optipfair 0.1.0__py3-none-any.whl

cli/__init__.py

@@ -0,0 +1,7 @@
+"""
+CLI package for OptiPFair.
+"""
+
+from .commands import cli
+
+__all__ = ["cli"]

cli/commands.py

@@ -0,0 +1,209 @@
+"""
+Command-line interface for OptiPFair.
+
+This module provides the CLI commands for pruning models and related operations.
+"""
+
+import click
+import logging
+import os
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+from .. import prune_model
+from ..pruning.utils import count_parameters
+
+logger = logging.getLogger(__name__)
+
+@click.group()
+def cli():
+    """OptiPFair: A library for structured pruning of large language models."""
+    pass
+
+@cli.command()
+@click.option('--model-path', required=True, help='Path or identifier of the model to prune.')
+@click.option('--pruning-type', default='MLP_GLU', type=click.Choice(['MLP_GLU']), 
+              help='Type of pruning to apply.')
+@click.option('--method', default='MAW', type=click.Choice(['MAW', 'VOW', 'PON']), 
+              help='Method to calculate neuron importance.')
+@click.option('--pruning-percentage', default=None, type=float, 
+              help='Percentage of neurons to prune (0-100).')
+@click.option('--expansion-rate', default=None, type=float, 
+              help='Target expansion rate in percentage (mutually exclusive with pruning-percentage).')
+@click.option('--output-path', required=True, help='Path to save the pruned model.')
+@click.option('--device', default='auto', help='Device to use for computation (auto, cpu, cuda, cuda:0, etc.)')
+@click.option('--dtype', default='auto', type=click.Choice(['auto', 'float32', 'float16', 'bfloat16']), 
+              help='Data type to load the model with.')
+@click.option('--verbose/--quiet', default=True, help='Whether to show verbose output.')
+def prune(model_path, pruning_type, method, pruning_percentage, expansion_rate, 
+          output_path, device, dtype, verbose):
+    """Prune a language model using the specified parameters."""
+    # Configure logging based on verbosity
+    log_level = logging.INFO if verbose else logging.WARNING
+    logging.basicConfig(level=log_level, format='%(message)s')
+    
+    # Validate inputs
+    if pruning_percentage is not None and expansion_rate is not None:
+        raise click.UsageError("--pruning-percentage and --expansion-rate are mutually exclusive.")
+    
+    if pruning_percentage is None and expansion_rate is None:
+        pruning_percentage = 10
+        logger.info(f"No pruning target specified, defaulting to {pruning_percentage}% pruning.")
+    
+    # Determine device
+    if device == 'auto':
+        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    
+    # Determine dtype
+    if dtype == 'auto':
+        if 'cuda' in device and torch.cuda.is_available():
+            if torch.cuda.get_device_capability()[0] >= 8:  # Ampere or newer
+                dtype = torch.bfloat16
+            else:
+                dtype = torch.float16
+        else:
+            dtype = torch.float32
+    else:
+        dtype_map = {
+            'float32': torch.float32,
+            'float16': torch.float16,
+            'bfloat16': torch.bfloat16,
+        }
+        dtype = dtype_map[dtype]
+    
+    logger.info(f"Loading model from {model_path} to {device} with {dtype}")
+    
+    # Load model and tokenizer
+    try:
+        model = AutoModelForCausalLM.from_pretrained(
+            model_path,
+            torch_dtype=dtype,
+            device_map=device
+        )
+        tokenizer = AutoTokenizer.from_pretrained(model_path)
+    except Exception as e:
+        logger.error(f"Error loading model: {str(e)}")
+        raise
+    
+    # Log original model parameters
+    original_params = count_parameters(model)
+    logger.info(f"Original model parameters: {original_params:,}")
+    
+    # Apply pruning
+    logger.info(f"Pruning model with {pruning_type} pruning, {method} neuron selection method")
+    if pruning_percentage is not None:
+        logger.info(f"Target: {pruning_percentage}% reduction in neurons")
+    else:
+        logger.info(f"Target: {expansion_rate}% expansion rate")
+    
+    pruned_model, stats = prune_model(
+        model=model,
+        pruning_type=pruning_type,
+        neuron_selection_method=method,
+        pruning_percentage=pruning_percentage,
+        expansion_rate=expansion_rate,
+        show_progress=verbose,
+        return_stats=True,
+    )
+    
+    # Log pruning statistics
+    logger.info("Pruning complete!")
