Upload app.py with huggingface_hub

app.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-OpenLLM Real Models App - Corrected version with exact model architecture
+OpenLLM Real Models App - Final fixed version with exact bias configuration
 """
 
 import gradio as gr
@@ -49,8 +49,8 @@ class GPT(nn.Module):
             ln_f = nn.LayerNorm(config.n_embd),
         ))
         
-        # Language model head
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=config.bias)
+        # Language model head - MUST have bias to match saved model
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=True)
         
         # Initialize weights
         self.apply(self._init_weights)
@@ -133,7 +133,7 @@ class Block(nn.Module):
         return x
 
 class CausalSelfAttention(nn.Module):
-    """Multi-head self-attention with causal masking"""
+    """Multi-head self-attention with causal masking - EXACT bias configuration"""
     def __init__(self, config):
         super().__init__()
         assert config.n_embd % config.n_head == 0
@@ -144,7 +144,13 @@ class CausalSelfAttention(nn.Module):
         self.n_head = config.n_head
         self.n_embd = config.n_embd
         self.dropout = config.dropout
-        self.bias = config.bias  # Add bias parameter
+        self.bias = config.bias
+        
+        # ADD THE BIAS PARAMETER that the saved model expects
+        if config.bias:
+            self.bias = nn.Parameter(torch.zeros(config.n_embd))
+        else:
+            self.register_parameter('bias', None)
     
     def forward(self, x):
         B, T, C = x.size()
@@ -161,6 +167,11 @@ class CausalSelfAttention(nn.Module):
         
         # Output projection
         y = self.resid_dropout(self.c_proj(y))
+        
+        # Add the bias if it exists
+        if self.bias is not None:
+            y = y + self.bias
+            
         return y
 
 class MLP(nn.Module):