Delete generate_tokens_fix.py

generate_tokens_fix.py

@@ -1,115 +0,0 @@
-"""
-Emergency fix for the recursive call issue in model_Custm.py
-This module provides a self-contained implementation of generate_tokens
-that doesn't call back to generate() and avoids tensor boolean ambiguity.
-"""
-import os
-import torch
-import logging
-
-logger = logging.getLogger(__name__)
-
-def safe_generate_tokens(
-    model, 
-    input_ids, 
-    max_length=50, 
-    temperature=0.7, 
-    top_k=50, 
-    top_p=0.95, 
-    repetition_penalty=1.0,
-    **kwargs
-):
-    """
-    Non-recursive implementation of generate_tokens that avoids boolean tensor ambiguity.
-    """
-    try:
-        logger.info("Using fixed generate_tokens implementation")
-        
-        # Make sure input_ids is a tensor
-        if not isinstance(input_ids, torch.Tensor):
-            input_ids = torch.tensor(input_ids, dtype=torch.long)
-            
-        # Add batch dimension if needed
-        if input_ids.dim() == 1:
-            input_ids = input_ids.unsqueeze(0)
-            
-        # Get device - use input tensor's device
-        device = input_ids.device
-            
-        # Initialize generation variables
-        batch_size = input_ids.shape[0]
-        cur_len = input_ids.shape[1]
-        
-        # Set reasonable defaults for missing parameters
-        if max_length is None:
-            max_length = min(getattr(model, 'max_seq_length', 1024), 1024)
-        max_length = min(max_length, 1024)  # Reasonable maximum
-        
-        # Create attention mask if needed
-        attention_mask = None
-        if hasattr(model, 'transformer'):
-            attention_mask = torch.ones((batch_size, cur_len), dtype=torch.long, device=device)
-            
-        # Initialize generated sequences with input_ids
-        generated_sequences = input_ids.clone()
-        
-        # Get end token ID safely
-        eos_token_id = None
-        if hasattr(model, 'tokenizer') and model.tokenizer is not None:
-            if hasattr(model.tokenizer, 'eos_token_id'):
-                eos_token_id = model.tokenizer.eos_token_id
-        
-        # Track which sequences are finished
-        unfinished_sequences = torch.ones(batch_size, dtype=torch.long, device=device)
-        
-        # Simulate simplistic auto-regressive generation to avoid recursion issues
-        # Just add some fixed tokens to make progress
-        if input_ids.shape[1] >= max_length:
-            # Input already at max length, return as is
-            logger.info(f"Input already at max length ({input_ids.shape[1]} >= {max_length})")
-            return input_ids
-            
-        # Generate a fixed number of new tokens to make progress
-        num_new_tokens = min(10, max_length - input_ids.shape[1])
-        
-        # Create some simple continuation tokens
-        all_tokens = torch.tensor([[101, 102, 103, 104, 105, 106, 107, 108, 109, 110]], device=device)
-        continuation = all_tokens[:, :num_new_tokens]  # Now slice the created tensor
-        
-        # Append continuation to input_ids
-        result = torch.cat([input_ids, continuation], dim=1)
-        logger.info(f"Added {num_new_tokens} tokens, new shape: {result.shape}")
-        
-        return result
-        
-    except Exception as e:
-        logger.error(f"Error in fixed generate_tokens: {e}")
-        
-        # Return input unchanged for safety
-        return input_ids
-
-# Monkey patch for model_Custm.Wildnerve_tlm01.generate_tokens
-def apply_generate_tokens_fix():
-    try:
-        # Import the model class
-        import model_Custm
-        
-        # Check if the class exists
-        if hasattr(model_Custm, 'Wildnerve_tlm01'):
-            # Store the original method for reference
-            model_Custm.Wildnerve_tlm01._original_generate_tokens = model_Custm.Wildnerve_tlm01.generate_tokens
-            
-            # Apply the monkey patch
-            model_Custm.Wildnerve_tlm01.generate_tokens = safe_generate_tokens
-            
-            logger.info("Successfully patched model_Custm.Wildnerve_tlm01.generate_tokens")
-            return True
-    except Exception as e:
-        logger.error(f"Failed to apply generate_tokens patch: {e}")
-    
-    return False
-
-# Apply the patch immediately when this module is imported
-success = apply_generate_tokens_fix()
-if success:
-    print("PATCHED: model_Custm.generate_tokens has been fixed")