| import torch |
| import numpy as np |
| from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig |
| from datasets import load_dataset |
| import gc |
| import os |
|
|
| |
| selected_model = "qwen-moe-a2.7b" |
| LOAD_PATH = "outputs/qwen-moe-a2.7b_20251009_084529" |
|
|
| input_file = os.path.join(LOAD_PATH, 'tokenized_input.npz') |
| metadata_file = os.path.join(LOAD_PATH, 'metadata.txt') |
| attention_file = os.path.join(LOAD_PATH, 'attention_matrices_multihead.npz') |
| routing_file = os.path.join(LOAD_PATH, 'routing_matrices.npz') |
|
|
| |
| print("--- Metadata ---") |
| with open(metadata_file, 'r') as f: |
| print(f.read()) |
|
|
| |
| print("\n--- Input data ---") |
|
|
| input_data = np.load(input_file) |
| print(f"File size: {os.path.getsize(input_file) / (1024**2):.2f} MB") |
|
|
| input_ids = input_data['input_ids'] |
| print(f"\nTokenized Input:") |
| print(f"Shape: {input_ids.shape}") |
| print(f"First 20 token IDs: {input_ids[:20]}") |
|
|
|
|
| input_text = str(input_data['input_text']) |
| print(f"\nOriginal text length: {len(input_text)} characters") |
| print(f"First 200 characters: {input_text[:200]}...") |
|
|
| |
| print("\n--- Attention matrices ---") |
| attn_data = np.load(attention_file, allow_pickle=True) |
| print(f"File size: {os.path.getsize(attention_file) / (1024**2):.2f} MB") |
| print(f"Available layers: {len(attn_data.files)}") |
| print(f"Layer names: {attn_data.files}") |
|
|
| |
| layer_0 = attn_data['layer_0'] |
| print(f"\nLayer 0 attention matrix:") |
| print(f"Shape: {layer_0.shape} [num_heads, seq_len, seq_len]") |
| print(f"Number of heads: {layer_0.shape[0]}") |
| print(f"Min value: {layer_0.min():.6f}") |
| print(f"Max value: {layer_0.max():.6f}") |
| print(f"Mean value: {layer_0.mean():.6f}") |
|
|
| print(f"\n Head 0 attention sample (first 5x5 tokens):") |
| print(layer_0[0, :5, :5]) |
|
|
|
|
| |
| print("\n--- Routing matrices ---") |
| routing_data = np.load(routing_file) |
| print(f"File size: {os.path.getsize(routing_file) / (1024**2):.2f} MB") |
| print(f"Available layers: {len(routing_data.files)}") |
| print(f"Layer names: {routing_data.files}") |
| layer_0 = routing_data['layer_0'] |
|
|
| print(f"Number of experts: {len(layer_0[0])}") |
|
|
| print(f"\nRouting logits sample (first 5 tokens, all experts):") |
| print(layer_0[:5, :]) |
|
|
