Python 排行榜

vLLM Qwen 3.6-27B (AWQ-INT4) + DFlash speculative decoding on AMD Strix Halo (gfx1151 iGPU, 128 GB UMA, ROCm 7.13). 24.8 t/s single-stream, vision, tool calling, 256K context, OpenAI-compatible, Docker. Matches DGX Spark FP8+DFlash+MTP at a third of the cost. No CUDA.

AI editor's assistant that learns your style from your own cuts. Turn interview footage into rough cuts for Final Cut, Premiere, and Resolve. Mac-native. Local-first.

Ollama with intel (i)GPU acceleration in docker and benchmark

Ollama Model Test - Figure out the best model for the task

The new DARNA. HI Version2.5. Aspires to be your self hosted Personal Health concierge on Linux. pwd 'health'

Always-current official Claude docs inside Claude Code. A 3 MB metadata plugin: live on-demand fetching from Anthropic's servers, zero content redistributed. Auto-discovery skill, instant search, interactive HTML courses, and doc changelog reports.

86 agent-executable skill packs converted from RefoundAI’s Lenny skills (unofficial). Works with Codex + Claude Code.

CRISP — The missing layer between vibe coding and building something people actually want.

[L0 CONSTITUTION] arifOS — constitutional MCP kernel. Law, identity, F1–F13, VAULT999. Judges but never executes. DITEMPA BUKAN DIBERI.

Train Llama 3 models from scratch. Any scale, any personality. By Arianna Method.

A lightweight but powerful library to build token indices for NLP tasks, compatible with major Deep Learning frameworks like PyTorch and Tensorflow.

Physics-inspired transformer modules based on mean-field dynamics of vector-spin models in JAX

Fully Configurable RAG Pipeline for Bengali Language RAG Applications. Supports both Local and Huggingface Models, Built with Langchain.

Pre-training Language Models for Japanese

[TITS] reliable dense matching based point cloud registration for autonomous driving

Train GEMMA on TPU/GPU! (Codebase for training Gemma-Ko Series)

Interpretable Pre-Trained Transformers for Heart Time-Series Data

Human parsing model for fashion and virtual try-on applications

Toy-scale unified multimodal model experiments — encoder-free understanding & generation with Mixture-of-Transformers on MLX/Apple Silicon

[⛔️ DEPRECATED] Friendli: the fastest serving engine for generative AI