chore(llm): clean up models & fix llama-cpp issue

modules/home/programs/terminal/nvim/config/lua/llm.lua

@@ -1,9 +1,10 @@
 local llm_endpoint = "https://llm-api.va.reichard.io"
-local llm_model = "qwen3-coder-30b-instruct"
+local llm_assistant_model = "gpt-oss-20b-thinking"
+local llm_infill_model = "qwen2.5-coder-3b-instruct"
 
 -- Default Llama - Toggle Llama & Copilot
-vim.g.copilot_filetypes = { ["*"] = false }
-local current_mode = "llama"
+-- vim.g.copilot_filetypes = { ["*"] = false }
+local current_mode = "copilot"
 local function toggle_llm_fim_provider()
 	if current_mode == "llama" then
 		vim.g.copilot_filetypes = { ["*"] = true }
@@ -24,8 +25,10 @@ vim.keymap.set("n", "<leader>cf", toggle_llm_fim_provider, { desc = "Toggle FIM
 -- Configure LLama LLM FIM
 vim.g.llama_config = {
 	endpoint = llm_endpoint .. "/infill",
-	model = llm_model,
-	n_predict = 1024,
+	model = llm_infill_model,
+	n_predict = 2048,
+	ring_n_chunks = 32,
+	enable_at_startup = false,
 }
 
 -- Configure Code Companion
@@ -39,7 +42,7 @@ require("codecompanion").setup({
 				return require("codecompanion.adapters").extend("openai_compatible", {
 					name = "llama-swap",
 					formatted_name = "LlamaSwap",
-					schema = { model = { default = llm_model } },
+					schema = { model = { default = llm_assistant_model } },
 					env = { url = llm_endpoint },
 				})
 			end,