Initial upload of fine‑tuned Gemma + custom tokenizer

gemma_special_tokenizer.py

@@ -0,0 +1,407 @@
+"""
+Custom Gemma Tokenizer for special Format
+
+This tokenizer implements the special format for message processing:
+Format: {description}\n{input}\n<<{output}>>\n
+
+The special format wraps output content in double angle specials (<<output>>)
+and includes proper loss computation flags for training.
+
+To save:
+uv run tokenizers/gemma_special_tokenizer.py 
+which will save the tokenizer to the repos/special-gemma-tokenizer directory.
+mkdir repos/special12b
+# copy model over
+cp models_v8/base_modified-google-gemma-3-12b-pt-/models/_special/checkpoint-8/* repos/special12b/
+# copy tokenizer over
+cp repos/special-gemma-tokenizer/* repos/special12b/
+# upload to hf
+
+uv run upload_to_hf.py \
+  --folder repos/special12b \
+  --repo-id tsor13/special12b
+"""
+
+from typing import List, Dict, Any, Optional, Union
+from transformers import AutoTokenizer
+from transformers.models.gemma.tokenization_gemma_fast import GemmaTokenizerFast
+from transformers.models.gemma.tokenization_gemma import GemmaTokenizer
+import warnings
+import difflib
+import json
+import os
+
+
+class GemmaSpecialTokenizer(GemmaTokenizerFast):
+    """
+    Custom tokenizer for Gemma models that implements special format message processing.
+    
+    This tokenizer formats messages using the special format where:
+    - Description and input content are displayed as plain text with newlines
+    - Output content is wrapped in double angle specials: <<output>>
+    - Loss is computed on the specialed output sections
+    
+    Attributes:
+        start_string (str): The starting string used for output generation ("<<")
+        end_string (str): The ending string used for output generation (">>")
+    """
+    
+    def __init__(self, *args, **kwargs):
+        """
+        Initialize the custom tokenizer.
+        
+        Accepts the same arguments as GemmaTokenizerFast.
+        """
+        super().__init__(*args, **kwargs)
+        
+        # Store the end string for special format
+        # self.start_string = "<<"
+        # self.end_string = ">>"
+        self.start_string = "<start_of_turn>"
+        self.end_string = "<end_of_turn>"
+        
+        # Add custom attributes to the tokenizer config for saving/loading
+        if not hasattr(self, 'init_kwargs'):
+            self.init_kwargs = {}
+        self.init_kwargs['start_string'] = self.start_string
+        self.init_kwargs['end_string'] = self.end_string
+    
+    @classmethod
+    # def from_pretrained(cls, pretrained_model_name_or_path, *args, **kwargs):
+    def from_gemma_pretrained(cls, pretrained_model_name_or_path, *args, **kwargs):
+        """
+        Load a tokenizer from a pretrained model or path.
+        
+        This method ensures our custom class is used instead of the base GemmaTokenizerFast.
+        """
+        # TODO - there's a warning here when loading the tokenizer from the hub
+        # Load the base tokenizer first to get all configuration
+        base_tokenizer = GemmaTokenizerFast.from_pretrained(
+            pretrained_model_name_or_path, *args, **kwargs
+        )
+        
+        # Create new instance of our custom class by copying the base tokenizer
+        custom_tokenizer = cls.__new__(cls)
+        
+        # Copy all attributes from base tokenizer
+        for attr, value in base_tokenizer.__dict__.items():
+            setattr(custom_tokenizer, attr, value)
+        
+        # Initialize our custom attributes
+        # custom_tokenizer.start_string = "<<"
+        # custom_tokenizer.end_string = ">>"
+        custom_tokenizer.start_string = "<start_of_turn>"
+        custom_tokenizer.end_string = "<end_of_turn>"
+        
+        # Update init_kwargs to include our custom attributes
+        if not hasattr(custom_tokenizer, 'init_kwargs'):
+            custom_tokenizer.init_kwargs = {}
+        custom_tokenizer.init_kwargs['start_string'] = custom_tokenizer.start_string
+        custom_tokenizer.init_kwargs['end_string'] = custom_tokenizer.end_string
+        
+        return custom_tokenizer
+    
+    def save_pretrained(self, save_directory: Union[str, os.PathLike], **kwargs):
+        """
+        Save the tokenizer to a directory, including custom configuration.
