Update app.py

app.py

@@ -31,37 +31,54 @@ model_kwargs={
             "max_new_tokens":2048,
             "temperature": 0.01,
             "truncate": 4096,
-            "seed" : 256,
+            "seed" : 42,
             "stop" : ["</s>","<|endoftext|>","<|end|>"],
             }
 
 bloom_model_kwargs={
             "max_new_tokens":1000,
             "temperature": 0.01,
-            "truncate": 1000,
-            "seed" : 256,
+            "truncate": 1512,
+            "seed" : 42,
             "stop" : ["</s>","<|endoftext|>","<|end|>","]]]"],
             }
 
 llm1 = HuggingFaceHub(repo_id=llama_repo, task="text-generation", model_kwargs=model_kwargs)
 llm2 = HuggingFaceHub(repo_id=starchat_repo, task="text-generation", model_kwargs=model_kwargs)
 llm3 = HuggingFaceHub(repo_id=bloom_repo, task="text-generation", model_kwargs=bloom_model_kwargs)
+
+def split_text_into_chunks(text, chunk_size=500):
+    lines = text.splitlines()
+    chunks = []
+    temp_chunk = ""
+    for line in lines:
+        # If adding the current line doesn't exceed the chunk size, add the line to the chunk
+        if len(temp_chunk) + len(line) <= chunk_size:
+            temp_chunk += line + '\n'
+        else:
+            # If adding the line exceeds chunk size, store the current chunk and start a new one
+            chunks.append(temp_chunk)
+            temp_chunk = line + '\n'
+    # Don't forget the last chunk
+    chunks.append(temp_chunk)
+    return chunks
     
 def translation(source, target, text):
     response = text
-    try:
-        input_prompt = bloom_template.replace("{source}", source)
-        input_prompt = input_prompt.replace("{target}", target)
-        input_prompt = input_prompt.replace("{query}", text)
-        response=llm3(input_prompt)
-    except Exception as e:
-        print(f"ERROR: LLM show {e}")
-        input_prompt = starchat_template.replace("{source}", source)
-        input_prompt = input_prompt.replace("{target}", target)
-        input_prompt = input_prompt.replace("{query}", text)
-        response=llm1(input_prompt)
-    for eot in bloom_model_kwargs['stop']:
-        response = response.replace(eot,"")
+    chunks = split_text_into_chunks(text)
+    for chunk in chunks:
+        response = ""
+        try:
+            input_prompt = bloom_template.replace("{source}", source)
+            input_prompt = input_prompt.replace("{target}", target)
+            input_prompt = input_prompt.replace("{query}", chunk)
+            stchunk = llm3(input_prompt)
+            for eot in bloom_model_kwargs['stop']:
+                stchunk = stchunk.replace(eot,"\n")
+            response += stchunk
+        except Exception as e:
+            print(f"ERROR: LLM show {e}")
+        response = response.strip()
     return response
 
 gr.Interface(translation, inputs=["text","text","text"], outputs="text").launch()