Download

Get PorchLM

One GGUF file. Runs in LM Studio, Ollama, llama.cpp, or a browser tab — nothing you type ever leaves your machine.

01 / Files

These builds are kept loaded and ready — pick one to try in chat straight away, or download the file.

PorchLM V10
Ready now
241 MB
An experimental model of PorchLM trained on 1 billion tokens. Not recommended for production use. Designed for testing and development purposes.
PorchLM-F16-V10.gguf
PorchLM V11
Ready nowRecommended
241 MB
The first PorchLM model that has search capabilities built in. It is a smaller model that is designed to be used for search and retrieval tasks.
PorchLM-F16-V11.gguf
PorchLM V3
Ready nowRecommended
115 MB
The most efficient and usable model of PorchLM as of the 30th of July, 2026. More sophisticated than all of the previous versions. Trained on 3 billion tokens. It is a 50M parameter model, which is a good balance between performance and efficiency.
PorchLM-F16-V3.gguf
PorchLM V9
Ready now
241 MB
An experimental model of PorchLM trained on 1 billion tokens. Not recommended for production use. Designed for testing and development purposes.
PorchLM-F16-V9.gguf
7 older builds available
Every version ever published, searchable. These load on demand rather than staying in memory.
See all models →

Served straight from this machine. See the changelog for what changed in each release.

For scale
PorchLM (Q4_K_M) ~6 MB
A phone photo ~4 MB
Llama-3-8B (Q4) ~4.7 GB
02 / Install
Browser

An in-browser demo (coming soon) will let you drop a .gguf file directly onto the page — no install, cached locally after the first load.

LM Studio

Place the file at ~/.lmstudio/models/PorchLM/PorchLM-F16-V10.gguf — the two-level publisher/model/file.gguf layout is required or it won’t appear.

Ollama

Create a Modelfile:

FROM ./PorchLM-F16-V10.gguf
PARAMETER temperature 0.9
PARAMETER top_k 40
PARAMETER top_p 0.95
PARAMETER repeat_penalty 1.15
PARAMETER num_ctx 1024

Then:

ollama create porchlm -f Modelfile && ollama run porchlm
llama.cpp
llama-cli -m PorchLM-F16-V10.gguf -c 1024 -cnv
Quantizing smaller
llama-quantize PorchLM-F16-V10.gguf PorchLM-Q4_K_M.gguf Q4_K_M

Very small models lose more to aggressive quantization than large ones — if quality drops noticeably, Q8_0 is the safer trade.

03 / Sampling defaults

Already embedded in the GGUF’s chat template — most tools pick these up automatically, no manual configuration needed.

Temperature 0.9
Top-k 40
Top-p 0.95
Repeat penalty 1.15
Context (num_ctx) 1024