Fix tokenizer configuration for proper MLX compatibility

- Add all special tokens to additional_special_tokens and added_tokens_decoder
- Add special tokens: <|system|>, <|user|>, <|assistant|>, <|end|>
- Embed chat_template directly in tokenizer_config.json
- Fix tokenizer_class from LlamaTokenizer to PreTrainedTokenizerFast
- Add proper token ID mappings for Phi-3 format
- Ensure compatibility with and without system prompts

This update ensures proper tokenization for MLX inference with the
Phi-3 chat format, supporting flexible usage with or without system
prompts in multi-turn conversations.

Tested with mlx-lm and verified correct token encoding.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

tokenizer_config.json

@@ -1,16 +1,72 @@
 {
   "add_prefix_space": null,
-  "additional_special_tokens": null,
-  "backend": "tokenizers",
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": ["<|system|>", "<|user|>", "<|assistant|>", "<|end|>"],
   "bos_token": "<s>",
+  "chat_template": "{% for message in messages %}{% if message['role'] == 'system' and message['content'] %}{{'<|system|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'user' %}{{'<|user|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'assistant' %}{{'<|assistant|>\n' + message['content'] + '<|end|>\n'}}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>\n' }}{% else %}{{ eos_token }}{% endif %}",
   "clean_up_tokenization_spaces": false,
   "eos_token": "<|endoftext|>",
-  "is_local": true,
   "model_max_length": 131072,
   "pad_token": "<|endoftext|>",
   "padding_side": "left",
-  "sp_model_kwargs": {},
-  "tokenizer_class": "LlamaTokenizer",
+  "tokenizer_class": "PreTrainedTokenizerFast",
   "unk_token": "<unk>",
   "use_default_system_prompt": false
 }