Açık kaynak radarı

Havuz

Takip ettiğimiz konularda yükselen açık kaynak projeler — yıldız ivmesi, gerçek geliştirme aktivitesi ve AI değerlendirmesiyle.

repo · sayfa 1 / 2

Sırala

Yıldız/gün (hız)Son gün (★/gün)Son hafta (★/gün)Son ay (★/gün)Toplam yıldız AI önem skoru Bana uyum Commit (28g)Yazar (28g)En yeni

Konu

Tümü AI / LLM18 Veri9 Geliştirici araçları13 Diğer3 SaaS altyapı13 Web11

Süz

Yalnız hareketli Yalnız analizli≥ 0 ≥ 100 ≥ 500 ≥ 2.000

Yalnız hareketli, son 28 günde commit almamış repoları gizler (yıldızı olan ama bırakılmış projeler). Commit sayısı henüz ölçülmemiş repolar da gizlenir.

EtiketTüm GitHub etiketleri — repo sahiplerinin koyduğu topic'lergöster

Sırala:

Etiketler yükleniyor…

Aktif mixture-of-experts

1
DevDock-AI/Speedy 49
Qwen3.8-Flash-Next (125B MoE) up to 1M context on one or two RTX cards (tuned on 3080 10gb). A measured, tuned and benchmarked fork of Str
8,2/gün ort. +6 son gün 75 commit 9 yazar C++ 01 Kas 2026 güncellendi dün
cuda inference llama llama-cpp llm long-context mixture-of-experts multi-gpu
49
2
ZhuLinsen/LiteMoE 29
A MoE inference engine you can read in an afternoon.
4,8/gün ort. +0 son gün 11 commit 1 yazar Python 01 Kas 2026 güncellendi 4 gün önce
gpu inference llm mixture-of-experts vllm
29
3
leonxger/deepseek-colab 56
Uncensored DeepSeek-V4.1-Flash IQ2_XS on Google Colab
7/gün ort. +3 son gün 1 commit 1 yazar Python 30 Eki 2026 güncellendi 8 gün önce
colab cuda deepseek deepseek-v4 gguf google-colab gpu llama-cpp
56
4
synchopate/llama-modchip 82
🐋 Run 284B DeepSeek-V4-Flash (<59GiB RAM, 44.4 tok/s pp, 9.1-14.7 tok/s) & 30B MoEs (25-111 tok/s) on a pure 64GB CPU at 256K ctx (0 GPUs).
9,1/gün ort. +9 son gün 41 commit 1 yazar Python 29 Eki 2026 güncellendi 2 gün önce
ai-agents apple-silicon avx512 claude-code cpu-inference deepseek deepseek-v4 gguf
82
5
benmaster82/Pithos 28
A dependency-free C inference engine for Liquid AI's LFM2 / LFM2.5 models. Terminal chat, an OpenAI-compatible server, a web agent and a ter
2,5/gün ort. +3 son gün — commit C 27 Eki 2026 güncellendi bugün
ai-agent avx2 c coding-agent cross-platform gguf inference-engine lfm2
28
6
zhangshuibai/molten 24
Run a coding model at 300+ tok/s on your Mac. Native Rust + Metal inference engine for LiquidAI LFM2.5-24B-A2B (MLX 4-bit) on Apple Silicon.
1,8/gün ort. +1 son gün — commit Metal 25 Eki 2026 güncellendi 9 gün önce
apple-silicon gpu-kernels inference-engine inference-optimization lfm2 liquid-ai llama-cpp llm
24
7
skydiscover-ai/jitengine 35
LLM serving engines that SkyDiscover builds on the fly for specific AI models and hardware: Qwen3.6-35B-A3B, Qwen3.8-Flash-Next and Flint on
2,5/gün ort. — commit Python 24 Eki 2026 güncellendi 3 gün önce
apple-silicon coding-agents llm llm-inference llm-serving macos metal mixture-of-experts
35
8
wici-ai/inferssd 352
Run models larger than your RAM or VRAM by streaming MoE experts from SSD. Built on llama.cpp.
22/gün ort. +20 son gün 203 commit 4 yazar C++ 22 Eki 2026 güncellendi bugün
apple-silicon cuda gguf inference llama-cpp llm metal mixture-of-experts
352
9
Magnvs7/Promethevs 134
PROMETHEVS: run the full DeepSeek-V4.1-Flash-1017 (510 GB) on a gaming PC with 64 GB RAM and a 12 GB GPU. PROMETHEVS engine (compress-then-c
7,9/gün ort. +26 son gün 12 commit 1 yazar 21 Eki 2026 güncellendi 15 gün önce
consumer-hardware deepseek deepseek-v4 llm llm-inference local-inference local-llm mixture-of-experts
134
10