+    logger.info(f"Original parameters: {stats['original_parameters']:,}")
+    logger.info(f"Pruned parameters: {stats['pruned_parameters']:,}")
+    logger.info(f"Reduction: {stats['reduction']:,} parameters ({stats['percentage_reduction']:.2f}%)")
+    
+    if stats['expansion_rate'] is not None:
+        logger.info(f"Final expansion rate: {stats['expansion_rate']:.2f}%")
+    
+    # Create output directory if it doesn't exist
+    os.makedirs(output_path, exist_ok=True)
+    
+    # Save model and tokenizer
+    logger.info(f"Saving pruned model to {output_path}")
+    pruned_model.save_pretrained(output_path)
+    tokenizer.save_pretrained(output_path)
+    
+    logger.info("Done!")
+
+@cli.command()
+@click.option('--model-path', required=True, help='Path or identifier of the model to analyze.')
+@click.option('--device', default='auto', help='Device to use for computation (auto, cpu, cuda, cuda:0, etc.)')
+def analyze(model_path, device):
+    """Analyze a model's architecture and parameter distribution."""
+    # Configure logging
+    logging.basicConfig(level=logging.INFO, format='%(message)s')
+    
+    # Determine device
+    if device == 'auto':
+        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    
+    logger.info(f"Loading model from {model_path} to {device}")
+    
+    # Load model
+    try:
+        model = AutoModelForCausalLM.from_pretrained(
+            model_path,
+            device_map=device
+        )
+    except Exception as e:
+        logger.error(f"Error loading model: {str(e)}")
+        raise
+    
+    # Log basic model information
+    total_params = count_parameters(model)
+    logger.info(f"Model: {model_path}")
+    logger.info(f"Total parameters: {total_params:,}")
+    
+    # Try to identify model architecture
+    from ..pruning.utils import get_model_layers, validate_model_for_glu_pruning
+    
+    layers = get_model_layers(model)
+    logger.info(f"Number of layers: {len(layers)}")
+    
+    # Check if model is compatible with GLU pruning
+    is_glu_compatible = validate_model_for_glu_pruning(model)
+    logger.info(f"Compatible with GLU pruning: {is_glu_compatible}")
+    
+    if is_glu_compatible and layers:
+        # Get information about the first layer
+        first_layer = layers[0]
+        mlp = first_layer.mlp
+        
+        hidden_size = mlp.gate_proj.in_features
+        intermediate_size = mlp.gate_proj.out_features
+        expansion_ratio = intermediate_size / hidden_size
+        
+        logger.info(f"Hidden size: {hidden_size}")
+        logger.info(f"Intermediate size: {intermediate_size}")
+        logger.info(f"Expansion ratio: {expansion_ratio:.2f}x ({expansion_ratio*100:.2f}%)")
+        
+        # Parameter distribution
+        attn_params = sum(p.numel() for name, p in first_layer.named_parameters() if 'self_attn' in name)
+        mlp_params = sum(p.numel() for name, p in first_layer.named_parameters() if 'mlp' in name)
+        other_params = sum(p.numel() for name, p in first_layer.named_parameters() 
+                           if 'self_attn' not in name and 'mlp' not in name)
+        
+        # Extrapolate to whole model
+        total_layer_params = attn_params + mlp_params + other_params
+        attn_percentage = (attn_params / total_layer_params) * 100
+        mlp_percentage = (mlp_params / total_layer_params) * 100
+        other_percentage = (other_params / total_layer_params) * 100
+        
+        logger.info("\nParameter distribution per layer:")
+        logger.info(f"Attention: {attn_params:,} ({attn_percentage:.2f}%)")
+        logger.info(f"MLP: {mlp_params:,} ({mlp_percentage:.2f}%)")
+        logger.info(f"Other: {other_params:,} ({other_percentage:.2f}%)")
+        
+        # Estimate potential parameter savings
+        for prune_percent in [10, 20, 30, 40, 50]:
+            new_intermediate_size = intermediate_size * (1 - prune_percent/100)
+            new_expansion_ratio = new_intermediate_size / hidden_size
+            param_reduction = (intermediate_size - new_intermediate_size) * (hidden_size + mlp.down_proj.out_features)
+            model_reduction_percent = (param_reduction * len(layers)) / total_params * 100
+            
+            logger.info(f"\nWith {prune_percent}% MLP neuron pruning:")
+            logger.info(f"  New expansion ratio: {new_expansion_ratio:.2f}x ({new_expansion_ratio*100:.2f}%)")
+            logger.info(f"  Estimated parameter reduction: {model_reduction_percent:.2f}% of model")
+
+if __name__ == '__main__':
+    cli()

