{"slug": "neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks", "title": "NeuronDump: Open a local LLM as a 3D graph, cut parts, see what breaks", "summary": "NeuronDump, a Rust-based open-source tool from developer aymen94, reads GGUF model files used by Ollama and llama.cpp and renders a local LLM as an explorable 3D graph, letting users cut layers, attention heads or token ranges and see the impact before writing a smaller GGUF. The tool reports parameters removed, new file size, what gets disconnected, whether llama.cpp still loads the file, and a verdict, refusing dangerous cuts unless forced; an optional trace feature runs the cut model in memory alongside the original to compare next tokens, KL divergence, logit lens per layer and perplexity. Ready-made Windows and Linux builds are on the GitHub releases page, and building from source requires Rust 1.88 or newer.", "body_md": "Look inside any local LLM, find the parts it doesn't need, and make it smaller.\n\nNeuronDump reads **GGUF** model files (the format used by **Ollama** and **llama.cpp**) and\nshows what is inside: layers, attention heads, vocabulary, chat template, weight statistics.\nIt can then remove unused parts and write a smaller GGUF that still runs in Ollama and llama.cpp.\n\nIt works with any model. It reads each file's own metadata; nothing is hard-coded per model.\n\n**Download:** ready-made Windows and Linux builds are on the\n**[releases page](https://github.com/aymen94/NeuronDump/releases)**. Unzip and run `neurondump list`.\nDownload steps and checks: **[install guide](https://github.com/aymen94/NeuronDump/blob/main/docs/install.md#download-a-ready-made-build)**.\n\n**Build from source:** you need [Git](https://git-scm.com) and [Rust](https://rustup.rs) 1.88 or newer.\nStep-by-step instructions: **[Windows and Linux install guide](https://github.com/aymen94/NeuronDump/blob/main/docs/install.md)**.\n\n**Windows** (PowerShell):\n\n```\nwinget install Git.Git\nwinget install Rustlang.Rustup              # then reopen PowerShell\ngit clone https://github.com/aymen94/NeuronDump\ncd NeuronDump\ncargo install --path crates/neurondump\nneurondump list\n```\n\n**Linux**:\n\n```\nsudo apt install build-essential git curl   # or dnf / pacman\ncurl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh && source \"$HOME/.cargo/env\"\ngit clone https://github.com/aymen94/NeuronDump\ncd NeuronDump\ncargo install --path crates/neurondump\nneurondump list\n```\n\nAdd `--features trace` to `cargo install` to also run models and watch each layer.\n\n```\nneurondump list                          # every local model\nneurondump inspect llama3.2:1b           # what's inside a model\nneurondump vocab llama3.2:1b --search hello\nneurondump template llama3.2:1b          # the chat template as a tree\nneurondump analyze llama3.2:1b --out plan.json\nneurondump slim llama3.2:1b --plan plan.json --only vocab --out small.gguf\nneurondump web --open                    # the same, in your browser\n```\n\nRun `neurondump` with no arguments to choose between the interactive CLI and the web interface.\n\nThe web interface draws the model as a graph you can open up, in 3D (tower, city or force layout) or 2D. Everything is bundled, so it works offline.\n\n- \n**Render inside any part:**\n  - double-click a layer to see its attention and feed-forward blocks;\n  - double-click attention to see its KV groups, then its individual heads;\n  - feed-forward opens into neuron buckets, and MoE into experts.\n- \n**Inspect:**\n  - hover any part for its size and flag;\n  - click it for its path, contents, connections, weight statistics, activations, importance,\nand whether it can be removed (`safe` ,`risky` or`breaks_model` ) and why.\n- \n**Search inside:** type a tensor name like`blk.3.ffn_down` . Closed layers open to show it.\n- \n**Cut (edit):** press**Cut this** on any part (a layer, a block, a head, a token range) to add\nit to a cut selection. Cut parts get a red outline.