yujunjun/LLMs-from-scratch
1
0
Fork 0
mirror of https://github.com/rasbt/LLMs-from-scratch.git synced 2025-08-31 03:50:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove reundant dropout in MLP module (#105)

Browse Source
This commit is contained in:
Sebastian Raschka 2024-04-03 20:19:08 -05:00 committed by GitHub
parent edcae09884
commit 5beff4e25a
11 changed files with 202 additions and 266 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

102
appendix-D/01_main-chapter-code/appendix-D.ipynb
View File

File diff suppressed because one or more lines are too long

1
appendix-D/01_main-chapter-code/previous_chapters.py
View File

@ -152,7 +152,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):

152
ch04/01_main-chapter-code/ch04.ipynb
View File

File diff suppressed because one or more lines are too long

24
ch04/01_main-chapter-code/exercise-solutions.ipynb
View File

@ -253,7 +253,7 @@
},
{
"cell_type": "code",
"execution_count": 6,
"execution_count": 1,
"id": "5fee2cf5-61c3-4167-81b5-44ea155bbaf2",
"metadata": {},
"outputs": [],
@ -265,7 +265,6 @@
" \"n_heads\": 12,\n",
" \"n_layers\": 12,\n",
" \"drop_rate_emb\": 0.1, # NEW: dropout for embedding layers\n",
" \"drop_rate_ffn\": 0.1, # NEW: dropout for feed forward module\n",
" \"drop_rate_attn\": 0.1, # NEW: dropout for multi-head attention \n",
" \"drop_rate_resid\": 0.1, # NEW: dropout for residual connections \n",
" \"qkv_bias\": False\n",
@ -274,26 +273,13 @@
},
{
"cell_type": "code",
"execution_count": 7,
"execution_count": 2,
"id": "5aa1b0c1-d78a-48fc-ad08-4802458b43f7",
"metadata": {},
"outputs": [],
"source": [
"import torch.nn as nn\n",
"from gpt import MultiHeadAttention, LayerNorm, GELU\n",
"\n",
"class FeedForward(nn.Module):\n",
" def __init__(self, cfg):\n",
" super().__init__()\n",
" self.layers = nn.Sequential(\n",
" nn.Linear(cfg[\"emb_dim\"], 4 * cfg[\"emb_dim\"]),\n",
" GELU(),\n",
" nn.Linear(4 * cfg[\"emb_dim\"], cfg[\"emb_dim\"]),\n",
" nn.Dropout(cfg[\"drop_rate_ffn\"]) # NEW: dropout for feed forward module\n",
" )\n",
"\n",
" def forward(self, x):\n",
" return self.layers(x)\n",
"from gpt import MultiHeadAttention, LayerNorm, GELU, FeedForward\n",
"\n",
"\n",
"class TransformerBlock(nn.Module):\n",
@ -356,7 +342,7 @@
},
{
"cell_type": "code",
"execution_count": 8,
"execution_count": 3,
"id": "1d013d32-c275-4f42-be21-9010f1537227",
"metadata": {},
"outputs": [],
@ -384,7 +370,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.6"
"version": "3.11.4"
}
},
"nbformat": 4,

1
ch04/01_main-chapter-code/gpt.py
View File

@ -144,7 +144,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):

