ollama: ship fewer models, improve notes on the remaining models

modules/services/ollama.nix

@@ -19,32 +19,32 @@ let
   modelSources = pkgs.symlinkJoin {
     name = "ollama-models";
     paths = with pkgs.ollamaPackages; [
-      aya-8b
+      # aya-8b  # it avoids generating code, only text
       # codegeex4-9b  # it's okaaay, seems to not give wrong code, just incomplete code.
       codestral-22b
-      deepseek-coder-7b
-      deepseek-coder-v2-16b
-      falcon2-11b
+      # deepseek-coder-7b  # subpar to deepseek-coder-v2 in nearly every way
+      deepseek-coder-v2-16b  # GREAT balance between speed and code quality. code is superior to qwen2_5 in some ways, and inferior in others
+      # falcon2-11b  # code examples are lacking
       # gemma2-9b  # fast, but not great for code
-      glm4-9b
-      hermes3-8b
-      llama3-chatqa-8b
+      # glm4-9b  # it generates invalid code
+      # hermes3-8b  # FAST, but unwieldy
+      # llama3-chatqa-8b  # it gets stuck
       # llama3_1-70b  # generates like 1 word/sec, decent output (comparable to qwen2_5-32b)
       llama3_2-3b
       magicoder-7b  # it generates valid, if sparse, code
       # mistral-7b  # it generates invalid code
-      mistral-nemo-12b
-      mistral-small-22b
+      # mistral-nemo-12b  # it generates invalid code
+      mistral-small-22b  # quality comparable to qwen2_5
       # mistral-large-123b  # times out launch on desko
       mixtral-8x7b  # generates valid, if sparse, code
-      phi3_5-3b
+      # phi3_5-3b  # generates invalid code
       # qwen2_5-7b   # notably less quality than 32b (i.e. generates invalid code)
       qwen2_5-14b  # *almost* same quality to 32b variant, but faster
       qwen2_5-32b  # generates 3~5 words/sec, but notably more accurate than coder-7b
       # qwen2_5-coder-7b  # fast, and concise, but generates invalid code
       # solar-pro-22b  # generates invalid code
       # starcoder2-15b-instruct  # it gets stuck
-      wizardlm2-7b
+      # wizardlm2-7b  # generates invalid code
       yi-coder-9b  # subpar to qwen2-14b, but it's still useful
     ];
   };