1picassoai/kraits 14
Run stock llama.cpp with the stock Unsloth Qwen3.6-35B-A3B 4-bit on a 16 GB Mac.
0,8/gün ort. +0 son gün — commit TypeScript 20 Eki 2026 güncellendi dün
apple-silicon ggml gguf llama-cpp llm llm-inference local-ai local-llm
14
11
BobClawblaw/deepseek-v41-flash-mlx-m5-ultra 14
DeepSeek-V4.1-Flash-1017 on one 256 GB M5 Ultra: 1M context, 70.7-104.9 tok/s decode with DSpark drafts, 1,739-1,867 tok/s cold prompts (8k-
0,7/gün ort. +0 son gün — commit Python 19 Eki 2026 güncellendi 15 gün önce
apple-silicon deepseek deepseek-v4 llm llm-inference long-context m5-ultra metal
14
12
surya-koritala/condor 34
Qwen3.8-Flash-Next (180B MoE) on one 12-16 GB NVIDIA GPU, Linux or Windows. Own engine, own compression.
1,4/gün ort. +0 son gün — commit Python 14 Eki 2026 güncellendi 17 gün önce
cuda inference linux llm local-llm mixture-of-experts nvidia qwen
34
13
Singularity-Compute/super-decider 19
Decisions, not essays: a choice, a yes or no, or a score with a calibrated probability. Open weights, Jev-compatible API, adaptive thinking.
0,8/gün ort. +0 son gün — commit Python 13 Eki 2026 güncellendi 12 gün önce
ai-agents calibration classification decision-model diffusion-llm guardrails jev llm-routing
19
14
RustingStudio/RustingStrata 119
Pure-Rust MoE inference engine for Qwen3.5/3.6/3.8 GGUF. CPU+GPU hybrid experts, NVRTC CUDA kernels, OpenAI-compatible server. 2x llama.cpp
4,3/gün ort. +36 son gün 443 commit 2 yazar Rust 10 Eki 2026 güncellendi bugün
cuda gguf inference llama-cpp llm local-llm mixture-of-experts moe
119
15
mw00/project-maya 2.611
Run GLM-5.3-Flash (321B), DeepSeek-V4.1-Flash (748B) and MiniMax-M3.1-Flash (428B) on your own GPUs - one engine, one dashboard, OpenAI- and
81,6/gün ort. +42 son gün 855 commit 15 yazar Cuda 06 Eki 2026 güncellendi bugün
anthropic-api cuda deepseek deepseek-v4 gguf glm glm-5 llm
2.611
16
DominikBucko/qwen38-flash-next-3090 13
Run Qwen3.8-Flash-Next on ONE RTX 3090 (24 GB) + 64 GB RAM: 128K context, up to 2,100 tok/s prefill, 43–51 tok/s decode. vLLM runtime with h
0,3/gün ort. +0 son gün — commit Python 30 Eyl 2026 güncellendi 30 Eyl 2026
128k-context 24gb-vram consumer-gpu cpu-offloading int4 kv-cache-quantization llm llm-inference
13
17
prabanta-dev/janas 654
37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cp
13,6/gün ort. +19 son gün 60 commit 1 yazar C 20 Eyl 2026 güncellendi bugün
avx2 c cpu-inference gguf inference-engine linux llm local-llm
654
18
DingoOz/llm-visuals 156
Live truecolor terminal dashboard for a locally running LLM: tok/s, TTFT, GPUs, context, MTP acceptance, layer and real expert routing (llam
2,7/gün ort. +0 son gün 96 commit 4 yazar Rust 11 Eyl 2026 güncellendi 4 gün önce
gpu-monitoring llama-cpp llm mixture-of-experts ratatui rust tui
156
19
fulvian/engraft-ngram 60
ENGRAFT: write facts into an LLM's n-gram (Engram) memory table as a removable overlay: 100 facts in a 9 MB file, no weights touched, verifi
1/gün ort. +0 son gün — commit Python 05 Eyl 2026 güncellendi 30 Eyl 2026
deepseek engram gguf interpretability knowledge-editing knowledge-injection llama-cpp llm
60
20
gpillon/ignis 56
OpenAI-compatible inference server for Qwen3.8 on one RTX 5090 / RTX PRO 6000: the 27B at ~170 tok/s (peaks ~370), Flash-Next (125B MoE) at
0,8/gün ort. +1 son gün — commit Rust 01 Eyl 2026 güncellendi bugün
blackwell coding-agents cuda inference-engine llm llm-inference local-llm mixture-of-experts
56
21