evaluation/__init__.py

@@ -0,0 +1,5 @@
+"""
+Evaluation package for OptiPFair.
+
+This package provides evaluation utilities for assessing pruned models.
+"""

evaluation/benchmarks.py

@@ -0,0 +1,157 @@
+"""
+Benchmarking utilities for evaluating pruned models.
+
+This module provides tools for evaluating the performance of pruned models
+on standard benchmarks like LAMBADA, BoolQ, etc.
+"""
+
+import logging
+from typing import Dict, List, Any, Union, Optional
+import time
+import torch
+from transformers import PreTrainedModel, AutoTokenizer
+
+logger = logging.getLogger(__name__)
+
+def time_inference(
+    model: PreTrainedModel,
+    tokenizer: AutoTokenizer,
+    prompt: str,
+    max_new_tokens: int = 100,
+    num_runs: int = 5,
+    warmup_runs: int = 2,
+) -> Dict[str, Any]:
+    """
+    Measure inference time for a model.
+    
+    Args:
+        model: Model to evaluate
+        tokenizer: Tokenizer to use
+        prompt: Input prompt for generation
+        max_new_tokens: Maximum number of tokens to generate
+        num_runs: Number of inference runs to average over
+        warmup_runs: Number of initial runs to discard (for warm-up)
+        
+    Returns:
+        Dictionary containing timing results
+    """
+    # Prepare input
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    
+    # Warmup runs
+    for _ in range(warmup_runs):
+        with torch.no_grad():
+            _ = model.generate(
+                inputs.input_ids,
+                max_new_tokens=max_new_tokens,
+                do_sample=False,
+            )
+    
+    # Timed runs
+    times = []
+    for _ in range(num_runs):
+        start_time = time.time()
+        with torch.no_grad():
+            output = model.generate(
+                inputs.input_ids,
+                max_new_tokens=max_new_tokens,
+                do_sample=False,
+            )
+        end_time = time.time()
+        times.append(end_time - start_time)
+    
+    generated_tokens = output.size(1) - inputs.input_ids.size(1)
+    
+    return {
+        "avg_time": sum(times) / len(times),
+        "min_time": min(times),
+        "max_time": max(times),
+        "tokens_per_second": generated_tokens / (sum(times) / len(times)),
+        "num_runs": num_runs,
+        "generated_tokens": generated_tokens,
+    }
+
+def compare_models_inference(
+    original_model: PreTrainedModel,
+    pruned_model: PreTrainedModel,
+    tokenizer: AutoTokenizer,
+    prompts: List[str],
+    max_new_tokens: int = 100,
+) -> Dict[str, Any]:
+    """
+    Compare inference performance between original and pruned models.
+    
+    Args:
+        original_model: Original model before pruning
+        pruned_model: Model after pruning
+        tokenizer: Tokenizer to use
+        prompts: List of input prompts for generation
+        max_new_tokens: Maximum number of tokens to generate
+        
+    Returns:
+        Dictionary containing comparison results
+    """
+    # Ensure models are in eval mode
+    original_model.eval()
+    pruned_model.eval()
+    
+    original_results = []
+    pruned_results = []
+    
+    for prompt in prompts:
+        logger.info(f"Testing prompt: {prompt[:50]}...")
+        
+        original_timing = time_inference(
+            original_model, tokenizer, prompt, max_new_tokens
+        )
+        original_results.append(original_timing)
+        
+        pruned_timing = time_inference(
+            pruned_model, tokenizer, prompt, max_new_tokens
+        )
+        pruned_results.append(pruned_timing)
+    
+    # Aggregate results
+    avg_original_time = sum(r["avg_time"] for r in original_results) / len(original_results)
+    avg_pruned_time = sum(r["avg_time"] for r in pruned_results) / len(pruned_results)
+    
+    avg_original_tps = sum(r["tokens_per_second"] for r in original_results) / len(original_results)
+    avg_pruned_tps = sum(r["tokens_per_second"] for r in pruned_results) / len(pruned_results)
+    
+    speedup = avg_original_time / avg_pruned_time if avg_pruned_time > 0 else float('inf')
+    tps_improvement = (avg_pruned_tps / avg_original_tps - 1) * 100 if avg_original_tps > 0 else float('inf')
+    
+    return {
+        "avg_original_time": avg_original_time,
+        "avg_pruned_time": avg_pruned_time,
+        "avg_original_tokens_per_second": avg_original_tps,
+        "avg_pruned_tokens_per_second": avg_pruned_tps,
+        "speedup": speedup,
+        "tps_improvement_percent": tps_improvement,
+        "num_prompts": len(prompts),
+    }
+
+# Note: This is a placeholder for future implementation
+def evaluate_on_lm_benchmarks(
+    model: PreTrainedModel,
+    benchmarks: List[str] = ["lambada", "boolq"],
+    num_few_shot: int = 0,
+) -> Dict[str, Any]:
+    """
+    Placeholder for evaluating a model on language modeling benchmarks.
+    
+    Args:
+        model: Model to evaluate
+        benchmarks: List of benchmarks to evaluate on
+        num_few_shot: Number of few-shot examples to use
+        
+    Returns:
+        Dictionary containing benchmark results
+    """
+    logger.warning("LM benchmark evaluation not implemented yet. Will be added in a future version.")
+    
+    # This would use libraries like lm-eval-harness, EleutherAI's evaluation framework, etc.
+    return {
+        "message": "Not implemented yet",
+        "benchmarks": benchmarks,
+    }

optipfair-0.1.0.dist-info/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [2025] [Pere Martra]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

