Upgrade to llama.cpp sha 615212

All done...

.github/workflows/cicd-mac.yml

@@ -39,8 +39,8 @@ jobs:
         uses: actions/checkout@v4
         with:
           repository: onicai/llama_cpp_onicai_fork
-          # ref: onicai  # Specify the branch name here
-          ref: onicai-615212  # While working on the upgrade...
+          ref: onicai  # Specify the branch name here
+          # ref: onicai-615212  # While working on the upgrade...
           path: src/llama_cpp_onicai_fork
           fetch-depth: 1 # Get just the last commit
           submodules: 'recursive'

README-contributors-guide.md

@@ -84,10 +84,15 @@ We need to rethink this logic, but for now it is ok...
 Do NOT merge the `onicai-<git-sha>` branch into the `onicai` branch, but replace it:
 
 ```
+# do the onicai branch management while master branch is checked out
+git checkout master
 git branch -m onicai onicai-<git-sha-old>
 git branch -m onicai-<git-sha-new> onicai
-git push origin onicai:onicai
+git push --force origin onicai:onicai
 git push origin onicai-<git-sha-old>:onicai-<git-sha-old>
+#
+# Switch to the onicai branch, which now contains the <git-sha-new> version
+git checkout onicai
 ```
 
 ## llama_cpp_canister

README.md

@@ -29,8 +29,12 @@ You can try out a variety of fully on-chain LLMs at https://icgpt.onicai.com
 # Capabilities 🔥
 
 - Deploy any LLM available as a gguf file.
-- Our largest so far is DeepSeek-R1 1.5B (See [X](https://x.com/onicaiHQ/status/1884339580851151089)).
 
+  *(The model must be able to produce at least 1 token per update call)*
+
+- Our largest so far is DeepSeek-R1 1.5B (See [X](https://x.com/onicaiHQ/status/1884339580851151089)).
+
+
 # Set up
 
 The build of the wasm must be done on a `Mac` !