[
  {
    "id": "llama-cpp-gh-1",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "llama cli -hf ggml-org/Qwen3.5-0.8B-GGUF",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-2",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "llama serve -hf ggml-org/Qwen3.5-0.8B-GGUF",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-3",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Built-in web UI against `llama serve` running Qwen 3.6",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-4",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "VLM session with `llama cli`",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-5",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Plain C/C++ implementation without any dependencies",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-6",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-7",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "AVX, AVX2, AVX512 and AMX support for x86 architectures",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-8",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-9",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP and Moore Threads GPUs via MUSA)",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-10",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-11",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Run with Docker - see our [Docker documentation](docs/docker.md)",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-12",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Download pre-built binaries from the [releases page](https://github.com/ggml-org/llama.cpp/releases)",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-13",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "[GBNF grammars](grammars/README.md)",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-gh-14",
    "tier": "github",
    "url": "https://github.com/ggml-org/llama.cpp",
    "excerpt": "Contributors can open PRs\n- Collaborators will be invited based on contributions",
    "fetchedAt": "2026-08-29T01:24:37.411Z"
  },
  {
    "id": "llama-cpp-comm-1",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35393284",
    "excerpt": "Author explains loading time performance is a huge win for usability, but the RAM usage reduction (mmap change) lacks a compelling theory yet, urging skepticism despite excitement.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-2",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35393284",
    "excerpt": "User reports on 16GiB RAM machine running the 30B model: performance was extremely slow, process wasn't CPU-limited, and it was clearly swapping a lot.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-3",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35393284",
    "excerpt": "\"Less memory than most Electron apps!\" — praising the dramatic RAM reduction achieved.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-4",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35100086",
    "excerpt": "User got llama.cpp working on M1 iMac trivially easily; performance was very impressive even without using Apple's neural compute hardware, and output lacked political correctness conditioning.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-5",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35100086",
    "excerpt": "On 32GB M1 Max, user reports getting 56.38 ms per token on the 7B model, calling it 'Very usable!'",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-6",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35100086",
    "excerpt": "User ran the 7B model on a 64GB M1 Max Macbook Pro, noting predict time of ~83ms per token and that it worked tremendously fast.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-7",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=35100086",
    "excerpt": "Praise for the minimal, dependency-free implementation: 'awesome being able to experiment with complex models without needing a billion python/c/cpp dependencies.'",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-8",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=36304143",
    "excerpt": "\"llama.cpp is great. It started off as CPU-only solution and now looks like it wants to support any computation device it can... totally detached from Python ML ecosystem and also popular.\"",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-9",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=36304143",
    "excerpt": "User using llama.cpp with python wrappers found the speed increase from CUDA acceleration great, but noted it seemed limited to a max of 40 N_GPU_LAYERS before updating.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-10",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=36304143",
    "excerpt": "Comment on CUDA GPU acceleration: only about a 2x speedup on a top-end 4090 card and limited to one CPU core, surprising given expectations, leaving performance on the table with modern multi-core CPUs.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-11",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=37140013",
    "excerpt": "Detailed breakdown: prompt ingestion runs fast in batches unlike token generation; llama.cpp's Q6_K quantization nearly matches FP16 perplexity while being much smaller; model splitting across GPU/CPU lets users run 70B models on desktop or 33B on a 16GB RAM laptop that otherwise couldn't run them.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-12",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=37140013",
    "excerpt": "User reports running llama.cpp on a 4-core i7 with 64GB RAM: ~0.5 tokens/s for 70B model, ~1 token/s for 30B model, expressing shock that such large models could run on a desktop CPU at all.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-13",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=43943047",
    "excerpt": "User found the vision feature 'works super well' after compiling from source, using llama-mtmd-cli with quantized multimodal models like Gemma-3, loading images via '/image image.png' in chat.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-14",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=43943047",
    "excerpt": "User used llama.cpp's vision support with Gemma3 4b to generate keywords/descriptions for trip photos, including basic OCR and context clues to identify photo locations, calling it 'very nice for something self-hosted.'",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-15",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=43943047",
    "excerpt": "Benchmark on M1 64GB Macbook Pro with gemma-3-4b-it: 25t/s prompt processing, 63t/s token generation, ~15 sec per image regardless of image size, with decent output quality.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-16",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=43943047",
    "excerpt": "User noted it was 'really sad' when vision support was removed from llama.cpp previously, and expressed thanks that it's been restored.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-comm-17",
    "tier": "community",
    "url": "https://news.ycombinator.com/item?id=43943047",
    "excerpt": "macOS user needed to run 'sudo xattr -rd com.apple.quarantine' on the compiled binaries to get the official release working after unzipping.",
    "fetchedAt": "2026-08-29T01:36:19.529Z"
  },
  {
    "id": "llama-cpp-probe-1",
    "tier": "probe",
    "url": "https://github.com/llms.txt",
    "excerpt": "PROBE llms.txt: HTTP 200 at https://github.com/llms.txt # GitHub\n\n> GitHub is a developer platform for building, shipping, and maintaining software. It provides cloud-based Git",
    "fetchedAt": "2026-08-29T01:38:26.058Z"
  }
]