Update app.py

app.py

@@ -1,27 +1,31 @@
 import os
 import gradio as gr
 from langchain.llms import HuggingFaceHub
-from langchain.prompts import PromptTemplate
-from langchain.chains import LLMChain
 
 model_repo = os.getenv('HF_MODEL_REPO')
-eos_string = "</s>"
-template = """<s>[INST]<<SYS>>You work as nation language translator. You job is translate user request from {source} to {target}<<SYS>>[/INST]</s>
-<s>[INST]{query}[/INST]</s>\n"""
+template = """[INST]<<SYS>>I want you to be a translator. You do translate {source} texts in the context into {target} then you return to me the whole translated context AND DO NOTHING ELSE.<</SYS>>
+Begin of the context:
+{query}
+End of the context.[/INST]
+{target} translation of the context:
 
-prompt = PromptTemplate(template=template, input_variables=["source","target","query"])
+"""
 
 model_kwargs={
-            "max_new_tokens":2048, 
-            "temperature":0.5,
-            "stop" : ["</s>","<|endoftext|>","<|end|>"]
+            "max_new_tokens":2048,
+            "temperature": 0.01,
+            "truncate": 4096,
+            "stop" : ["</s>","<|endoftext|>","<|end|>"],
             }
 
 llm = HuggingFaceHub(repo_id=model_repo, task="text-generation", model_kwargs=model_kwargs)
-chain = LLMChain(prompt=prompt, llm=llm)
 
 def translation(source, target, text):
-    response=chain.run(source=source,target=target,query=text)
-    return response.partition(eos_string)[0]
+    input_prompt = template.replace("{source}", source)
+    input_prompt = input_prompt.replace("{target}", target)
+    input_prompt = input_prompt.replace("{query}", text)
+    print(input_prompt)
+    response=llm(input_prompt)
+    return response
 
 gr.Interface(translation, inputs=["text","text","text"], outputs="text").launch()