optipfair-0.1.0.dist-info/METADATA

@@ -0,0 +1,174 @@
+Metadata-Version: 2.1
+Name: optipfair
+Version: 0.1.0
+Summary: A library for structured pruning & Bias visualization of large language models
+Home-page: https://github.com/peremartra/optipfair
+Author: Pere Martra
+Author-email: peremartra@uadla.com
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: OS Independent
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: torch >=1.10.0
+Requires-Dist: transformers >=4.25.0
+Requires-Dist: tqdm >=4.62.0
+Requires-Dist: click >=8.0.0
+Provides-Extra: dev
+Requires-Dist: pytest >=7.0.0 ; extra == 'dev'
+Requires-Dist: black >=22.0.0 ; extra == 'dev'
+Requires-Dist: flake8 >=4.0.0 ; extra == 'dev'
+Requires-Dist: isort >=5.10.0 ; extra == 'dev'
+Requires-Dist: mypy >=0.900 ; extra == 'dev'
+Provides-Extra: eval
+Requires-Dist: datasets >=2.0.0 ; extra == 'eval'
+Requires-Dist: numpy >=1.20.0 ; extra == 'eval'
+Requires-Dist: pandas >=1.3.0 ; extra == 'eval'
+Requires-Dist: matplotlib >=3.5.0 ; extra == 'eval'
+Requires-Dist: seaborn >=0.11.0 ; extra == 'eval'
+
+# OptiPFair
+
+A Python library for structured pruning of large language models, with a focus on GLU architectures.
+
+## Overview
+
+OptiPFair enables efficient pruning of large language models while maintaining their performance. It implements various structured pruning methods, starting with MLP pruning for GLU architectures (as used in models like LLaMA, Mistral, etc.).
+
+Key features:
+- GLU architecture-aware pruning that preserves model structure 
+- Multiple neuron importance calculation methods
+- Support for both pruning percentage and target expansion rate
+- Simple Python API and CLI interface
+- Progress tracking and detailed statistics
+
+## Installation
+
+```bash
+# From PyPI (not yet available)
+pip install optipfair
+
+# From source
+git clone https://github.com/yourusername/optipfair.git
+cd optipfair
+pip install -e .
+```
+
+## Quick Start
+
+### Python API
+
+```python
+from transformers import AutoModelForCausalLM
+from optipfair import prune_model
+
+# Load a pre-trained model
+model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-1B")
+
+# Prune 20% of neurons from MLP layers using MAW method
+pruned_model, stats = prune_model(
+    model=model,
+    pruning_type="MLP_GLU",
+    neuron_selection_method="MAW",
+    pruning_percentage=20,
+    show_progress=True,
+    return_stats=True
+)
+
+# Print pruning statistics
+print(f"Original parameters: {stats['original_parameters']:,}")
+print(f"Pruned parameters: {stats['pruned_parameters']:,}")
+print(f"Reduction: {stats['reduction']:,} parameters ({stats['percentage_reduction']:.2f}%)")
+
+# Save pruned model
+pruned_model.save_pretrained("./pruned-llama-model")
+```
+
+### Command-Line Interface
+
+```bash
+# Prune a model with default settings (10% pruning, MAW method)
+optipfair prune --model-path meta-llama/Llama-3.2-1B --output-path ./pruned-model
+
+# Prune with custom settings
+optipfair prune \
+  --model-path meta-llama/Llama-3.2-1B \
+  --pruning-type MLP_GLU \
+  --method MAW \
+  --pruning-percentage 20 \
+  --output-path ./pruned-model
+
+# Use expansion rate instead of pruning percentage
+optipfair prune \
+  --model-path meta-llama/Llama-3.2-1B \
+  --expansion-rate 140 \
+  --output-path ./pruned-model
+
+# Analyze a model's architecture
+optipfair analyze --model-path meta-llama/Llama-3.2-1B
+```
+
+## Neuron Selection Methods
+
+OptiPFair supports three methods for calculating neuron importance:
+
+1. **MAW (Maximum Absolute Weight)** - Default method that identifies influential neurons based on the magnitude of their connections. Typically provides the best pruning results.
+
+2. **VOW (Variance of Weights)** - Identifies neurons based on the variance of their weights. May be useful for specific architectures.
+
+3. **PON (Product of Norms)** - Uses the product of L1 norms to identify important neurons. This method may be applicable in certain contexts.
+
+## Documentation
+
+Complete documentation is available at [https://yourusername.github.io/optipfair/](https://yourusername.github.io/optipfair/).
+
+## Supported Models
+
+OptiPFair is designed to work with transformer-based language models that use GLU architecture in their MLP layers, including:
+
+- LLaMA family (LLaMA, LLaMA-2, LLaMA-3)
+- Mistral models
+- And other models with similar GLU architecture
+
+## Expansion Rate vs Pruning Percentage
+
+OptiPFair supports two ways to specify the pruning target:
+
+1. **Pruning Percentage** - Directly specify what percentage of neurons to remove (e.g., 20%)
+
+2. **Expansion Rate** - Specify the target expansion rate (ratio of intermediate size to hidden size) as a percentage (e.g., 140% instead of the default 400%)
+
+The expansion rate approach is often more intuitive when comparing across different model scales.
+
+## Future Roadmap
+
+- Support for attention layer pruning
+- Whole block pruning
+- Quantization-aware pruning
+- Integrated evaluation benchmarks
+- Iterative pruning capabilities
+
+## Citation
+
+If you use OptiPFair in your research, please cite:
+
+```
+@software{optipfair2025,
+  author = {Pere Martra},
+  title = {OptiPFair: A Library for Structured Pruning of Large Language Models},
+  year = {2025},
+  url = {https://github.com/yourusername/optipfair}
+}
+```
+
+## License
+
+Apache 2.0
+
+## Contributing
+
+Contributions are welcome! Please check out our [contributing guidelines](CONTRIBUTING.md) for details.

