From 019066c4dcba33d1f43d67c87f3e564cf94da49f Mon Sep 17 00:00:00 2001
From: Hyperling <me@hyperling.com>
Date: Sat, 10 Oct 2026 16:42:19 -0700
Subject: [PATCH] llm, further reduce contexts based on testing, now using
 2.5:14b for Complete successfully alongside the RAM needs of 3:30b on Cline.

---
 tasks/workstation/linux/software/llm.yml | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/tasks/workstation/linux/software/llm.yml b/tasks/workstation/linux/software/llm.yml
index 9574514..e633acc 100644
--- a/tasks/workstation/linux/software/llm.yml
+++ b/tasks/workstation/linux/software/llm.yml
@@ -62,8 +62,8 @@
         # Conversely:
         #   - qwen3.5:4b would be 3.3GB * 1.3 = 4290, rather than 4000.
         llm_size_coef: 1.3
-        # Take the GB of RAM and use it as the MB of context.
-        llm_context: "{{ [(ansible_memtotal_mb | round(-3, method='ceil')), 32000] | max | int }}"
+        # Use the MB of RAM as the context, with a maximum to prevent swapping.
+        llm_context: "{{ [(ansible_memtotal_mb | round(-3, method='ceil')), 32000] | min | int }}"
 
     - name: Workstation | Linux | Software | LLMs | Facts 02
       ansible.builtin.set_fact:
@@ -591,7 +591,7 @@
           [providers.models.ollama.{{ llm.provider }}]
           model = "{{ llm.model }}"
           timeout_secs = 3600 # 1 hour, default is 120 (2 minutes), issue happens between 5 and 6 minutes though.
-          context_window = {{ [llm_context, llm.context, 24000] | min | int }}
+          context_window = {{ [llm_context, llm.context, 16000] | min | int }}
           max_tokens = {{ [llm_tokens, (llm.context / 4), 4000] | min | int }}
 
           {% endif %}
@@ -625,14 +625,14 @@
               provider: ollama
               model: {{ llm.model }}
               defaultCompletionOptions:
-                contextLength: {{ [llm_context, llm.context, 32000] | min | int }}
+                contextLength: {{ [llm_context, llm.context, 24000] | min | int }}
                 maxTokens: {{ [llm_tokens, (llm.context / 4), 8000] | min | int }}
 
             - name: {{ llm.provider }}_autocomplete
               provider: ollama
               model: {{ llm.model }}
               defaultCompletionOptions:
-                contextLength: {{ [llm_context, llm.context, 16000] | min | int }}
+                contextLength: {{ [llm_context, llm.context, 12000] | min | int }}
                 maxTokens: {{ [llm_tokens, (llm.context / 4), 2000] | min | int }}
               roles:
                 - autocomplete