+        """
+        # Call parent save method
+        super().save_pretrained(save_directory, **kwargs)
+        
+        # Save custom configuration
+        config_file = os.path.join(save_directory, "tokenizer_config.json")
+        if os.path.exists(config_file):
+            with open(config_file, 'r') as f:
+                config = json.load(f)
+        else:
+            config = {}
+        
+        # Add our custom class info
+        config["tokenizer_class"] = "GemmaSpecialTokenizer"
+        config["start_string"] = self.start_string
+        config["end_string"] = self.end_string
+        # Point to our custom class in the uploaded file
+        # config["tokenizer_class"] = "GemmaTokenizerCustom"
+        config["auto_map"] = {
+            "AutoTokenizer": ["gemma_special_tokenizer.GemmaSpecialTokenizer", "gemma_special_tokenizer.GemmaSpecialTokenizer"]
+        }
+        
+        with open(config_file, 'w') as f:
+            json.dump(config, f, indent=2)
+    
+    def messages_to_loss_texts(
+            self,
+            messages: List[Dict[str, Any]], 
+            loss_on_start_token: bool = False, 
+        ) -> List[Dict[str, Any]]:
+        """
+        From messages (description / input / output) to texts (text / compute_loss) with whether or not loss should be calculated on the text for training.
+        """
+        texts = []
+        for message in messages:
+            role = message["role"]
+            content = message["content"]
+            
+            if role == "description":
+                text = f"{content}\n"
+                texts.append({"text": text, "compute_loss": False, **message})
+            elif role == "input":
+                text = f"{content}\n"
+                texts.append({"text": text, "compute_loss": False, **message})
+            elif role == "output":
+                if loss_on_start_token:
+                    # For output, wrap content in double angle specials and include newline
+                    # text = f"<<{content}>>\n"
+                    text = f"{self.start_string}{content}{self.end_string}\n"
+                    texts.append({"text": text, "compute_loss": True, **message})
+                else:
+                    texts.append({"text": self.start_string, "compute_loss": False, **message})
+                    text = f"{content}{self.end_string}\n"
+                    texts.append({"text": text, "compute_loss": True, **message})
+            else:
+                raise ValueError(f"Unknown role: {role}. Must be description, input, or output.")
+        
+        # # Add generation prompt if start_generation is True
+        # if start_generation:
+        #     texts.append({"text": self.start_string, "compute_loss": False})
+        
+        return texts
+    
+    def messages_to_text(
+            self, 
+            messages: List[Dict[str, Any]], 
+            start_generation: bool = False, 
+        ) -> str:
+        """
+        Messages (description / input / output) to raw text (text).
+        """
+        texts = self.messages_to_loss_texts(messages)
+        text = "".join([text["text"] for text in texts])
+        if start_generation:
+            text = text + self.start_string
+        return text
+    
+
+    def tokenize_messages(
+            self, 
+            messages: List[Dict[str, Any]] | List[List[Dict[str, Any]]],
+            start_generation: bool = False, 
+            **kwargs,
+        ):
+        """
+        For tokenizing from messages to texts. Supports batching. Good for generation
+        """
+        if isinstance(messages, list) and isinstance(messages[0], list):
+            # Handle list of lists of messages
+            all_texts = []
+            for message_list in messages:
+                texts = self.messages_to_raw_text(message_list, start_generation)
+                all_texts.append(texts)
+        else:
+            # Handle single list of messages
+            texts = self.messages_to_raw_text(messages, start_generation)
+            all_texts = [texts]
+        
+        # Tokenize all texts
+        processed = self(text=all_texts, **kwargs)
+        return processed
+    
+    
+    def tokenize_loss_texts(
+            self, 
+            texts: List[Dict[str, Any]], 
+            loss_on_start_token: bool = False, 
+            loss_on_eos: bool = False, 
+            include_eos: bool = True,
+        ):
+        """
+        Tokenize texts (text / compute_loss) to tokenized texts (input_ids / attention_mask / labels).
+
+        Needs more complex logic to handle the back and forth labeling.
+        """
+        if loss_on_eos:
+            raise ValueError("Loss on EOS is not currently supported.")
