# MarvinLabs

> Public dataset of measured local-LLM fine-tuning and inference findings: a 17-version LoRA fine-tune program of a 4B command router (unsloth/gemma-3-4b-it) on 8 GB consumer NVIDIA cards, September 2026, plus cross-GPU memory probes on an AMD RX 7900 XTX, October 2026. Every number is measured and dated. Each page has a Markdown twin; the raw data is JSON.

## Pages

- [MarvinLabs overview](https://marvinlabs.org/index.md): Public dataset of measured local-LLM fine-tuning and inference findings: a 17-version LoRA program on a 4B router, trained on 8 GB NVIDIA cards.
- [Version ledger](https://marvinlabs.org/versions.md): The ledger of every trained version: config, training stats and measured results, plus the test-set eras.
- [VRAM measurements](https://marvinlabs.org/vram.md): Every GPU memory measurement from the program: peak MB by rank, dropout, DoRA, bf16 and sequence length on an 8 GB NVIDIA card and a 24 GB AMD card, plus the probe calibration.
- [Inference speed](https://marvinlabs.org/inference.md): Ollama load time, prompt speed, generation speed and resident VRAM for seven small models on an AMD RX 7900 XTX.
- [Untuned baselines](https://marvinlabs.org/baselines.md): The untuned model tournament and the untuned 12B scored on each test-set era.
- [Findings](https://marvinlabs.org/findings.md): Every finding with its evidence and measured date, plus the methods tried and closed and the ideas left untried.
- [Gotchas](https://marvinlabs.org/gotchas.md): Environment and tooling failures hit during the program, with symptom, cause, fix and versions.
- [Training recipe](https://marvinlabs.org/recipe.md): The shared training recipe: model, quantization, LoRA settings, hyperparameters, memory fixes, export path and release rule.
- [About: method and scope](https://marvinlabs.org/about.md): How the numbers were measured, what the terms mean, what is not published, and how to report a correction.

## Data

- [baselines.json](https://marvinlabs.org/data/baselines.json): Untuned (off-the-shelf) models, measured before and during the fine-tune program through Ollama. 'routing' = the production command-routing prompt (~5.6k tokens); 'content' = 11 generation workloads...
- [findings.json](https://marvinlabs.org/data/findings.json): Lessons from 17+ LoRA fine-tunes of gemma-3-4b-it as a command router on one 8 GB consumer GPU. Each claim is limited to this task, this base model and this recipe. Accuracy numbers are passed /...
- [gotchas.json](https://marvinlabs.org/data/gotchas.json): Environment and tooling failures hit during the program, with symptom, cause, fix and versions. Versions are those recorded at the time; see recipe.json for the final stack.
- [inference.json](https://marvinlabs.org/data/inference.json): Ollama inference speed for seven small models on one AMD Radeon RX 7900 XTX (24 GB, Windows 11, Ollama 0.30.10). Each model was unloaded first, then given synthetic prompts of about 1,550 and 4,550...
- [recipe.json](https://marvinlabs.org/data/recipe.json): The training recipe shared by every fresh-from-base version (v1-v3, v5-v17). v4 differs only in continuing from v3's adapter at 8 epochs. Values come from the training scripts and the run headers...
- [versions.json](https://marvinlabs.org/data/versions.json): LoRA fine-tunes of unsloth/gemma-3-4b-it as a closed-set command router (one JSON tool call per utterance), trained and served on 8 GB NVIDIA consumer cards, 13-25 Sep 2026.
- [vram.json](https://marvinlabs.org/data/vram.json): Every VRAM measurement from the program. The cross_gpu section adds one-step probes on an AMD RX 7900 XTX (24 GB) and a second NVIDIA card. Training GPU: one 8 GB NVIDIA consumer card (an RTX 3070 Ti...

## Optional

- [llms-full.txt](https://marvinlabs.org/llms-full.txt): every Markdown page in one file
- [sitemap.xml](https://marvinlabs.org/sitemap.xml): all URLs