118
ch05/01_main-chapter-code/ch05.ipynb
View File

File diff suppressed because one or more lines are too long

66
ch05/01_main-chapter-code/exercise-solutions.ipynb
View File

@ -222,7 +222,7 @@
},
{
"cell_type": "code",
"execution_count": 1,
"execution_count": 4,
"id": "a61a4034-797a-4635-bf42-ddfff1b07125",
"metadata": {},
"outputs": [],
@ -253,7 +253,7 @@
},
{
"cell_type": "code",
"execution_count": 2,
"execution_count": 5,
"id": "ee95a272-b852-43b4-9827-ea7e1dbd5724",
"metadata": {},
"outputs": [],
@ -264,7 +264,7 @@
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": 6,
"id": "4ab43658-3240-484a-9072-a40a0ed85be6",
"metadata": {},
"outputs": [
@ -273,11 +273,7 @@
"output_type": "stream",
"text": [
"Output text:\n",
" Every effort moves you?\"\n",
"\n",
"\"Yes--quite insensible to the irony. She wanted him vindicated--and by me!\"\n",
"\n",
"\n"
" Every effort moves you know,\" was one of the axioms he laid down across the Sevres and silver of an exquisitely appointed lun\n"
]
}
],
@ -298,7 +294,7 @@
},
{
"cell_type": "code",
"execution_count": 4,
"execution_count": 7,
"id": "ebb22d06-393a-42d3-ab64-66646d33b39b",
"metadata": {},
"outputs": [
@ -307,11 +303,7 @@
"output_type": "stream",
"text": [
"Output text:\n",
" Every effort moves you?\"\n",
"\n",
"\"Yes--quite insensible to the irony. She wanted him vindicated--and by me!\"\n",
"\n",
"\n"
" Every effort moves you know,\" was one of the axioms he laid down across the Sevres and silver of an exquisitely appointed lun\n"
]
}
],
@ -332,7 +324,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 8,
"id": "75469f24-47cc-458d-a200-fe64c648131d",
"metadata": {},
"outputs": [
@ -341,11 +333,7 @@
"output_type": "stream",
"text": [
"Output text:\n",
" Every effort moves you?\"\n",
"\n",
"\"Yes--quite insensible to the irony. She wanted him vindicated--and by me!\"\n",
"\n",
"\n"
" Every effort moves you know,\" was one of the axioms he laid down across the Sevres and silver of an exquisitely appointed lun\n"
]
}
],
@ -412,7 +400,7 @@
"model = GPTModel(GPT_CONFIG_124M)\n",
"model.load_state_dict(checkpoint[\"model_state_dict\"])\n",
"\n",
"optimizer = torch.optim.AdamW(model.parameters(), lr=5e-4, weight_decay=0.1)\n",
"optimizer = torch.optim.AdamW(model.parameters(), lr=0.0004, weight_decay=0.1)\n",
"optimizer.load_state_dict(checkpoint[\"optimizer_state_dict\"])\n",
"model.to(device)\n",
"model.train();"
@ -497,9 +485,9 @@
"name": "stdout",
"output_type": "stream",
"text": [
"Ep 1 (Step 000000): Train loss 0.523, Val loss 6.445\n",
"Ep 1 (Step 000005): Train loss 0.422, Val loss 6.541\n",
"Every effort moves you?\" \"Yes--quite insensible to the irony. She wanted him vindicated--and by me!\" \"Oh, and I remember getting off a prodigious phrase about the sketch of the donkey. \"There were days when I\n"
"Ep 1 (Step 000000): Train loss 0.271, Val loss 6.545\n",
"Ep 1 (Step 000005): Train loss 0.244, Val loss 6.614\n",
"Every effort moves you?\" \"Yes--quite insensible to the irony. She wanted him vindicated--and by me!\" He laughed again, and threw back his head to look up at the sketch of the donkey. \"There were days when I\n"
]
}
],
@ -558,7 +546,7 @@
},
{
"cell_type": "code",
"execution_count": 1,
"execution_count": 12,
"id": "68d162d6-bbb9-4d6d-82ee-1c410694f872",
"metadata": {},
"outputs": [],
@ -586,7 +574,7 @@
},
{
"cell_type": "code",
"execution_count": 2,
"execution_count": 13,
"id": "d8373461-7dad-47da-a489-3e23f0799b23",
"metadata": {},
"outputs": [
@ -612,7 +600,7 @@
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": 14,
"id": "cdd44873-d6c2-4471-a20f-f639b09fdcd3",
"metadata": {},
"outputs": [],
@ -637,7 +625,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 15,
"id": "c7d562e4-33f6-4611-9b75-6ad1cb441d3b",
"metadata": {},
"outputs": [],
@ -652,7 +640,7 @@
},
{
"cell_type": "code",
"execution_count": 6,
"execution_count": 16,
"id": "46eda9ea-ccb0-46ee-931b-3c07502b2544",
"metadata": {},
"outputs": [],
@ -705,7 +693,7 @@
},
{
"cell_type": "code",
"execution_count": 7,
"execution_count": 17,
"id": "4e3574a2-687d-47a2-a2f6-457fe9d595f1",
"metadata": {},
"outputs": [
@ -713,8 +701,8 @@
"name": "stdout",
"output_type": "stream",
"text": [
"Training loss: 3.754748503367106\n",
"Validation loss: 3.559617757797241\n"
"Training loss: 3.7547483444213867\n",
"Validation loss: 3.5596189498901367\n"
]
}
],
@ -739,7 +727,7 @@
},
{
"cell_type": "code",
"execution_count": 13,
"execution_count": 18,
"id": "1a79a4b6-fe8f-40c2-a018-e731dcf391b3",
"metadata": {},
"outputs": [
@ -754,8 +742,8 @@
"File already exists and is up-to-date: gpt2/1558M/model.ckpt.index\n",
"File already exists and is up-to-date: gpt2/1558M/model.ckpt.meta\n",
"File already exists and is up-to-date: gpt2/1558M/vocab.bpe\n",
"Training loss: 3.3046312597062855\n",
"Validation loss: 3.119514226913452\n"
"Training loss: 3.3046313656700983\n",
"Validation loss: 3.1195149421691895\n"
]
}
],
@ -812,7 +800,7 @@
},
{
"cell_type": "code",
"execution_count": 1,
"execution_count": 19,
"id": "31e0972b-e85e-4904-a0f5-24c3eacd5fa2",
"metadata": {},
"outputs": [],
@ -838,7 +826,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 20,
"id": "b641ee88-f9d4-43ec-a787-e34199eed356",
"metadata": {},
"outputs": [
@ -882,7 +870,7 @@
},
{
"cell_type": "code",
"execution_count": 6,
"execution_count": 21,
"id": "c98f56f4-98fc-43b4-9ee5-726e9d17c73f",
"metadata": {},
"outputs": [],
@ -893,7 +881,7 @@
},
{
"cell_type": "code",
"execution_count": 7,
"execution_count": 22,
"id": "b1f7853c-6e81-4f1f-a1d0-61e2c7d33a20",
"metadata": {},
"outputs": [

1
ch05/01_main-chapter-code/previous_chapters.py
View File

@ -144,7 +144,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):

1
ch05/02_alternative_weight_loading/previous_chapters.py
View File

@ -144,7 +144,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):

1
ch05/03_bonus_pretraining_on_gutenberg/previous_chapters.py
View File

@ -146,7 +146,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):

1
ch05/05_bonus_hparam_tuning/previous_chapters.py
View File

@ -149,7 +149,6 @@ class FeedForward(nn.Module):
nn.Linear(cfg["emb_dim"], 4 * cfg["emb_dim"]),
GELU(),
nn.Linear(4 * cfg["emb_dim"], cfg["emb_dim"]),
nn.Dropout(cfg["drop_rate"])
)
def forward(self, x):