carloslfu/slotstream önem 64gpt6-luna uyum 38gpt6-luna 655
Slotstream, büyük bir MoE modelini SSD’den parça parça yükleyip yoğun kullanılan expert’leri bellekte önbelleğe alarak Apple Silicon Mac’lerde çalıştıran yerel bir inference engine’dir.
Neden yükseliyor · README, 48 GB M5 Pro üzerinde belirli koşullarda 15.86 tok/s ölçümü ve prompt işleme iyileştirmeleri yayımlıyor; son 28 günde 271 commit ve 5 farklı yazar da yoğun geliştirme faaliyetine işaret ediyor. Ancak proje yalnızca 30 günlük ve mevcut sürüm için nitelikli tam yanıt hızı ölçümünün hâlâ beklediği belirtiliyor.
9,2/gün ort. +3 son gün 278 commit 3 yazar Swift 28 Ağu 2026 güncellendi 3 gün önce
apple-silicon claude-code inference-engine llm llm-inference local-ai local-llm macos
655
22
ruashots/flashnext-2x3090 18
What I run on my inference server: Qwen3.8-Flash-Next on two RTX 3090s, and how I got here
0,3/gün ort. +1 son gün — commit Python 27 Ağu 2026 güncellendi 26 gün önce
benchmark llama-cpp llm-inference local-llm mixture-of-experts qwen rtx-3090
18
23
peonist-ai/crucible 13
MoE expert compression toolkit — prune, merge, and forge specialist models for local inference.
0,2/gün ort. +0 son gün — commit Python 27 Ağu 2026 güncellendi 4 gün önce
expert-pruning gguf llama-cpp llm llm-inference local-llm machine-learning mixture-of-experts
13
24
peonist-ai/halogen-flash-server önem 56gpt6-luna uyum 25gpt6-luna 1.248
halogen-flash-server, Qwen3.8-Flash-Next modelini AMD Strix Halo (gfx1151) donanımında çalıştırmak için özelleştirilmiş bir inference server'dır.
Neden yükseliyor · 32 günlük repo için 722 yıldız ve son 28 günde 99 commit dikkat çekiyor. README, belirli donanım üzerindeki ölçümleri ve ölçüm koşullarını açıklıyor; ancak performans iddiaları proje tarafından sunulmuş ve bağımsız doğrulama bilgisi burada yok. 99 commit'in tek yazardan gelmesi geliştirme hızını gösterirken bakım kapasitesine dair kanıtı sınırlıyor.
17,1/gün ort. +10 son gün 67 commit 1 yazar Shell 26 Ağu 2026 güncellendi dün
amd gfx1151 gpu-inference hip inference-engine llm llm-serving local-llm
1.248
25
siris9476/pulsarforge 23
0,3/gün ort. +0 son gün — commit C 25 Ağu 2026 güncellendi 10 Eyl 2026
avx2 c11 cpu-inference cpu-only gguf inference llama-cpp llm
23
26
tuff-app/TUFF 70
The local AI app built for the Mac. Run LLMs like Qwen, Gemma and GPT-OSS on Apple Silicon, including models far bigger than your Mac's memo
0,8/gün ort. +0 son gün — commit Swift 16 Ağu 2026 güncellendi 5 gün önce
ai-chat apple-silicon gemma gpt-oss llm llm-inference local-ai local-ai-models
70
27
shyringo/deepseek-v4-flash-0731-in-c 28
Run the native 284B-A13B DeepSeek-V4-Flash-0731 LLM locally on one laptop CPU: pure C, 8 GB RAM minimum, no GPU, best TPOT 0.892 s/token, re
0,3/gün ort. +0 son gün — commit C 16 Ağu 2026 güncellendi 09 Eyl 2026
c deepseek deepseek-v4 inference-engine llm llm-inference local-llm mixture-of-experts
28
28
WestWaters/pollard-weights 33
A tiered-memory system design for workloads that don't fit in RAM: measure the working set, pin the hot tier, stream the cold tier from flas
0,4/gün ort. +0 son gün — commit Python 07 Ağu 2026 güncellendi 3 gün önce
diagnostics exl3 gguf gptq llama-cpp llm low-resource mac
33
29
JustVugg/lumabri 213
Run large AI models across your computers and servers. Powered by Colibri. Pure C.
2,2/gün ort. +0 son gün — commit C 04 Ağu 2026 güncellendi bugün
c cluster-computing colibri distributed-inference llm-inference local-ai mixture-of-experts peer-to-peer