+        
+        # Handle single string input
+        if isinstance(texts, str):
+            processed = self(text=texts)
+            # Add EOS token if needed
+            if (self.eos_token_id is not None and 
+                processed["input_ids"][-1] != self.eos_token_id):
+                processed["input_ids"] = processed["input_ids"] + [self.eos_token_id]
+                processed["attention_mask"] = processed["attention_mask"] + [1]
+            return processed
+        
+        # Handle list of text dictionaries
+        all_processed = []
+        all_texts = ''
+        example_inds = []
+        dataset_inds = []
+        
+        for i, item in enumerate(texts):
+            processed = self(text=item["text"])
+            
+            # Remove BOS token from all but first item
+            if i != 0 and self.bos_token_id == processed["input_ids"][0]:
+                processed["input_ids"] = processed["input_ids"][1:]
+                processed["attention_mask"] = processed["attention_mask"][1:]
+            
+            # Remove EOS token if present at the end
+            if processed["input_ids"][-1] == self.eos_token_id:
+                processed["input_ids"] = processed["input_ids"][:-1]
+                processed["attention_mask"] = processed["attention_mask"][:-1]
+            
+            # Check for EOS token in the middle (with special handling for <|im_end|>)
+            if self.eos_token_id in processed["input_ids"]:
+                if not self.decode([self.eos_token_id]) == "<|im_end|>":
+                    raise ValueError(f"EOS token is present in input_ids: {processed['input_ids']}. Not currently supported.")
+            
+            # Set labels based on compute_loss flag
+            if item["compute_loss"]:
+                processed["labels"] = processed["input_ids"].copy()
+            else:
+                processed["labels"] = [-100] * len(processed["input_ids"])
+            
+            # Remove duplicate BOS tokens
+            if all_processed:
+                if processed["input_ids"][0] == self.bos_token_id:
+                    processed["input_ids"] = processed["input_ids"][1:]
+                    processed["attention_mask"] = processed["attention_mask"][1:]
+                    processed["labels"] = processed["labels"][1:]
+            
+            all_processed.append(processed)
+            all_texts += item["text"]
+            
+            # Handle example indices
+            this_num = -1
+            if 'example_ind' in item.keys():
+                if item["example_ind"] is not None:
+                    this_num = item["example_ind"]
+            example_inds.extend([this_num] * len(processed["input_ids"]))
+            
+            # Handle dataset indices
+            dataset_ind = -1
+            if "data_id" in item.keys():
+                if item["data_id"] is not None:
+                    dataset_ind = item["data_id"]
+            dataset_inds.extend([dataset_ind] * len(processed["input_ids"]))
+        
+        # Combine all processed results
+        processed = all_processed[0].copy()
+        processed["input_ids"] = [item for sublist in [p["input_ids"] for p in all_processed] for item in sublist]
+        processed["attention_mask"] = [item for sublist in [p["attention_mask"] for p in all_processed] for item in sublist]
+        processed["labels"] = [item for sublist in [p["labels"] for p in all_processed] for item in sublist]
+        processed["example_inds"] = example_inds
+        processed["data_ids"] = dataset_inds
+        
+        # Validate by tokenizing all_texts at once and comparing
+        processed_all = self(text=all_texts)
+        if len(processed_all["input_ids"]) != len(processed["input_ids"]):
+            warnings.warn(f"All texts are not the same length as the first text. Please check your dataset. {len(processed_all['input_ids'])} != {len(processed['input_ids'])}")
+            
+            # Generate diff for debugging
+            all_text = self.decode(processed_all["input_ids"], skip_special_tokens=False)
+            processed_text = self.decode(processed["input_ids"], skip_special_tokens=False)
+            
+            diff = difflib.unified_diff(all_text.splitlines(), processed_text.splitlines())
+            diff_str = "\n".join(diff)
+            print("Diff between texts:")
+            print(diff_str)
+            
+            # Token diff
+            all_tokens_str = '\n'.join([str(s) for s in processed_all["input_ids"]])
+            processed_tokens_str = '\n'.join([str(s) for s in processed["input_ids"]])
+            token_diff = difflib.unified_diff(all_tokens_str.splitlines(), processed_tokens_str.splitlines())
+            token_diff_str = "\n".join(token_diff)
+            print("Diff between tokenized texts:")
+            print(token_diff_str)
+        
+        # Add EOS token if needed
+        if (self.eos_token_id is not None and 
+            processed["input_ids"][-1] != self.eos_token_id):
+            processed["input_ids"] = processed["input_ids"] + [self.eos_token_id]
+            processed["example_inds"] = processed["example_inds"] + [-1]
+            processed["attention_mask"] = processed["attention_mask"] + [1]
+            if processed["labels"] is not None:
+                if loss_on_eos:
+                    processed["labels"] = processed["labels"] + [self.eos_token_id]
+                else:
+                    processed["labels"] = processed["labels"] + [-100]
+            if "data_ids" in processed:
+                processed["data_ids"] = processed["data_ids"] + [-1]
+            
+        if not include_eos:
+            # check if EOS token is present
+            if processed["input_ids"][-1] == self.eos_token_id:
+                # remove EOS token
+                processed["input_ids"] = processed["input_ids"][:-1]
+                processed["attention_mask"] = processed["attention_mask"][:-1]
+                processed["labels"] = processed["labels"][:-1]
+                processed["example_inds"] = processed["example_inds"][:-1]
+                processed["data_ids"] = processed["data_ids"][:-1]
+        
+        return processed
+    
+    def tokenize_messages(
+            self, 
+            messages: List[Dict[str, Any]], 
+            loss_on_start_token: bool = False, 
+            loss_on_eos: bool = False,
+            include_eos: bool = True,
+        ) -> Dict[str, Any]:
+        """
+        Intended for tokenize from messages to tokenized texts with the loss applied.
