Upload 8 files

rnn/dot/best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e769ed3b93d68a94881fc269f9a8fbc40d10862b0039f6db3079a6d608caeab2
+size 443012514

rnn/dot/config.yaml

@@ -0,0 +1,45 @@
+# ------------- 分词器 -----------------
+tokenizer: tokenizer.NltkTokenizer     # 可改为自定义类的全路径
+
+# ------------- 模型结构 (针对 100K 数据 & 8G 显存优化) ----------------
+model:
+  type: "rnn"
+  encoder_embedding_dim: 256
+  decoder_embedding_dim: 256
+  hidden_dim: 512       # GRU 隐藏层维度
+  n_layers: 2           # 作业要求: 2 layers
+  dropout: 0.1
+  # 关键参数: 探索不同的 attention 方式
+  # 可选项: "dot" (默认), "general", "concat"
+  attention_method: "dot"
+
+# ------------- 训练超参 ----------------
+train:
+  batch_size: 32
+  epochs: 10
+  lr: 0.0005
+  weight_decay: 0.0001
+  clip: 1.0             # 梯度裁剪阈值
+  # 学习率调度参数
+  lr_patience: 2
+  lr_factor: 0.5
+  save_dir: runs/train/Nltk_100k
+  num_workers: 0
+
+# ------------- 数据路径 ----------------
+data:
+  raw_train:      data/train_100k.jsonl # 输入文件不变
+  raw_val:        data/valid.jsonl
+  raw_test:       data/test.jsonl
+
+  processed_dir:  data/processed_nltk_100k
+  train_processed: data/processed_nltk_100k/train.jsonl
+  val_processed:   data/processed_nltk_100k/val.jsonl
+  test_processed:  data/processed_nltk_100k/test.jsonl
+
+  src_vocab:      data/processed_nltk_100k/src_vocab.pkl
+  tgt_vocab:      data/processed_nltk_100k/tgt_vocab.pkl
+  min_freq: 2
+
+# ------------- 其余 --------------------
+seed: 3407          # 固定随机种子，保证可复现

