From a9080a6d1276f16345d520e0a8deb7884369c8d0 Mon Sep 17 00:00:00 2001
From: Hyperling <me@hyperling.com>
Date: Thu, 8 Oct 2026 21:09:24 -0700
Subject: [PATCH] zeroclaw, a few more small edits while learning how things
 work.

---
 tasks/workstation/linux/software/zeroclaw.yml | 41 ++++++++++---------
 1 file changed, 22 insertions(+), 19 deletions(-)

diff --git a/tasks/workstation/linux/software/zeroclaw.yml b/tasks/workstation/linux/software/zeroclaw.yml
index c47216e..664dd14 100644
--- a/tasks/workstation/linux/software/zeroclaw.yml
+++ b/tasks/workstation/linux/software/zeroclaw.yml
@@ -20,6 +20,7 @@
 #
 # References
 #   - https://docs.zeroclaw.com/master/en/reference/config.html
+#   - https://docs.ollama.com
 
 # Currently only testing this on Arch based distros (pacman).
 - name: Workstation | Linux | Software | ZeroClaw
@@ -30,7 +31,7 @@
         ollama: ollama
         zeroclaw: zeroclaw
         zerocode: zerocode
-        llm_concurrency: "{{ ansible_processor_vcpus / 4 | round(0, method='ceil') | int }}"
+        llm_concurrency: 1 # "{{ ansible_processor_vcpus / 4 | round(0, method='ceil') | int }}"
         llm_installs: []
         llm_tier0_cpu: 1
         llm_tier0_ram: 2000
@@ -41,35 +42,36 @@
         llm_tier1_cpu: 2
         llm_tier1_ram: 4000
         llm_tier1:
-          - { "model": "qwen3.5:2b",         "provider": "qwen_35_0020" }
-          - { "model": "qwen2.5-coder:3b",   "provider": "qwen_25_coder_0030" }
+          - { "model": "qwen3.5:2b",       "provider": "qwen_35_0020" }
+          - { "model": "qwen2.5-coder:3b", "provider": "qwen_25_coder_0030" }
         llm_tier2_cpu: 3
         llm_tier2_ram: 6000
         llm_tier2:
-          - { "model": "qwen3.5:4b",         "provider": "qwen_35_0040" }
-          - { "model": "qwen2.5-coder:7b",   "provider": "qwen_25_coder_0070" }
+          - { "model": "qwen3.5:4b",       "provider": "qwen_35_0040" }
+          - { "model": "qwen2.5-coder:7b", "provider": "qwen_25_coder_0070" }
         llm_tier3_cpu: 4
         llm_tier3_ram: 8000
         llm_tier3:
-          - { "model": "qwen3.5:9b",         "provider": "qwen_35_0090" }
-          - { "model": "qwen2.5-coder:14b",  "provider": "qwen_25_coder_0140" }
+          - { "model": "qwen3.5:9b",        "provider": "qwen_35_0090" }
+          - { "model": "qwen2.5-coder:14b", "provider": "qwen_25_coder_0140" }
         llm_tier4_cpu: 6
         llm_tier4_ram: 12000
-        llm_tier4:
-        llm_tier5_cpu: 8
-        llm_tier5_ram: 16000
+        llm_tier4: []
+        llm_tier5_cpu: 10
+        llm_tier5_ram: 20000
         llm_tier5:
-        llm_tier6_cpu: 10
+          #- { "model": "devstral-small-2:24b", "provider": "devstral_20_0240_small" }
+          #- { "model": "qwen3.8:27b",          "provider": "qwen_38_0270" }
+        llm_tier6_cpu: 12
         llm_tier6_ram: 30000
         llm_tier6:
-          #- { "model": "qwen3.8:27b",        "provider": "qwen_38_0270" }
-          #- { "model": "qwen3.6:35b",        "provider": "qwen_36_0350" }
-          - { "model": "qwen3.5:35b",        "provider": "qwen_35_0350" }
-          - { "model": "qwen3-coder:30b",    "provider": "qwen_30_coder_0300" }
-          - { "model": "qwen2.5-coder:32b",  "provider": "qwen_25_coder_0320" }
-        llm_tier7_cpu: 12
+          #- { "model": "qwen3.6:35b",       "provider": "qwen_36_0350" }
+          - { "model": "qwen3.5:35b",       "provider": "qwen_35_0350" }
+          - { "model": "qwen3-coder:30b",   "provider": "qwen_30_coder_0300" }
+          - { "model": "qwen2.5-coder:32b", "provider": "qwen_25_coder_0320" }
+        llm_tier7_cpu: 16
         llm_tier7_ram: 64000
-        llm_tier7:
+        llm_tier7: []
         llm_tier8_cpu: 32
         llm_tier8_ram: 320000
         llm_tier8:
@@ -302,8 +304,9 @@
               export OLLAMA_REQUEST_TIMEOUT=3600 # default 300  (unconfirmed, not in the list of --help vars)
               export OLLAMA_LOAD_TIMEOUT=30m # default is 5m
               export OLLAMA_MAX_LOADED_MODELS={{ llm_concurrency }} # default is 3 if using CPUs, or 3 * NumGPUs
-              export OLLAMA_MAX_QUEUE=999999999 # default is 512 (unconfirmed)
+              export OLLAMA_MAX_QUEUE=999 # default is 512 (unconfirmed)
               export OLLAMA_NUM_PARALLEL={{ llm_concurrency }} # default is 1 and applies per-model (unconfirmed)
+              export OLLAMA_CONTEXT_LENGTH=999999 # default is 4096, https://docs.ollama.com/faq#how-can-i-specify-the-context-window-size
               exec chpst -u {{ user }} ollama serve
             dest: "{{ ollama_runit_script }}"
             mode: "0755"