From 5719ec6b60bfb3fde40745fa8ba7374c7ecffc19 Mon Sep 17 00:00:00 2001
From: Hyperling <me@hyperling.com>
Date: Wed, 21 Oct 2026 04:47:16 -0700
Subject: [PATCH] llm, add optional open-terminal package list, helpful but
 required Internet, may have to do an Internet check and then choose which
 docker run command to execute. Also fix and enhance the ollama_thread
 checker.

---
 tasks/workstation/linux/software/llm.yml | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/tasks/workstation/linux/software/llm.yml b/tasks/workstation/linux/software/llm.yml
index bda0364..2793bff 100644
--- a/tasks/workstation/linux/software/llm.yml
+++ b/tasks/workstation/linux/software/llm.yml
@@ -560,6 +560,9 @@
                 $image run --port {{ open_terminal_port }}
 
               # -p {{ open_terminal_port }}:8000
+
+              # This requires internet access to be successful...
+              #-e OPEN_TERMINAL_PACKAGES="xz-utils libglu1-mesa libglu1-mesa ninja-build libnotify-bin libgtk-3-dev liblzma-dev libstdc++-12-dev libc6:amd64 libstdc++6 lib32z1 libbz2-1.0 default-jdk libncurses-dev android-udev-rules" \
             mode: "0755"
 
         - name: Workstation | Linux | Software | LLMs | runit | open-terminal | Script | Finish
@@ -910,12 +913,23 @@
             if [[ -z $model ]]; then
               model="qwen3.5:4b"
             fi
+
+            #echo " Stopping all ollama usage..."
+            sudo sv restart ollama >/dev/null 2>&1
+
             # Benchmark Thread Count
             for t in {1..{{ ansible_processor_vcpus | int }}}; do
-              echo "`date` - Model $model w/ $t Threads`"
-              curl -s http://localhost:11434/api/generate \
-                -d '{"model":"$model","prompt":"Hello","stream":false,"options":{"num_thread":$t}}' \
-                | jq .eval_count / .eval_duration
+              echo -n "`date` - Model $model w/ $t Thread(s), tokens/sec: "
+
+              speed=$(curl -s http://localhost:11434/api/generate \
+                -d '{"model":"'$model'","prompt":"Hello","stream":false,"keep_alive":1,"options":{"num_thread":'$t'}}' \
+                | jq '.eval_count * 1000000000 / .eval_duration'
+              )
+              ctx="`ollama ps | grep $model | awk '{print $7}'`"
+
+              echo "$speed @ context=$ctx"
+
+              ollama stop $model >/dev/null 2>&1
             done
           }
           alias ollama-threads="ollama_threads "