optipfair-0.1.0.dist-info/RECORD

@@ -0,0 +1,16 @@
+cli/__init__.py,sha256=zXFlgks4qK6E94fExZsJToZIMkEgcBWcSpKKL0vhjjw,80
+cli/commands.py,sha256=tVc0cRvqriWqjXAfnQwXuaslaip33k0POjy06AxUU90,8716
+evaluation/__init__.py,sha256=4Bv7kFHji5lP6LnlcbYtwBrnw8MC7MUwRY85Y2-RgNg,114
+evaluation/benchmarks.py,sha256=lb603AqR7udHA8ru9hKd3eDfV_Pp7Aa-KUXStloEvro,5054
+pruning/__init__.py,sha256=t86KAxqZluSXz4Cz2LofmQpsf8o7thGUcejKOUL7q6o,425
+pruning/mlp_glu.py,sha256=8r7idI_2Dzey16dZ4Q2No0CGj8YRNpRkKYR51koSM0A,10338
+pruning/utils.py,sha256=RlyHmomzrwaRNGhSbUV5-7Lm4mMFCDJM521-zDvfdu4,5051
+pruning/future_extensions/__init__.py,sha256=TJIpxmqHPXBKFaTskK-O9sScAkBWa4_SrteOgk1-oLc,130
+pruning/future_extensions/attention.py,sha256=K-BAgoUpTGyQIoDVfilxbiNA4NBHuQ3MZCm0N_5xLzw,697
+pruning/future_extensions/block.py,sha256=AvOvXYOqKQuCH5-Q4V2T6sM0nLbGVgx1r7q7PEwPLDI,678
+optipfair-0.1.0.dist-info/LICENSE,sha256=jePzZyOOJNMJn9MaXviqb7sKrkuNFnrtMwnBYs22Ilc,11345
+optipfair-0.1.0.dist-info/METADATA,sha256=9ikleULsAieZbljpwv5GtAAWnTbspN2VR5oybtouGSQ,5583
+optipfair-0.1.0.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+optipfair-0.1.0.dist-info/entry_points.txt,sha256=g_jYd69Iz2iPaddlBoXx99W4ATZ8_fQQYRKP0BwYa2s,48
+optipfair-0.1.0.dist-info/top_level.txt,sha256=m910PciT6fKdznCidcHuUZ1K4vxhQprVnpnSm-pBCzs,23
+optipfair-0.1.0.dist-info/RECORD,,

optipfair-0.1.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (75.3.2)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

optipfair-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+optipfair = optipfair.cli:cli

optipfair-0.1.0.dist-info/top_level.txt

@@ -0,0 +1,3 @@
+cli
+evaluation
+pruning

pruning/__init__.py

@@ -0,0 +1,16 @@
+"""
+Pruning package for OptiPFair.
+
+This package provides various pruning methods for transformer-based models.
+"""
+
+from .mlp_glu import prune_model_mlp_glu
+from .utils import validate_model_for_glu_pruning, get_model_layers, count_parameters, get_pruning_statistics
+
+__all__ = [
+    "prune_model_mlp_glu",
+    "validate_model_for_glu_pruning",
+    "get_model_layers",
+    "count_parameters",
+    "get_pruning_statistics",
+]

pruning/future_extensions/__init__.py

@@ -0,0 +1,5 @@
+"""
+Future extensions for OptiPFair pruning methods.
+
+This package will contain additional pruning methods in future releases.
+"""

pruning/future_extensions/attention.py

@@ -0,0 +1,20 @@
+"""
+Attention module pruning (placeholder for future implementation).
+
+This module will implement pruning techniques for attention layers in transformer models.
+"""
+
+# This is a placeholder file for future implementation
+def prune_attention_heads(model, head_importance_method="ATTENTION_WEIGHTS", prune_percentage=10):
+    """
+    Placeholder for future implementation of attention head pruning.
+    
+    Args:
+        model: Model to prune
+        head_importance_method: Method to calculate head importance
+        prune_percentage: Percentage of heads to prune
+        
+    Returns:
+        model: Pruned model
+    """
+    raise NotImplementedError("Attention pruning is not yet implemented.")

pruning/future_extensions/block.py

@@ -0,0 +1,20 @@
+"""
+Block-level pruning (placeholder for future implementation).
+
+This module will implement pruning techniques for entire transformer blocks.
+"""
+
+# This is a placeholder file for future implementation
+def prune_transformer_blocks(model, block_importance_method="PERPLEXITY", prune_percentage=10):
+    """
+    Placeholder for future implementation of transformer block pruning.
+    
+    Args:
+        model: Model to prune
+        block_importance_method: Method to calculate block importance
+        prune_percentage: Percentage of blocks to prune
+        
+    Returns:
+        model: Pruned model
+    """
+    raise NotImplementedError("Block pruning is not yet implemented.")