\n- \n**Direct impact, instantly:** the panel shows:\n  - parameters removed or zeroed, and the new file size;\n  - what gets disconnected and what depends on the cut;\n  - whether llama.cpp still loads the file;\n  - a verdict.\n Dangerous cuts are refused unless you force them.\n- \n**Simulated impact** (`--features trace` ): the cut model runs in memory next to the original.\nYou compare their next tokens, KL divergence, generated text, the logit lens per layer and\nperplexity. Nothing is written.\n- \n**Sweep:** cuts each layer (or head) in turn, ranks them by damage, and colors the graph with it.\n- \n**Save as plan → Slim:** writes the smaller GGUF and an Ollama Modelfile.\n\nThe same works from the command line:\n\n```\nneurondump graph llama3.2:1b --expand layer.7.attn                  # inside a layer's attention\nneurondump node llama3.2:1b layer.7.attn.head.3                     # inspect one head\nneurondump ablate llama3.2:1b --cut layer.7 --static                # direct impact\nneurondump ablate llama3.2:1b --cut layer.7 --prompt \"The capital of Italy is\"   # simulated impact\nneurondump sweep llama3.2:1b --corpus texts.txt --save-plan plan.json --top 2\nneurondump slim llama3.2:1b --plan plan.json --out llama-14.gguf\n```\n\nDetails: [Graph and ablation](https://github.com/aymen94/NeuronDump/blob/main/docs/graph-and-ablation.md).\n\n| Page | What it covers | \n|---|---|\n| [Install and run](https://github.com/aymen94/NeuronDump/blob/main/docs/install.md) | Windows and Linux, step by step | \n| [Getting started](https://github.com/aymen94/NeuronDump/blob/main/docs/getting-started.md) | find your models, first commands | \n| [How it works](https://github.com/aymen94/NeuronDump/blob/main/docs/internals.md) | what NeuronDump reads, and how | \n| [CLI reference](https://github.com/aymen94/NeuronDump/blob/main/docs/cli.md) | every command and flag, and the REPL | \n| [Web interface](https://github.com/aymen94/NeuronDump/blob/main/docs/web.md) | the browser UI and the HTTP API | \n| [Analyze and slim](https://github.com/aymen94/NeuronDump/blob/main/docs/analyze-and-slim.md) | making models smaller, safely | \n| [Trace](https://github.com/aymen94/NeuronDump/blob/main/docs/trace.md) | running a model and looking at each layer | \n| [Graph and ablation](https://github.com/aymen94/NeuronDump/blob/main/docs/graph-and-ablation.md) | the 3D / 2D graph: look inside layers, inspect, search, cut, direct and simulated impact | \n| [What's inside real models](https://github.com/aymen94/NeuronDump/blob/main/docs/models.md) | Qwen3 8B vs DeepSeek-R1, and Qwen 3.8 | \n| [Building](https://github.com/aymen94/NeuronDump/blob/main/docs/building.md) | build options and Windows notes | \n| [Manual test plan](https://github.com/aymen94/NeuronDump/blob/main/TEST.md) | step-by-step checks | \n\nA short illustrated explanation is on the [website](https://aymen94.github.io/NeuronDump/)\n(source: [`docs/index.html`](https://github.com/aymen94/NeuronDump/blob/main/docs/index.html)).\n\nEarly version, tested on Windows with Ollama models. Known gaps are listed at the end of\n[TEST.md](https://github.com/aymen94/NeuronDump/blob/main/TEST.md). Bug reports and ideas: [issues](https://github.com/aymen94/NeuronDump/issues).\n\nLicensed under the [MIT License](https://github.com/aymen94/NeuronDump/blob/main/LICENSE).", "url": "https://wpnews.pro/news/neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks", "canonical_source": "https://github.com/aymen94/NeuronDump", "published_at": "2026-10-25 19:45:09+00:00", "updated_at": "2026-10-25 20:12:44.842524+00:00", "lang": "en", "topics": ["large-language-models", "ai-tools", "developer-tools", "machine-learning"], "entities": ["NeuronDump", "aymen94", "GGUF", "Ollama", "llama.cpp", "Rust", "GitHub"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks", "markdown": "https://wpnews.pro/news/neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks.md", "text": "https://wpnews.pro/news/neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks.txt", "jsonld": "https://wpnews.pro/news/neurondump-open-a-local-llm-as-a-3d-graph-cut-parts-see-what-breaks.jsonld"}}