rnn/dot/log.txt

@@ -0,0 +1,103 @@
+[Info] Logging started. Output will be saved to runs\train\Nltk_100k\rnn\20251227_2104\log.txt
+[Info] Model Architecture: RNN
+Seq2SeqRNN(
+  (encoder): EncoderRNN(
+    (embedding): Embedding(34122, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+  )
+  (decoder): DecoderRNN(
+    (embedding): Embedding(28647, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+    (attention): Attention(
+      (attn): Linear(in_features=512, out_features=512, bias=True)
+    )
+    (concat): Linear(in_features=1024, out_features=512, bias=True)
+    (out): Linear(in_features=512, out_features=28647, bias=True)
+  )
+)
+[Info] Total Parameters: 37,069,543
+[Info] Trainable Parameters: 37,069,543
+[Info] Model Size (approx): 141.41 MB
+===== Epoch 1/20 =====
+[Info] Logging started. Output will be saved to runs\train\Nltk_100k\rnn\20251227_2104\log.txt
+[Info] Model Architecture: RNN
+Seq2SeqRNN(
+  (encoder): EncoderRNN(
+    (embedding): Embedding(34122, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+  )
+  (decoder): DecoderRNN(
+    (embedding): Embedding(28647, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+    (attention): Attention()
+    (concat): Linear(in_features=1024, out_features=512, bias=True)
+    (out): Linear(in_features=512, out_features=28647, bias=True)
+  )
+)
+[Info] Total Parameters: 36,806,887
+[Info] Trainable Parameters: 36,806,887
+[Info] Model Size (approx): 140.41 MB
+===== Epoch 1/20 =====
+Epoch 01 | Time: 22m29s | train loss: 6.1225 | val loss: 6.1888
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 1
+===== Epoch 2/20 =====
+Epoch 02 | Time: 19m42s | train loss: 5.4501 | val loss: 5.9591
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 2
+===== Epoch 3/20 =====
+Epoch 03 | Time: 19m45s | train loss: 5.1753 | val loss: 5.8372
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 3
+===== Epoch 4/20 =====
+Epoch 04 | Time: 19m45s | train loss: 4.9817 | val loss: 5.7523
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 4
+===== Epoch 5/20 =====
+Epoch 05 | Time: 19m51s | train loss: 4.7887 | val loss: 5.6182
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 5
+===== Epoch 6/20 =====
+Epoch 06 | Time: 19m46s | train loss: 4.6178 | val loss: 5.5535
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 6
+===== Epoch 7/20 =====
+Epoch 07 | Time: 19m47s | train loss: 4.4872 | val loss: 5.5063
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 7
+===== Epoch 8/20 =====
+Epoch 08 | Time: 19m43s | train loss: 4.3829 | val loss: 5.4677
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 8
+===== Epoch 9/20 =====
+Epoch 09 | Time: 19m46s | train loss: 4.2980 | val loss: 5.4273
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 9
+===== Epoch 10/20 =====
+Epoch 10 | Time: 19m46s | train loss: 4.2265 | val loss: 5.4023
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 10
+===== Epoch 11/20 =====
+Epoch 11 | Time: 19m49s | train loss: 4.1654 | val loss: 5.3904
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 11
+===== Epoch 12/20 =====
+Epoch 12 | Time: 19m45s | train loss: 4.1105 | val loss: 5.4086
+===== Epoch 13/20 =====
+Epoch 13 | Time: 19m43s | train loss: 4.0633 | val loss: 5.3844
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 13
+===== Epoch 14/20 =====
+Epoch 14 | Time: 19m45s | train loss: 4.0223 | val loss: 5.3724
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 14
+===== Epoch 15/20 =====
+Epoch 15 | Time: 19m44s | train loss: 3.9840 | val loss: 5.3649
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 15
+===== Epoch 16/20 =====
+Epoch 16 | Time: 19m45s | train loss: 3.9500 | val loss: 5.3559
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 16
+===== Epoch 17/20 =====
+Epoch 17 | Time: 19m43s | train loss: 3.9202 | val loss: 5.3502
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 17
+===== Epoch 18/20 =====
+Epoch 18 | Time: 19m48s | train loss: 3.8926 | val loss: 5.3545
+===== Epoch 19/20 =====
+Epoch 19 | Time: 19m44s | train loss: 3.8676 | val loss: 5.3607
+===== Epoch 20/20 =====
+Epoch 20 | Time: 19m47s | train loss: 3.7420 | val loss: 5.3310
+New best model saved to runs\train\Nltk_100k\rnn\20251227_2104\best_model.pt, best epoch: 20
+Model for epoch 20 saved to runs\train\Nltk_100k\rnn\20251227_2104\model_epoch_20.pt
+
+Training finished! Total Epochs: 20 Total Time: 398m12s

rnn/dot/model_epoch_20.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69a74bee49ed6b8d6184ab954ca49a3864f68eaa698efd970640046e804dbc9e
+size 443012826

rnn/general/best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cb4db4628635881ed1df9ca9d54f0104a339f881edcbb3e82f4293a14794e95
+size 446166690

rnn/general/config.yaml

@@ -0,0 +1,45 @@
+# ------------- 分词器 -----------------
+tokenizer: tokenizer.NltkTokenizer     # 可改为自定义类的全路径
+
+# ------------- 模型结构 (针对 100K 数据 & 8G 显存优化) ----------------
+model:
+  type: "rnn"
+  encoder_embedding_dim: 256
+  decoder_embedding_dim: 256
+  hidden_dim: 512       # GRU 隐藏层维度
+  n_layers: 2           # 作业要求: 2 layers
+  dropout: 0.1
+  # 关键参数: 探索不同的 attention 方式
+  # 可选项: "dot" (默认), "general", "concat"
+  attention_method: "general"
+
+# ------------- 训练超参 ----------------
+train:
+  batch_size: 32
+  epochs: 20
+  lr: 0.0005
+  weight_decay: 0.0001
+  clip: 1.0             # 梯度裁剪阈值
+  # 学习率调度参数
+  lr_patience: 2
+  lr_factor: 0.5
+  save_dir: runs/train/Nltk_100k
+  num_workers: 0
+
+# ------------- 数据路径 ----------------
+data:
+  raw_train:      data/train_100k.jsonl # 输入文件不变
+  raw_val:        data/valid.jsonl
+  raw_test:       data/test.jsonl
+
+  processed_dir:  data/processed_nltk_100k
+  train_processed: data/processed_nltk_100k/train.jsonl
+  val_processed:   data/processed_nltk_100k/val.jsonl
+  test_processed:  data/processed_nltk_100k/test.jsonl
+
+  src_vocab:      data/processed_nltk_100k/src_vocab.pkl
+  tgt_vocab:      data/processed_nltk_100k/tgt_vocab.pkl
+  min_freq: 2
+
+# ------------- 其余 --------------------
+seed: 3407          # 固定随机种子，保证可复现

