llm, begin refactoring the script into an ai folder, also increase parallel.

local.yml

@@ -131,7 +131,7 @@
 
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/docker.yml
 
-            - ansible.builtin.include_tasks: tasks/workstation/linux/software/llm.yml
+            - ansible.builtin.include_tasks: tasks/workstation/linux/software/ai/llm.yml
 
           when: ansible_system == "Linux"
             and internet

tasks/workstation/linux/software/ai/llm.yml

@@ -66,7 +66,15 @@
         # Also setting this to a minimum of 3 for using Codium with extensions.
         # 20261014: Going down to 1 since only using CPUs, no GPUs. Otherwise
         # the models seem to take excessively long to answer even small things.
-        llm_parallel: 1 #"{{ [(ansible_processor_vcpus / 2) | round(0, method='ceil'), 3] | max | int }}"
+        #llm_parallel: 1 #"{{ [(ansible_processor_vcpus / 2) | round(0, method='ceil'), 3] | max | int }}"
+        # 20261027 Increased to 3 as a maximum.
+        # After more research, this is a more memory-based parameter. Basically
+        # the model allocates this number of KV caches, that way when it is
+        # running in multiple chats it does not need to drop context and reload.
+        # Still only uses X threads for all chats, but does not waste much time.
+        # So if set to 3, and doing 64K context, it really allocates 192k in RAM
+        # and supposedly if only doing 1 chat the model can use it all.
+        llm_parallel: "{{ [(ansible_memtotal_mb / 8000) | round(0, method='floor'), 3] | min | int }}"
         # Most LLMs take a coefficient of about 1.2 to 1.5 of their file size in
         #    RAM, which is usually less than guessing a usage of weight*1000.
         # Example:
@@ -272,7 +280,7 @@
               ### Resource Usage ###
               export OLLAMA_KEEP_ALIVE=-1 # default is 5m. Good frontends keep the model alive as long as it needs to be, this is not a source of timeout.
               export OLLAMA_MAX_LOADED_MODELS={{ llm_concurrency }} # default is 3 if using CPUs, or 3 * NumGPUs
-              export OLLAMA_NUM_PARALLEL={{ llm_parallel }} # default is 1 and applies per-model (unconfirmed)
+              export OLLAMA_NUM_PARALLEL={{ llm_parallel }} # default is 1 and applies per-model
               export OLLAMA_CONTEXT_LENGTH={{ llm_context }} # default is 4096, https://docs.ollama.com/faq#how-can-i-specify-the-context-window-size
               export OLLAMA_MAX_QUEUE=999 # default is 512 (unconfirmed)
               export OLLAMA_NUM_THREADS={{ (ansible_processor_vcpus / 2) | int }} # not real, but maybe some day.

tasks/workstation/linux/software/hugo.yml

@@ -22,35 +22,56 @@
         - name: Workstation | Linux | Software | Hugo | pacman | Install New
           ansible.builtin.package:
             name:
+              - hugo
               - "{{ sass_current }}"
             state: present
 
       when: ansible_pkg_mgr == "pacman"
 
     # This must exist after nodejs.yml so that NPM is installed.
-  - name: Workstation | Linux | Software | Hugo | npm
+    - name: Workstation | Linux | Software | Hugo | apt
       block:
 
-    - name: Workstation | Linux | Software | Hugo | npm | Facts
+        - name: Workstation | Linux | Software | Hugo | apt | sass | Facts
           ansible.builtin.set_fact:
             sass_old: node-sass
             sass_current: sass
 
-    - name: Workstation | Linux | Software | Hugo | npm | Gather
+        - name: Workstation | Linux | Software | Hugo | apt | sass | Gather
           ansible.builtin.shell: npm prefix -g
           register: npm_bin_dir # Provides /usr/local/ on Devuan, /usr/ on Artix.
           changed_when: false
 
-    - name: Workstation | Linux | Software | Hugo | npm | Remove Old
+        - name: Workstation | Linux | Software | Hugo | apt | sass | Remove Old
           ansible.builtin.shell: "npm remove -g {{ sass_old }}"
           args:
             creates: "{{ npm_bin_dir.stdout }}/bin/sass"
 
-    - name: Workstation | Linux | Software | Hugo | npm | Install New
+        - name: Workstation | Linux | Software | Hugo | apt | sass | Install New
           ansible.builtin.shell: "npm install -g {{ sass_current }}"
           args:
             creates: "{{ npm_bin_dir.stdout }}/bin/sass"
 
+        - name: Workstation | Linux | Software | Hugo | apt | hugo | Remove Old
+          ansible.builtin.package:
+            name:
+              - hugo
+            state: absent
+
+        - name: Workstation | Linux | Software | Hugo | apt | hugo | Install Dependencies
+          ansible.builtin.package:
+            name:
+              - golang
+              - gcc
+            state: present
+
+        - name: Workstation | Linux | Software | Hugo | apt | hugo | Build
+          ansible.builtin.shell: |
+            CGO_ENABLED=1 go install -tags extended github.com/gohugoio/hugo@latest
+            ln -s $HOME/go/bin/hugo /usr/local/bin/hugo
+          register: shell_check
+          changed_when: shell_check.rc == 0
+
       when: ansible_pkg_mgr == "apt"
 
   when: coding is true

test.yml

@@ -28,7 +28,7 @@
           ansible.builtin.include_tasks: tasks/workstation/linux/software/docker.yml
 
         - name: LLM
-          ansible.builtin.include_tasks: tasks/workstation/linux/software/llm.yml
+          ansible.builtin.include_tasks: tasks/workstation/linux/software/ai/llm.yml
 
     #######################
     #######################