pruning/mlp_glu.py

@@ -0,0 +1,251 @@
+"""
+MLPGLUPruning - Module for pruning MLP layers with GLU architecture in transformer models.
+
+This module provides functionality to prune neurons in MLP layers that follow the
+Gated Linear Unit (GLU) architecture, as used in models like LLaMA. The pruning
+is structured to maintain the paired nature of gate_proj and up_proj layers.
+"""
+
+import torch
+from torch import nn
+import logging
+from typing import Tuple, Dict, List, Optional, Callable, Union, Any
+from tqdm import tqdm
+from transformers import PreTrainedModel
+
+from .utils import validate_model_for_glu_pruning, get_model_layers
+
+logger = logging.getLogger(__name__)
+
+def compute_neuron_pair_importance_maw(gate_weight: torch.Tensor, up_weight: torch.Tensor) -> torch.Tensor:
+    """
+    Compute neuron pair importance scores using Maximum Absolute Weight method.
+    
+    Args:
+        gate_weight: Weight matrix from the gate_proj layer
+        up_weight: Weight matrix from the up_proj layer
+        
+    Returns:
+        importance_scores: Importance scores for each neuron pair
+    """
+    gate_max_abs = torch.max(gate_weight, dim=1).values + torch.abs(torch.min(gate_weight, dim=1).values)
+    up_max_abs = torch.max(up_weight, dim=1).values + torch.abs(torch.min(up_weight, dim=1).values)
+    importance_scores = gate_max_abs + up_max_abs
+    return importance_scores
+
+def compute_neuron_pair_importance_vow(gate_weight: torch.Tensor, up_weight: torch.Tensor) -> torch.Tensor:
+    """
+    Compute neuron pair importance scores using Variance of Weights method.
+    
+    Args:
+        gate_weight: Weight matrix from the gate_proj layer
+        up_weight: Weight matrix from the up_proj layer
+        
+    Returns:
+        importance_scores: Importance scores for each neuron pair
+    """
+    gate_variance = torch.var(gate_weight, dim=1)
+    up_variance = torch.var(up_weight, dim=1)
+    importance_scores = gate_variance + up_variance
+    return importance_scores
+
+def compute_neuron_pair_importance_pon(gate_weight: torch.Tensor, up_weight: torch.Tensor) -> torch.Tensor:
+    """
+    Compute neuron pair importance scores using Product of Norms method.
+    
+    Args:
+        gate_weight: Weight matrix from the gate_proj layer
+        up_weight: Weight matrix from the up_proj layer
+        
+    Returns:
+        importance_scores: Importance scores for each neuron pair
+    """
+    gate_norms = torch.norm(gate_weight, p=1, dim=1)
+    up_norms = torch.norm(up_weight, p=1, dim=1)
+    importance_scores = gate_norms * up_norms
+    return importance_scores
+
+# Dictionary mapping method names to their respective functions
+IMPORTANCE_FUNCTIONS = {
+    "MAW": compute_neuron_pair_importance_maw,
+    "VOW": compute_neuron_pair_importance_vow,
+    "PON": compute_neuron_pair_importance_pon,
+}
+
+def prune_neuron_pairs(
+    mlp: nn.Module,
+    prune_percentage: float,
+    importance_fn: Callable = compute_neuron_pair_importance_maw
+) -> Tuple[nn.Linear, nn.Linear, nn.Linear, int]:
+    """
+    Prune a specific percentage of neurons from the MLP layers (GLU architecture).
+    
+    Args:
+        mlp: MLP module containing gate_proj, up_proj, and down_proj layers
+        prune_percentage: Percentage of neurons to prune (0-100)
+        importance_fn: Function to compute neuron pair importance
+        
+    Returns:
+        new_gate_proj: Pruned gate_proj layer
+        new_up_proj: Pruned up_proj layer
+        new_down_proj: Pruned down_proj layer
+        k: New intermediate size after pruning
+    """
+    # Extract the weights from the MLP layers
+    gate_weight = mlp.gate_proj.weight.data.float()
+    up_weight = mlp.up_proj.weight.data.float()
+    
+    # Compute importance scores for neuron pairs
+    importance_scores = importance_fn(gate_weight, up_weight)
+    
+    # Determine the new intermediate size
+    original_intermediate_size = gate_weight.size(0)
+    num_neuron_pairs_to_prune = min(int(prune_percentage / 100 * original_intermediate_size), original_intermediate_size - 1)
+    k = original_intermediate_size - num_neuron_pairs_to_prune
+    
+    # Validate the new size
+    if k <= 0:
+        raise ValueError(f"Invalid number of neuron pairs to keep: {k}. Reduce pruning percentage.")
+    
+    # Select the neurons to keep based on importance scores
+    _, indices_to_keep = torch.topk(importance_scores, k, largest=True)
+    indices_to_keep = indices_to_keep.sort().values
+    
+    # Create new layers with reduced dimensions
+    new_gate_proj = nn.Linear(mlp.gate_proj.in_features, k, bias=mlp.gate_proj.bias is not None)
+    new_up_proj = nn.Linear(mlp.up_proj.in_features, k, bias=mlp.up_proj.bias is not None)
+    new_down_proj = nn.Linear(k, mlp.down_proj.out_features, bias=mlp.down_proj.bias is not None)
+    
+    # Copy selected weights to the new layers
+    new_gate_proj.weight.data = gate_weight[indices_to_keep, :]
+    if mlp.gate_proj.bias is not None:
+        new_gate_proj.bias.data = mlp.gate_proj.bias.data[indices_to_keep]
+    
+    new_up_proj.weight.data = up_weight[indices_to_keep, :]
+    if mlp.up_proj.bias is not None:
+        new_up_proj.bias.data = mlp.up_proj.bias.data[indices_to_keep]
+    
+    new_down_proj.weight.data = mlp.down_proj.weight.data[:, indices_to_keep]
