Upload 2 files

lctlm1.py

@@ -1,6 +1,8 @@
 import torch
 from torch import nn
 from typing import Optional
+import torch.nn.functional as F
+from tokenizers import Tokenizer
 class LCMBlock (nn.Module) :
   """
     LCm (Laten Connected Model ) block, looking attention as two preception and icreasing it
@@ -84,3 +86,35 @@ class LMLCT1(nn.Module):
         x = self.decoder_mlp(x)   
         logits = self.out(x)
         return logits
+
+
+def generate_tesk(model: LMLCT1, texts: str, tokenizer: Tokenizer, temperature: float = 1.0):
+    texts = "sos " + texts
+    input_ids = tokenizer.encode(texts).ids
+    start_index = len(input_ids)
+
+    input_ids = torch.tensor(input_ids, dtype=torch.long).unsqueeze(0) 
+    zeros_tolerant = 0 
+    response_tokens = []
+
+    for _ in range(start_index,500):   
+          
+        if zeros_tolerant >= 3 :
+           break  
+        with torch.no_grad():
+            logits = model(input_ids)        
+            logits = logits[:, -1, :]        
+
+            logits = logits / temperature
+
+            probs = F.softmax(logits, dim=-1)
+            next_token = torch.multinomial(probs, num_samples=1)
+        
+        if next_token == 0 :
+           zeros_tolerant +=1 
+        next_token_id = next_token.item()
+
+        response_tokens.append(next_token_id)
+        input_ids = torch.cat([input_ids, next_token], dim=1)
+
+    return tokenizer.decode(response_tokens).strip()

lctlm1_finetuned.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9af67ced6736aa23454b15d6025f2dd692ba43b8de267001419bee55386c71f3
+size 176885767