What worked, what didn’t, and what the numbers actually mean.
I’m Gökhan Çava, an independent developer. I published LCGA Local, a source-only research package for local AI on small GPUs. This is the short technical story behind the release, including the parts that did not work.
The test machine is an RTX 3050 Laptop GPU with 6 GB of VRAM. That is enough for a small model, but not enough to load a 4B model in bf16 alongside a normal KV cache and a desktop session. The release contains two different answers to that constraint.
The serving path uses Qwen3-4B-Instruct-2507 in int8 with layer streaming and one bf16 layer slot. It is deliberately boring engineering: keep the resident working set below the limit, move weights in as needed, and avoid allocating the whole model at once.
Measured on the 6 GB RTX 3050 Laptop:
This is not fast by datacenter standards. It is useful for local chat, experimentation, and checking whether a 4B model fits at all.
The RAG layer is CPU-only BM25 over a 326-document Wikipedia-derived corpus (4.57 MiB). It returns a passage and a citation, and it declines when the corpus does not support an answer.
On the internal exp48 evaluation (50 covered prompts plus 20 controls):
The one keyword miss is not a factual failure: the answer used the correct Turkish synonym "Büyük Okyanus" instead of the literal "Pasifik". The baseline keyword score was 42/50.
The important limit: this is a small, hand-built evaluation, not a public benchmark. The manual review was performed in the session by the agent and the developer, not by an independent human reviewer. The repository says so in "results/exp48_manual_review.md".
The M5 experiment asks a different question: can the same laptop update the full parameter set of a 3B model, not a LoRA adapter?
Using block streaming and Adafactor, M5 updates all 3.09B parameters on the 6 GB RTX 3050 Laptop: This is a demo, not a convergence study. A 150-step run is enough to show that the memory path works; it is not enough to claim a useful fine-tune. The full report is in "M5_RAPOR.md".
The repository also contains the LCGA block-sparse attention experiments. The current quality result is negative: in the measured setup, dense attention reached 12.46 PPL and the sparse variant reached 16.56 PPL. The release includes that result rather than hiding it. The serving and training work is independently useful; the sparse-attention idea still needs more work.
Model weights, checkpoints, caches, and "tools/llama.cpp" are excluded. Qwen3-4B must be downloaded separately. The code is MIT-licensed; the RAG corpus is CC BY-SA 4.0 with attribution in "data/rag/README.md".
If you run this on a smaller GPU, or if you find a mistake in the evaluation, open an issue. Negative results and corrections are welcome. Disclosure: This post was drafted by an AI agent from the project's reports; the developer authorized publication. All measurements are from the repository.