Differences

This shows you the differences between two versions of the page.

--- infrastructure:llama-cpp-on-dreamflasher [2026/10/10 22:09] – created diamond
+++ infrastructure:llama-cpp-on-dreamflasher [2026/10/10 22:32] (current) – diamond
@@ Line 2: / Line 2: @@
 - admin: @diamond
-- address: https://llama-cpp.dreamflasher.dma.space/
+- addresses:
+  - external: https://llama-cpp.dreamflasher.dma.space/ (login required)
+  - internal: http://dreamflasher.int.dma.computer:12345/
+- deployment: https://codeberg.org/dma/infra/src/commit/main/nixos/machines/dreamflasher/_llama-server.nix
+## Inference Engine
+- tool: llama.cpp
+- model name: Qwen/Qwen3.6-35B-A3B (`Qwen3.6-35B-A3B-UD-IQ2_XXS.gguf`)
+- model size: 10.01 GB
+- context size: 131,072 tokens
+- training context: 262,144 tokens
+- parameters: 34.66B
+- embedding size: 2,048
+- vocabulary size: 248,320 tokens
+- parallel slots: 4
 ## Usage Guidelines
@@ Line 10: / Line 25: @@
 - TODO: expand
-I'm hosting this to allow people to have something to play and learn with, without needing the money to pay for an AI subscription service. When using this, please keep that in mind! It is a **shared** service, so don't hog all of its resources for yourself if possible.
+I'm hosting this to allow people to have something to play and learn with, without needing the money to pay for an AI subscription service. When using this, please keep that in mind! It is a **shared** service, so don't hog all of its resources for yourself.
+> **Note**: We don't store chat history. Inference is done completely offline.