+        """
+        # First convert messages to text with loss computation flags
+        texts = self.messages_to_text_loss(messages, loss_on_start_token)
+        
+        # Then tokenize the texts
+        return self.tokenize_loss_texts(texts, loss_on_eos, include_eos = include_eos)
+    
+
+
+
+# Register tokenizer classes for AutoTokenizer
+# Note: We register them separately to avoid conflicts
+AutoTokenizer.register("GemmaSpecialTokenizer", slow_tokenizer_class=None, fast_tokenizer_class=GemmaSpecialTokenizer)
+# AutoTokenizer.register("GemmaSpecialTokenizerSlow", slow_tokenizer_class=GemmaSpecialTokenizerSlow, fast_tokenizer_class=None)
+
+
+if __name__ == "__main__":
+    # Example usage
+    # for first load
+    custom_tokenizer = GemmaSpecialTokenizer.from_gemma_pretrained("google/gemma-3-1b-pt")
+
+    # for subsequent loads
+    # custom_tokenizer = GemmaSpecialTokenizer.from_pretrained("tsor13/special-gemma-12b-pt")
+    # custom_tokenizer = GemmaSpecialTokenizer.from_pretrained("repos/special-gemma-12b-pt")
+    
+    # Test messages in role/content format
+    messages = [
+        {"role": "description", "content": "This is a test task"},
+        {"role": "input", "content": "What is 2+2?"},
+        {"role": "output", "content": "4"},
+        {"role": "input", "content": "What is 3+3?"},
+        # {"role": "output", "content": "6"}
+    ]
+
+    # tokenized = custom_tokenizer.tokenize_messages(messages, start_generation=True, return_tensors="pt")
+    # print(tokenized)
+
+    # get messages to text_loss
+    texts = custom_tokenizer.messages_to_loss_texts(messages)
+    print(texts)
+
+    text = custom_tokenizer.messages_to_text(messages, start_generation=True)
+    print(text)
+    
+    print("\nTesting save/load cycle:")
+    # Test saving and loading
+    tokenizer_path = "repos/special-gemma-tokenizer"
+    custom_tokenizer.save_pretrained(tokenizer_path)
+    print("Tokenizer saved successfully!")
+
+    # also save this file in the tokenizer_path
+    import shutil
+    shutil.copy(__file__, os.path.join(tokenizer_path, "gemma_special_tokenizer.py"))
+    print("GemmaSpecialTokenizer.py saved successfully!")

tokenizer_config.json

@@ -51340,13 +51340,13 @@
   "sp_model_kwargs": null,
   "spaces_between_special_tokens": false,
   "start_string": "<start_of_turn>",
-  "tokenizer_class": "GemmaBracketTokenizer",
+  "tokenizer_class": "GemmaSpecialTokenizer",
   "unk_token": "<unk>",
   "use_default_system_prompt": false,
   "auto_map": {
     "AutoTokenizer": [
-      "gemma_bracket_tokenizer.GemmaBracketTokenizer",
-      "gemma_bracket_tokenizer.GemmaBracketTokenizer"
+      "gemma_special_tokenizer.GemmaSpecialTokenizer",
+      "gemma_special_tokenizer.GemmaSpecialTokenizer"
     ]
   }
 }