Disable flaky test points

It is not yet clear what causes Ollama to not be reliable in these situations, but we do see differences. Will need to analyse separately and report upstream.

Author: ccreutzi

tests/tollamaChat.m

@@ -50,8 +50,13 @@ function extremeTopK(testCase)
         end
 
         function extremeTfsZ(testCase)
-            % setting tfs_z to z=0 leaves no random choice,
-            % so we expect to get a fixed response.
+            %% This should work, and it does on some computers. On others, Ollama
+            %% receives the parameter, but either Ollama or llama.cpp fails to
+            %% honor it correctly.
+            testCase.assumeTrue(false,"disabled due to Ollama/llama.cpp not honoring parameter reliably");
+
+            % setting tfs_z to z=0 leaves no random choice, but degrades to
+            % greedy sampling, so we expect to get a fixed response.
             chat = ollamaChat("mistral",TailFreeSamplingZ=0);
             prompt = "Sampling with tfs_z=0 returns a definite answer.";
             response1 = generate(chat,prompt);
@@ -70,6 +75,11 @@ function stopSequences(testCase)
         end
 
         function seedFixesResult(testCase)
+            %% This should work, and it does on some computers. On others, Ollama
+            %% receives the parameter, but either Ollama or llama.cpp fails to
+            %% honor it correctly.
+            testCase.assumeTrue(false,"disabled due to Ollama/llama.cpp not honoring parameter reliably");
+
             chat = ollamaChat("mistral");
             response1 = generate(chat,"hi",Seed=1234);
             response2 = generate(chat,"hi",Seed=1234);