+    if mlp.down_proj.bias is not None:
+        new_down_proj.bias.data = mlp.down_proj.bias.data.clone()
+    
+    return new_gate_proj, new_up_proj, new_down_proj, k
+
+def calculate_pruning_percentage_from_expansion_rate(
+    current_intermediate_size: int,
+    current_hidden_size: int,
+    target_expansion_rate: float
+) -> float:
+    """
+    Calculate the pruning percentage needed to achieve a target expansion rate.
+    
+    Args:
+        current_intermediate_size: Current size of the intermediate layer
+        current_hidden_size: Current size of the hidden layer
+        target_expansion_rate: Target expansion rate in percentage (e.g., 140 for 140%)
+        
+    Returns:
+        pruning_percentage: Percentage of neurons to prune
+    """
+    current_expansion_rate = (current_intermediate_size / current_hidden_size) * 100
+    target_intermediate_size = (target_expansion_rate / 100) * current_hidden_size
+    
+    if target_intermediate_size >= current_intermediate_size:
+        raise ValueError(
+            f"Target expansion rate ({target_expansion_rate}%) would increase the model size. "
+            f"Current expansion rate is {current_expansion_rate:.2f}%."
+        )
+    
+    pruning_percentage = (1 - (target_intermediate_size / current_intermediate_size)) * 100
+    return pruning_percentage
+
+def prune_model_mlp_glu(
+    model: PreTrainedModel,
+    neuron_selection_method: str = "MAW",
+    pruning_percentage: Optional[float] = 10,
+    expansion_rate: Optional[float] = None,
+    show_progress: bool = True,
+) -> PreTrainedModel:
+    """
+    Prune the MLP layers in a model with GLU architecture.
+    
+    Args:
+        model: Pre-trained model to prune
+        neuron_selection_method: Method to use for calculating neuron importance ("MAW", "VOW", or "PON")
+        pruning_percentage: Percentage of neurons to prune (0-100)
+        expansion_rate: Target expansion rate in percentage (mutually exclusive with pruning_percentage)
+        show_progress: Whether to show progress during pruning
+        
+    Returns:
+        model: Pruned model
+    """
+    # Validate the model for GLU pruning
+    if not validate_model_for_glu_pruning(model):
+        raise ValueError("Model is not compatible with GLU pruning. It must have gate_proj, up_proj, and down_proj layers.")
+    
+    # Select the appropriate importance function
+    if neuron_selection_method not in IMPORTANCE_FUNCTIONS:
+        raise ValueError(f"Invalid neuron selection method: {neuron_selection_method}. "
+                         f"Choose from {list(IMPORTANCE_FUNCTIONS.keys())}.")
+    
+    importance_fn = IMPORTANCE_FUNCTIONS[neuron_selection_method]
+    
+    # Handle mutually exclusive parameters
+    if pruning_percentage is not None and expansion_rate is not None:
+        raise ValueError("pruning_percentage and expansion_rate are mutually exclusive. Provide only one.")
+    
+    if expansion_rate is not None:
+        # Get the first MLP layer to calculate current expansion rate
+        layers = get_model_layers(model)
+        if not layers:
+            raise ValueError("Could not find MLP layers in the model.")
+        
+        first_mlp = layers[0].mlp
+        current_intermediate_size = first_mlp.gate_proj.out_features
+        current_hidden_size = first_mlp.gate_proj.in_features
+        
+        pruning_percentage = calculate_pruning_percentage_from_expansion_rate(
+            current_intermediate_size, current_hidden_size, expansion_rate
+        )
+        
+        logger.info(f"Calculated pruning percentage: {pruning_percentage:.2f}% to achieve "
+                   f"expansion rate of {expansion_rate}%")
+    
+    # Ensure pruning_percentage is within valid range
+    if not 0 <= pruning_percentage <= 100:
+        raise ValueError(f"pruning_percentage must be between 0 and 100, got {pruning_percentage}")
+    
+    # Get all layers to prune
+    layers = get_model_layers(model)
+    if not layers:
+        raise ValueError("Could not find MLP layers in the model.")
+    
+    new_intermediate_size = None
+    
+    # Wrap with tqdm if show_progress is True
+    layer_iterator = tqdm(enumerate(layers), total=len(layers), desc="Pruning layers") if show_progress else enumerate(layers)
+    
+    # Iterate through layers and apply pruning
+    for idx, layer in layer_iterator:
+        mlp = layer.mlp
+        
+        # Prune neuron pairs in the MLP layer
+        new_gate_proj, new_up_proj, new_down_proj, new_size = prune_neuron_pairs(
+            mlp, pruning_percentage, importance_fn
+        )
+        
+        # Replace original layers with pruned layers
+        mlp.gate_proj = new_gate_proj
+        mlp.up_proj = new_up_proj
+        mlp.down_proj = new_down_proj
+        
+        # Store the new intermediate size (should be the same for all layers)
+        if new_intermediate_size is None:
+            new_intermediate_size = new_size
+    
+    # Update model configuration
+    if hasattr(model, "config") and hasattr(model.config, "intermediate_size"):
+        model.config.intermediate_size = new_intermediate_size
+        logger.info(f"Updated model config: intermediate_size = {new_intermediate_size}")
+    
+    return model