213
30
rayl15/Godwit 25
Run GPT-OSS-120B on a 16 GB Mac. Streams mixture-of-experts weights from SSD so only the 2 GiB trunk stays resident. Swift + Metal, no depen
0,3/gün ort. +0 son gün — commit Swift 04 Ağu 2026 güncellendi 07 Ağu 2026
apple-silicon gpt-oss inference llm local-llm macos metal mixture-of-experts
25
31
FareedKhan-dev/kimi-k3-in-c önem 38gpt6-luna uyum 20gpt6-luna 11.001
kimi-k3-in-c, Kimi K3 modelinde inference çalıştırmayı hedefleyen bağımlılıksız bir C99 motorudur.
Neden yükseliyor · 57 günde 8.683 yıldız alması ve Kimi K3'ü 8 GB RAM'li CPU'da çalıştırma iddiası dikkat çekiyor; ancak son 28 gündeki 12 commit'in tek yazardan gelmesi, ilginin bağımsız teknik doğrulama veya geniş katkıcı tabanıyla desteklendiğini göstermiyor.
112,3/gün ort. +234 son gün 0 commit 0 yazar C 01 Ağu 2026 güncellendi 02 Eki 2026
avx2 c99 cpu-inference deep-learning from-scratch inference-engine kimi-k3 linear-attention
11.001
32
melonmath/microkimi 26
The Kimi K3 and DeepSeek-V4-Flash-0731 architectures, reimplemented from scratch in pure Rust (zero dependencies) and verified 1:1 against t
0,3/gün ort. +0 son gün — commit Rust 31 Tem 2026 güncellendi 19 Ağu 2026
from-scratch inference-engine kimi linear-attention llm machine-learning mixture-of-experts moe
26
33
pablo-reyes8/kimi-k3-pytorch 14
Build and train a mini Kimi K3 from scratch — a pure-PyTorch playground from a custom 88M model to 2T-scale LLM profiles.
0,1/gün ort. +0 son gün — commit Python 28 Tem 2026 güncellendi 5 gün önce
deep-learning generative-ai kimi-k3 kv-cache llm llm-training llm-training-from-scratch machine-learning
14
34
Tylogi/TyloQuant 95
Get more intelligence from every bit. Better quantization formats and smarter calibration let larger, stronger models run smoothly on the ha
0,9/gün ort. +0 son gün — commit C++ 24 Tem 2026 güncellendi 2 gün önce
apple-silicon cuda deepseek gemma gguf llm llm-inference llm-quantization
95
35
prayangshuuu/hummingbird 62
hummingbird is a lightweight, zero dependency runtime for massive open source Mixture of Experts (MoE) language models. It unifies SSD, RAM,
0,6/gün ort. +0 son gün — commit C 21 Tem 2026 güncellendi 31 Tem 2026
ai-inference artificial-intelligence c17 cross-platform deepseek glm gpt-oss high-performance-computing
62
36
FareedKhan-dev/glm-5.2-in-c 75
GLM-5.2, a 744 billion parameter mixture of experts model, in a pure C inference engine: quantized to int4, experts streamed from disk, depl
0,7/gün ort. +0 son gün — commit C 15 Tem 2026 güncellendi 15 Tem 2026
c consumer-hardware cpu-inference cuda deep-learning edge-ai glm inference-engine
75
37
drmihirbrahme/midge 13
Tiny engine, immense models — run large MoE LLMs (gpt-oss, Mixtral, Qwen3-MoE) on ordinary machines by streaming experts from disk. OpenAI-c
0,1/gün ort. +0 son gün — commit Python 14 Tem 2026 güncellendi 21 Ağu 2026
apple-silicon cpu-inference gpt-oss llm low-memory mixture-of-experts mlx moe
13
38
giannisanni/pulsar 234
SSD-streaming inference engine for giant MoE models (Rust + CUDA). GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs
2/gün ort. +0 son gün 0 commit 0 yazar Rust 13 Tem 2026 güncellendi 01 Eyl 2026
cuda gguf glm inference-engine llm local-llm mixture-of-experts moe
234
39
Helldez/BigMoeOnEdge 712
Run MoE models bigger than your RAM. Frontier-size MoE on a phone, laptop, edge. Lossless, on stock llama.cpp. An adaptive streaming optimiz
