chore: move debugging scripts to dedicated folder

Author: solderzzc

scripts/debugging/list_keys.py

@@ -0,0 +1,13 @@
+from safetensors import safe_open
+from glob import glob
+import sys
+
+paths = glob("/Users/simba/.cache/huggingface/hub/models--mlx-community--gemma-4-26b-a4b-it-4bit/snapshots/*/*.safetensors")
+found = False
+for path in paths:
+    with safe_open(path, framework="np") as f:
+        for k in f.keys():
+            if "layers.3.experts" in k:
+                print(k, f.get_tensor(k).shape)
+                found = True
+        if found: break

scripts/debugging/test_tokens.py

@@ -0,0 +1,8 @@
+from transformers import AutoTokenizer
+
+tok = AutoTokenizer.from_pretrained("mlx-community/gemma-4-26b-a4b-it-4bit")
+messages = [{"role": "user", "content": "What is 2+2?"}]
+prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+tokens = tok.encode(prompt)
+print("PROMPT:", repr(prompt))
+print("TOKENS:", len(tokens), tokens)

scripts/debugging/test_tokens.swift

@@ -0,0 +1,21 @@
+import Foundation
+import MLXLMCommon
+import MLX
+
+func run() async throws {
+    let factory = LLMModelFactory.shared
+    let config = ModelConfiguration(id: "mlx-community/gemma-4-26b-a4b-it-4bit")
+    let container = try await factory.loadContainer(from: HubClient.default, using: TokenizersLoader(), configuration: config)
+    let context = try await container.context()
+    
+    let messages: [[String: Any]] = [
+        ["role": "user", "content": "What is 2+2?"]
+    ]
+    let prompt = try context.tokenizer.applyChatTemplate(messages: messages)
+    let tokens = context.tokenizer.encode(text: prompt)
+    print("PROMPT TEXT:", prompt)
+    print("TOKENS:", tokens)
+}
+
+Task { try await run(); exit(0) }
+RunLoop.main.run()

scripts/debugging/test_update.swift

@@ -0,0 +1,17 @@
+import Foundation
+import MLX
+import MLXNN
+
+class DummyModule: Module {
+    @ModuleInfo(key: "my_proj") var myProj: Module
+    override init() {
+        self._myProj.wrappedValue = Linear(10, 10)
+        super.init()
+    }
+}
+
+let m = DummyModule()
+print("Module initialized")
+let updates = [("my_proj", Linear(10, 10))]
+m.update(modules: updates)
+print("Module updated successfully!")

scripts/debugging/test_weights.swift

@@ -0,0 +1,24 @@
+import Foundation
+import MLX
+import MLXLLM
+import MLXLMCommon
+
+@main
+struct TestWeights {
+    static func main() async throws {
+        let config = ModelConfiguration(id: "mlx-community/gemma-4-26b-a4b-it-4bit")
+        let modelContainer = try await LLMModelFactory.shared.loadContainer(from: HubClient.default, using: TokenizersLoader(), configuration: config)
+        let context = try await modelContainer.context()
+
+        if let gemma4 = context.model as? Gemma4Model {
+            print("Successfully cast to Gemma4Model!")
+            let layer0 = gemma4.model.layers[0]
+            print("Layer 0 norms:")
+            print("post1:", layer0.postFeedforwardLayerNorm1?.weight.max().item(Float.self) ?? "NIL")
+            print("pre2:", layer0.preFeedforwardLayerNorm2?.weight.max().item(Float.self) ?? "NIL")
+            print("post2:", layer0.postFeedforwardLayerNorm2?.weight.max().item(Float.self) ?? "NIL")
+            
+            print("Router proj:", layer0.expertsBlock?.router.proj.weight.shape ?? [0])
+        }
+    }
+}