try to have continos generation

app.py

@@ -6,14 +6,17 @@ from huggingface_hub import login
 
 login(os.getenv('HF_LOGIN'))
 
-#pipe = pipeline("text-generation", model="utter-project/EuroLLM-9B-Instruct")
-pipe = pipeline("text-generation", model="utter-project/EuroLLM-1.7B-Instruct")
-textik = st.text_area('enter some input!')
-message = [
-    {"role": "user", "content": textik},
-]
+model_id = "utter-project/EuroLLM-1.7B-Instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_id, torch_dtype=torch.bfloat16)
+model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
+model.generation_config.pad_token_id = tokenizer.pad_token_id
 
-if textik:
-  out = pipe(message, max_new_tokens=1024)[0]['generated_text'][1]
-  st.write('## '+out['role'])
-  st.write(out['content'])
+inner = st.text_area('enter some input!')
+text = '<|im_start|><|im_end|>user\n'+inner+'<|im_end|>\n<|im_start|>assistant\n'
+
+inputs = tokenizer(text, return_tensors="pt")
+outputs = inputs['input_ids']
+if inner:
+    while (not torch.any(outputs[0][-token_step_size:] == 4)):
+        outputs = model.generate(input_ids=outputs, attention_mask=torch.ones_like(outputs),max_new_tokens=token_step_size)
+        st.write(tokenizer.decode(outputs[0][-token_step_size:], skip_special_tokens=True))#, end=' ', flush=True)