From 76d9b8509bf8112e75398d10777ebdaff8572f8c Mon Sep 17 00:00:00 2001
From: charlie-rasberry <charlie.rasberry@outlook.com>
Date: Fri, 20 Feb 2026 19:17:22 +0000
Subject: [PATCH] Model almost complete, need to work on loss functions soon

---
 src/__pycache__/dataset.cpython-313.pyc | Bin 0 -> 2184 bytes
 src/model.py                            |  31 ++++++++++++++++++++++--
 2 files changed, 29 insertions(+), 2 deletions(-)
 create mode 100644 src/__pycache__/dataset.cpython-313.pyc

diff --git a/src/__pycache__/dataset.cpython-313.pyc b/src/__pycache__/dataset.cpython-313.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..bf36b23c03d222b4d345d682f24bc94b31904039
GIT binary patch
literal 2184
zcmcIl&2JM&6rc63*N(CCMF_-1iR&1mZ9+uFN1&n(EvS(I(UOe?5|&2GdXsEfug%OD
zD9V9eDkP{xkRnqqBwRvH$uUR%gsgnPnub$Pxm0Nb_rBS+1C5ZXN@b+|&6_uG-q*f)
zvz<u95R9K5R_#|&gnk!`<_I^4?QejXM>^7_i--iXL_)9*U6d~<L`eyrd@+1MC29&C
zKxw2aNu-DK$_Y8pK_b(Vrf!PGWliFd^9D1h#ehtFU16ok(oM^??^(nQ!~+a97Tcc#
zF^}?yNIC+$5|Q<Y9!jDgB%<hY5|Oa5M^rrwc2q6O6BE{L+q%=Rve^RXO{JktOu{Si
zZBGEqqw9ddm}uH;GGI=loOD787&+wZNL))fjpz`%cGTI?KaG~9%UYO+$^{-JmQlba
zz0E^~Dc-`Gaq?KvxQiXjon|wd!WC*cQ(P$<Y=$Jj5@`jXLSqUB&zN84bJQX<e{RMg
zj&0>n9{u82o)R-(2ogC`{*gy9wq2WH%)9>){z#KA{xt0cxQ{kEyQ)J^hMpE`odXY)
zN0E9@`cZ2=nfaw{8}^ae2v^YWlQj$rNY>6?4t*?L%~WL&5qZ}_T!uJFT+k9}gNX_a
zDbfMRBUoeseBQU)XP@nhw}T+K+}8Q{cy)Ac^x^D!^dRX1rn&PkU?Gh8|1=BSc?8Y@
zWB2%&!#PP5y6&SJf^*Hf|KM;=@YN3(4N?#3vaaai(_vT6$+{|zP6$tS4L#CK3rI!(
zUFy1jD}|B?k+@w0D)PTZ*CK2-efZzC$3Cg{J)cy2{6p<}Vnod#t@ScjiQhu}n>Yz~
znQ4ko^GMk!6l`~z#~7)&rorrz%Ug-XD#XRia%qWB-eSAu3d44R5>cbP)nKqCL~+re
zH+k$vWf~K!Tq2COO<4vUwqSj$Vo}CbgO)9m@%CWGRGcgVXe#LkZ6qb&GXZG<0|GJt
zTExF*T(O;!`M#0x2nEH~J&T7~iI_874cvr?#Qoy3UARk<f{I$MSrWnR^q%0B?!kE4
zVm7mia8rAB-_)J0PS8RB0Jx9p9s8>L=k`~7=X#%-UPpT2l-Du*K;4R?t^?KVT(;ic
zT|e0WeC*e;`hg5k_5OjaJ*X?ah19Or2hpv=D4AXu@_I&I_nh>4POjdn^_;Hv9bO+C
z_4-b|?mO%Coqg3&>pQoR?0-J|Y<Ph!-C4Y|G`l#vdd?d>wbttmep^e9Z}exL&pw-d
zd46Sld3@#4@};%C-tbrJ+3&pJ?`!>6Ho6a1ohQ!1@uktl(WNtsXI5oz;P~n_Zy;Oi
z9@|JAT8h7jznomTx_osdznoti@Q!5H$F6usCTc1D&qSm>wuK@sF~1jSvY&guM*P&n
zJGY6T)4_8Do!K>t7UV95aj{gWID(F2e5+zOjh!}(r))x*W4o4H!kCC;lKldP0Qf$P
zDT87&F_aRqZ&a8?F<wT#asMw7(Vu(}4Ty^N`ZIbPlEa7oib>(#cWNpexa!MjQ8Fm>
zs%6717?g;H!`uD(7^yJZp(EmHA)=%5I3tEjr%I#<ZxB&{71%l^DA9<6Dv6#6V#|_I
n;E5(rMY(Rtk|b?Mk(Bri?R|qXZ_weliYCP#kH3Q#A#mw$>HoRe

literal 0
HcmV?d00001

diff --git a/src/model.py b/src/model.py
index 1969b3d..0996802 100644
--- a/src/model.py
+++ b/src/model.py
@@ -24,13 +24,40 @@ class Model(nn.Module):
         self.feature_head = nn.Linear(hidden_size, 2)
         self.aspect_head = nn.Linear(hidden_size, 6)
         self.aspect_sentiment_head = nn.Linear(hidden_size, 3)
+
+    # Pass through encoder then extract the token representation
+    # Apply droupout to it, take scores for each head, return them in a dictionary
     def forward(self, input_ids, attention_mask):
         outputs = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
+        output = outputs.last_hidden_state[:, 0, :]
 
+        output = self.dropout(output)
 
+        # Logits for each head:
+        bug_logits = self.bug_head(output)
+        feature_logits = self.feature_head(output)
+        aspect_logits = self.aspect_head(output)
+        aspect_sentiment = self.aspect_sentiment_head(output)
+        return {
+            'bug_report': bug_logits,
+            'feature_request': feature_logits,
+            'aspect': aspect_logits,
+            'aspect_sentiment': aspect_sentiment
+        }
+    
+if __name__ == "__main__":
+    from dataset import ReviewDataset
+    from transformers import AutoTokenizer
+    from torch.utils.data import DataLoader
 
+    tokenizer = AutoTokenizer.from_pretrained("FacebookAI/xlm-roberta-base")
+    dataset = ReviewDataset("data/processed/original_train.csv", tokenizer)
+    loader = DataLoader(dataset, batch_size=2)
 
+    batch = next(iter(loader))
 
+    model = Model()
+    outputs = model(batch["input_ids"], batch["attention_mask"])
 
-tokenizer = AutoTokenizer.from_pretrained("FacebookAI/xlm-roberta-base")
-model = AutoModelForMaskedLM.from_pretrained("FacebookAI/xlm-roberta-base")
\ No newline at end of file
+    for k, v in outputs.items():
+        print(k, v.shape)
\ No newline at end of file