delete "Unbabel/TowerInstruct-7B-v0.2", "HuggingFaceTB/SmolLM3-3B"; add YanoljaNEXT-Rosetta-4B

app.py

@@ -17,26 +17,25 @@ iso1toall = {iso[1]: (iso[0], iso[2], iso[3]) for iso in non_empty_isos} # {'ro'
 langs = list(favourite_langs.keys())
 langs.extend(list(all_langs.keys())) # Language options as list, add favourite languages first
 
-models = ["Helsinki-NLP", "QUICKMT", "Argos", "Google", "HPLT", "HPLT-OPUS",
+models = ["Helsinki-NLP", "QUICKMT", "Argos", "HPLT", "HPLT-OPUS", "Google",
           "Helsinki-NLP/opus-mt-tc-bible-big-mul-mul", "Helsinki-NLP/opus-mt-tc-bible-big-mul-deu_eng_nld",
           "Helsinki-NLP/opus-mt-tc-bible-big-mul-deu_eng_fra_por_spa", "Helsinki-NLP/opus-mt-tc-bible-big-deu_eng_fra_por_spa-mul",
           "Helsinki-NLP/opus-mt-tc-bible-big-roa-deu_eng_fra_por_spa", "Helsinki-NLP/opus-mt-tc-bible-big-deu_eng_fra_por_spa-roa", "Helsinki-NLP/opus-mt-tc-bible-big-roa-en",
           "facebook/nllb-200-distilled-600M", "facebook/nllb-200-distilled-1.3B", "facebook/nllb-200-1.3B", "facebook/nllb-200-3.3B",
           "facebook/mbart-large-50-many-to-many-mmt", "facebook/mbart-large-50-one-to-many-mmt", "facebook/mbart-large-50-many-to-one-mmt",
-          "facebook/m2m100_418M", "facebook/m2m100_1.2B", "alirezamsh/small100",
           "facebook/hf-seamless-m4t-medium", "facebook/seamless-m4t-large", "facebook/seamless-m4t-v2-large",
+          "facebook/m2m100_418M", "facebook/m2m100_1.2B",
+          "alirezamsh/small100", "naist-nlp/mitre_466m", "naist-nlp/mitre_913m",
           "bigscience/mt0-small", "bigscience/mt0-base", "bigscience/mt0-large", "bigscience/mt0-xl",
           "bigscience/bloomz-560m", "bigscience/bloomz-1b1", "bigscience/bloomz-1b7", "bigscience/bloomz-3b",
           "google/madlad400-3b-mt", "jbochi/madlad400-3b-mt",
           "NiuTrans/LMT-60-0.6B", "NiuTrans/LMT-60-1.7B", "NiuTrans/LMT-60-4B",
-          "naist-nlp/mitre_466m", "naist-nlp/mitre_913m",
           "Lego-MT/Lego-MT", "BSC-LT/salamandraTA-2b-instruct",
           "winninghealth/WiNGPT-Babel", "winninghealth/WiNGPT-Babel-2", "winninghealth/WiNGPT-Babel-2.1",
-          "Unbabel/Tower-Plus-2B", "HuggingFaceTB/SmolLM3-3B", "Unbabel/TowerInstruct-7B-v0.2",
-          "utter-project/EuroLLM-1.7B", "utter-project/EuroLLM-1.7B-Instruct",
+          "Unbabel/Tower-Plus-2B", "utter-project/EuroLLM-1.7B", "utter-project/EuroLLM-1.7B-Instruct",
+          "yanolja/YanoljaNEXT-Rosetta-4B-2511", "yanolja/YanoljaNEXT-Rosetta-4B",
           "google-t5/t5-small", "google-t5/t5-base", "google-t5/t5-large",
-          "google/flan-t5-small", "google/flan-t5-base", "google/flan-t5-large", "google/flan-t5-xl"
-          ]
+          "google/flan-t5-small", "google/flan-t5-base", "google/flan-t5-large", "google/flan-t5-xl"]
 DEFAULTS = [langs[0], langs[1], models[0]]
 
 def timer(func):
@@ -114,6 +113,31 @@ class Translators:
         result = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
         return result
     
+    def rosetta(self):
+        model = AutoModelForCausalLM.from_pretrained(
+            self.model_name,
+            dtype=torch.bfloat16, # float32 slow
+            low_cpu_mem_usage=False, # True
+            device_map="auto")
+        tokenizer = AutoTokenizer.from_pretrained(self.model_name)
+        system = f"Translate the user's text to {self.tl}. Provide the final translation in a formal tone immediately immediately without any other text."
+        messages = [
+            {"role": "system", "content": system},
+            {"role": "user", "content": self.input_text},
+        ]     
+        prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+        inputs = tokenizer(prompt, return_tensors="pt").to(self.device)
+        input_length = inputs["input_ids"].shape[1]       
+        model.eval()
+        with torch.inference_mode():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=self.max_new_tokens,
+            )   
+        generated_tokens = outputs[0][input_length:]
+        translation = tokenizer.decode(generated_tokens, skip_special_tokens=True)
+        return translation    
+    
     def niutrans(self):
         tokenizer = AutoTokenizer.from_pretrained(self.model_name, padding_side='left')
         model = AutoModelForCausalLM.from_pretrained(self.model_name)
@@ -303,23 +327,6 @@ class Translators:
         translated_text = translator(text, max_length=512)
         return translated_text[0]['translation_text']
     
-    def smollm(self):
-        tokenizer = AutoTokenizer.from_pretrained(self.model_name)
-        model = AutoModelForCausalLM.from_pretrained(self.model_name)
-        prompt = f"""Translate the following {self.sl} text to {self.tl}, generating only the translated text and maintaining the original meaning and tone:
-        {self.input_text}
-        Translation:"""
-        inputs = tokenizer(prompt, return_tensors="pt")
-        outputs = model.generate(
-            inputs.input_ids,
-            max_length=len(inputs.input_ids[0]) + 150,
-            temperature=0.3,
-            do_sample=True
-        ) 
-        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        print(response)
-        return response.split("Translation:")[-1].strip()
-
     def flan(self):
         tokenizer = T5Tokenizer.from_pretrained(self.model_name, legacy=False)
         model = T5ForConditionalGeneration.from_pretrained(self.model_name)
@@ -605,6 +612,9 @@ def translate_text(input_text: str, s_language: str, t_language: str, model_name
         elif "small100" in model_name.lower():
             translated_text = Translators(model_name, sl, tl, input_text).smallonehundred()
 
+        elif "rosetta" in model_name.lower():
+            translated_text = Translators(model_name, s_language, t_language, input_text).rosetta()
+        
         elif "lego" in model_name.lower():
             translated_text = Translators(model_name, sl, tl, input_text).LegoMT()
 
@@ -657,9 +667,6 @@ def translate_text(input_text: str, s_language: str, t_language: str, model_name
         elif 'Unbabel' in model_name:   
             translated_text = Translators(model_name, s_language, t_language, input_text).unbabel()
             
-        elif model_name == "HuggingFaceTB/SmolLM3-3B":
-            translated_text = Translators(model_name, s_language, t_language, input_text).smollm()
-
         elif "winninghealth/WiNGPT" in model_name:      
             translated_text = Translators(model_name, s_language, t_language, input_text).wingpt()