From 5620e5a223cdf3e0679163e27416a242dab42897 Mon Sep 17 00:00:00 2001
From: Hyperling <me@hyperling.com>
Date: Wed, 21 Oct 2026 02:57:12 -0700
Subject: [PATCH] llm, add function for benchmarking whether more threads adds
 latency.

---
 tasks/workstation/linux/software/llm.yml | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/tasks/workstation/linux/software/llm.yml b/tasks/workstation/linux/software/llm.yml
index 41c68cf..b76573c 100644
--- a/tasks/workstation/linux/software/llm.yml
+++ b/tasks/workstation/linux/software/llm.yml
@@ -904,6 +904,23 @@
           }
           alias ollama-sizes="ollama_sizes "
           alias model-sizes="ollama_sizes "
+
+          function ollama_threads() {
+            model="$1"
+            if [[ -z $model ]]; then
+              model="qwen3.5:4b"
+            fi
+            # Benchmark Thread Count
+            for t in {1..{{ ansible_processor_vcpus }}}; do
+              echo "`date` - $t Threads`"
+              curl -s http://localhost:11434/api/generate \
+                -d '{"model":"$model","prompt":"Hello","stream":false,"options":{"num_thread":$t}}' \
+                | jq .eval_count / .eval_duration
+            done
+          }
+          alias ollama-threads="ollama_threads "
+          alias ollama-cpus="ollama_threads "
+          alias ollama-cpu="ollama_threads "
         state: present
         create: false
         backup: true