pruning/utils.py

@@ -0,0 +1,148 @@
+"""
+Utility functions for the OptiPFair library.
+
+This module provides helper functions for model compatibility checking,
+layer extraction, and other common tasks needed across different pruning methods.
+"""
+
+import torch
+from typing import List, Optional, Union, Any, Dict
+import logging
+from transformers import PreTrainedModel
+
+logger = logging.getLogger(__name__)
+
+def validate_model_for_glu_pruning(model: PreTrainedModel) -> bool:
+    """
+    Validate that a model is compatible with GLU pruning.
+    
+    Args:
+        model: Model to validate
+        
+    Returns:
+        bool: True if the model is compatible, False otherwise
+    """
+    # Check if the model has the expected structure
+    try:
+        layers = get_model_layers(model)
+        if not layers:
+            logger.warning("Could not find decoder layers in the model")
+            return False
+        
+        # Check the first layer for GLU components
+        first_layer = layers[0]
+        if not hasattr(first_layer, 'mlp'):
+            logger.warning("Model layers do not have 'mlp' attribute")
+            return False
+        
+        mlp = first_layer.mlp
+        required_attributes = ['gate_proj', 'up_proj', 'down_proj']
+        for attr in required_attributes:
+            if not hasattr(mlp, attr):
+                logger.warning(f"MLP does not have required attribute: {attr}")
+                return False
+            
+            # Verify these are linear layers
+            layer = getattr(mlp, attr)
+            if not isinstance(layer, torch.nn.Linear):
+                logger.warning(f"{attr} is not a Linear layer")
+                return False
+        
+        # Verify gate_proj and up_proj have the same dimensions
+        if mlp.gate_proj.in_features != mlp.up_proj.in_features:
+            logger.warning("gate_proj and up_proj have different input dimensions")
+            return False
+            
+        if mlp.gate_proj.out_features != mlp.up_proj.out_features:
+            logger.warning("gate_proj and up_proj have different output dimensions")
+            return False
+            
+        if mlp.down_proj.in_features != mlp.gate_proj.out_features:
+            logger.warning("down_proj input dimensions don't match gate_proj output dimensions")
+            return False
+            
+        return True
+    
+    except Exception as e:
+        logger.warning(f"Error validating model for GLU pruning: {str(e)}")
+        return False
+
+def get_model_layers(model: PreTrainedModel) -> List[Any]:
+    """
+    Extract transformer layers from a pre-trained model.
+    Currently supports LLaMA, Mistral, and similar model architectures.
+    
+    Args:
+        model: Pre-trained model
+        
+    Returns:
+        List of decoder layers that contain MLP blocks
+    """
+    # Try different attribute paths based on common model architectures
+    if hasattr(model, 'model') and hasattr(model.model, 'layers'):
+        # LLaMA, Mistral, and similar architectures
+        return list(model.model.layers)
+    elif hasattr(model, 'transformer') and hasattr(model.transformer, 'h'):
+        # GPT-2 and similar architectures
+        return list(model.transformer.h)
+    elif hasattr(model, 'encoder') and hasattr(model.encoder, 'layer'):
+        # BERT and similar architectures
+        return list(model.encoder.layer)
+    elif hasattr(model, 'layers'):
+        # Direct layers attribute
+        return list(model.layers)
+        
+    logger.warning("Could not find layers in the model")
+    return []
+
+def count_parameters(model: torch.nn.Module) -> int:
+    """
+    Count the number of trainable parameters in a model.
+    
+    Args:
+        model: PyTorch model
+        
+    Returns:
+        Number of trainable parameters
+    """
+    return sum(p.numel() for p in model.parameters() if p.requires_grad)
+
+def get_pruning_statistics(
+    original_model: torch.nn.Module,
+    pruned_model: torch.nn.Module,
+) -> Dict[str, Any]:
+    """
+    Calculate statistics about the pruning operation.
+    
+    Args:
+        original_model: Original model before pruning
+        pruned_model: Model after pruning
+        
+    Returns:
+        Dictionary containing pruning statistics
+    """
+    original_params = count_parameters(original_model)
+    pruned_params = count_parameters(pruned_model)
+    
+    reduction = original_params - pruned_params
+    percentage_reduction = (reduction / original_params) * 100
+    
+    # Get expansion rate if possible
+    expansion_rate = None
+    try:
+        layers = get_model_layers(pruned_model)
+        if layers:
+            first_mlp = layers[0].mlp
+            intermediate_size = first_mlp.gate_proj.out_features
+            hidden_size = first_mlp.gate_proj.in_features
+            expansion_rate = (intermediate_size / hidden_size) * 100
+    except Exception:
+        pass
+    
+    return {
+        "original_parameters": original_params,
+        "pruned_parameters": pruned_params,
+        "reduction": reduction,
+        "percentage_reduction": percentage_reduction,
+        "expansion_rate": expansion_rate
+    }