5,9/gün ort. +3 son gün 162 commit 3 yazar C++ 10 Tem 2026 güncellendi bugün
android cpp edge-ai gemma gguf inference llama-cpp llm
712
40
Joakimpalm-Zen/xyntetik-runner 30
Local inference you can prove. A single-binary GGUF engine in C: serves, scores, and trains LoRA through the quantized weights it deploys, b
0,3/gün ort. +0 son gün — commit C 10 Tem 2026 güncellendi bugün
agents c constrained-decoding cuda edge-ai fine-tuning gguf inference
30
41
AMD-AGI/Instella-MoE 80
Instella MoE is a fully open state-of-the-art Mixture-of-Experts language model, trained end-to-end on AMD Instinct™ GPUs.
0,7/gün ort. +0 son gün — commit Python 07 Tem 2026 güncellendi 02 Eyl 2026
amd instella llm mixture-of-experts rocm
80
42
arizqi/cpubrrr 57
Frontier-class LLM inference on a laptop CPU — gpt-oss:20b at ~110 tok/s on Apple M4 Max, 7.5x llama.cpp, no GPU. From-scratch NEON/SME kern
0,5/gün ort. +0 son gün — commit Rust 04 Tem 2026 güncellendi 11 Ağu 2026
apple-silicon cpu gpt-oss inference llm mixture-of-experts neon quantization
57
43
AarambhDevHub/aarambh-studio 112
🦀 Decoder-only LLM built from scratch in pure Rust using Candle — no Python, no PyTorch. Gated DeltaNet + sparse attention, fine-grained MoE
0,8/gün ort. +0 son gün — commit Rust 24 Haz 2026 güncellendi 21 Ağu 2026
ai-agents candle decoder-only deep-learning dora from-scratch gguf grpo
112
44
gittensor-ai-lab/sparkinfer 93
Fastest MoE/LLM inference runtime for consumer and edge Blackwell GPUs. SN74 on Gittensor.
0,7/gün ort. +0 son gün — commit C++ 22 Haz 2026 güncellendi 27 gün önce
agentic-ai bittensor blackwell cuda edge-ai gittensor gpu gpu-programming
93
45
DaoyuanLi2816/tracedistill 85
Distill teacher chains-of-thought into a LoRA adapter via a strict boxed-answer format contract + two-phase Train→Nudge (silver-medal NVIDIA
0,6/gün ort. +0 son gün — commit Python 22 Haz 2026 güncellendi 27 Tem 2026
chain-of-thought fine-tuning kaggle llm lora mamba mixture-of-experts nemotron
85
46
Dewaldnel11/LightLX 18
Run models too big for RAM on Apple Silicon — any Hugging Face model, dense or MoE — by streaming weights from disk
0,1/gün ort. +0 son gün — commit Python 22 Haz 2026 güncellendi 15 Eyl 2026
apple-silicon inference llm local-llm macos mixture-of-experts mlx mlx-lm
18
47
hec-ovi/gamentic 29
Self-hosted, browser-based AI dungeon RPG, fully local on an AMD Strix Halo APU: narrator + per-character agents on one local LLM (Gemma 4 2
0,2/gün ort. +1 son gün — commit Python 09 Haz 2026 güncellendi 26 Tem 2026
agentic ai-dungeon ai-rpg amd-strix-halo comfyui docker-compose fastapi flux
29
48
Achyuthan-S/moe-bench 16
Open benchmark for expert collapse and routing efficiency in sparse MoE LLMs
0,1/gün ort. +0 son gün — commit Python 08 Haz 2026 güncellendi 17 gün önce
benchmarking language-models llm machine-learning mixture-of-experts moe nlp pytorch
16
49
Abineshabee/MoEWatch 17
👻 MoEWatch: A lightweight diagnostic and audit library for Mixture-of-Experts (MoE) models in HuggingFace Transformers — the pytest for MoE
0,1/gün ort. +0 son gün — commit Python 02 Haz 2026 güncellendi 17 gün önce
deep-learning deepseek expert-collapse huggingface llm llmops load-balancing machine-learning
17
50
egesabanci/reap-mlx 40
MLX-compatible REAP for pruning MoE models on Apple Silicon
0,2/gün ort. -1 son gün — commit Python 29 May 2026 güncellendi 3 gün önce
apple-silicon llm mixture-of-experts mlx mlx-lm moe model-compression reap
40