rnn/general/log.txt

@@ -0,0 +1,80 @@
+[Info] Logging started. Output will be saved to runs\train\Nltk_100k\rnn\20251228_0557\log.txt
+[Info] Model Architecture: RNN
+Seq2SeqRNN(
+  (encoder): EncoderRNN(
+    (embedding): Embedding(34122, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+  )
+  (decoder): DecoderRNN(
+    (embedding): Embedding(28647, 256)
+    (gru): GRU(256, 512, num_layers=2, batch_first=True, dropout=0.1)
+    (dropout): Dropout(p=0.1, inplace=False)
+    (attention): Attention(
+      (attn): Linear(in_features=512, out_features=512, bias=True)
+    )
+    (concat): Linear(in_features=1024, out_features=512, bias=True)
+    (out): Linear(in_features=512, out_features=28647, bias=True)
+  )
+)
+[Info] Total Parameters: 37,069,543
+[Info] Trainable Parameters: 37,069,543
+[Info] Model Size (approx): 141.41 MB
+===== Epoch 1/20 =====
+Epoch 01 | Time: 21m34s | train loss: 6.1383 | val loss: 6.1445
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 1
+===== Epoch 2/20 =====
+Epoch 02 | Time: 22m49s | train loss: 5.3770 | val loss: 5.9185
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 2
+===== Epoch 3/20 =====
+Epoch 03 | Time: 21m51s | train loss: 5.0793 | val loss: 5.7854
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 3
+===== Epoch 4/20 =====
+Epoch 04 | Time: 21m30s | train loss: 4.8973 | val loss: 5.6876
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 4
+===== Epoch 5/20 =====
+Epoch 05 | Time: 21m36s | train loss: 4.7746 | val loss: 5.6771
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 5
+===== Epoch 6/20 =====
+Epoch 06 | Time: 21m45s | train loss: 4.6856 | val loss: 5.6152
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 6
+===== Epoch 7/20 =====
+Epoch 07 | Time: 21m45s | train loss: 4.6145 | val loss: 5.5739
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 7
+===== Epoch 8/20 =====
+Epoch 08 | Time: 21m43s | train loss: 4.5574 | val loss: 5.5883
+===== Epoch 9/20 =====
+Epoch 09 | Time: 21m46s | train loss: 4.5090 | val loss: 5.5381
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 9
+===== Epoch 10/20 =====
+Epoch 10 | Time: 21m45s | train loss: 4.4682 | val loss: 5.5439
+===== Epoch 11/20 =====
+Epoch 11 | Time: 21m35s | train loss: 4.4376 | val loss: 5.5325
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 11
+===== Epoch 12/20 =====
+Epoch 12 | Time: 21m44s | train loss: 4.4089 | val loss: 5.5030
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 12
+===== Epoch 13/20 =====
+Epoch 13 | Time: 21m48s | train loss: 4.3860 | val loss: 5.5052
+===== Epoch 14/20 =====
+Epoch 14 | Time: 21m42s | train loss: 4.3655 | val loss: 5.4956
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 14
+===== Epoch 15/20 =====
+Epoch 15 | Time: 21m59s | train loss: 4.3467 | val loss: 5.4969
+===== Epoch 16/20 =====
+Epoch 16 | Time: 21m51s | train loss: 4.3303 | val loss: 5.4816
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 16
+===== Epoch 17/20 =====
+Epoch 17 | Time: 21m38s | train loss: 4.3171 | val loss: 5.4770
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 17
+===== Epoch 18/20 =====
+Epoch 18 | Time: 21m43s | train loss: 4.3050 | val loss: 5.4545
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 18
+===== Epoch 19/20 =====
+Epoch 19 | Time: 21m40s | train loss: 4.2943 | val loss: 5.4553
+===== Epoch 20/20 =====
+Epoch 20 | Time: 21m41s | train loss: 4.2881 | val loss: 5.4534
+New best model saved to runs\train\Nltk_100k\rnn\20251228_0557\best_model.pt, best epoch: 20
+Model for epoch 20 saved to runs\train\Nltk_100k\rnn\20251228_0557\model_epoch_20.pt
+
+Training finished! Total Epochs: 20 Total Time: 435m49s

rnn/general/model_epoch_20.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70d60b7a5f0764a05f7b127741a175beabd41bebc303d5a19b4338a24eaa2cf2
+size 446167034