare with no visibility into monetization or ecosystem lock-in. [S2026-07-14]

What would change our mind

• 30d star velocity re-accelerates materially (e.g., back above prior highs) confirming the momentum score is broad-based rather than a single-metric artifact. [Mgh_stars S2026-07-14]
• New data families (models, apps, sdk, hiring, attention) populate the score, indicating ecosystem or commercial signals beyond raw stars. [S2026-07-14]
• Fork/subscriber counts show positive 30d velocity, evidencing renewed developer engagement rather than stagnant community metrics. [Mghforks Mghsubscribers]
• The open prediction resolves, confirming or denying top-decile 90-day cohort star growth. [P1]

Verdict

llama.cpp shows an eye-catching momentum score but the underlying stars metric is nearly flat on a 7-day basis and the score is derived from a single, narrow data family — treat the 97.1 headline with skepticism pending broader signal confirmation.

Confidence: 0.35

Sources

• [Mgh_stars] metric gh_stars (latest 120327)
• [Mgh_forks] metric gh_forks (latest 20572)
• [Mgh_open_issues] metric ghopenissues (latest 1844)
• [S2026-07-14] score 2026-07-14 · momentum 97
• [E4780] gh_release 2026-07-14 — ggml-org/llama.cpp b10001 — source
• [E4781] gh_release 2026-07-14 — ggml-org/llama.cpp b10000 — source
• [E4776] gh_release 2026-07-14 — ggml-org/llama.cpp b9999 — source
• [E675] gh_release 2026-07-14 — ggml-org/llama.cpp b9996 — source
• [E676] gh_release 2026-07-14 — ggml-org/llama.cpp b9995 — source
• [E5] gh_release 2026-07-14 — ggml-org/llama.cpp b9994 — source
• [E6] gh_release 2026-07-13 — ggml-org/llama.cpp b9993 — source
• [E7] gh_release 2026-07-13 — ggml-org/llama.cpp b9992 — source
• [E8] gh_release 2026-07-13 — ggml-org/llama.cpp b9990 — source
• [E9] gh_release 2026-07-13 — ggml-org/llama.cpp b9988 — source
• [E10] gh_release 2026-07-13 — ggml-org/llama.cpp b9987 — source
• [E11] gh_release 2026-07-13 — ggml-org/llama.cpp b9986 — source
• [E12] gh_release 2026-07-13 — ggml-org/llama.cpp b9985 — source
• [E13] gh_release 2026-07-13 — ggml-org/llama.cpp b9984 — source
• [E14] gh_release 2026-07-13 — ggml-org/llama.cpp b9982 — source
• [Mgh_subscribers] metric gh_subscribers (latest 779)
• [P1] prediction: ggml-org/llama.cpp will rank in the top decile of its cohort's 90-day star growth