[MOD] Change pipeline charging between Carballo models

app.py

@@ -56,30 +56,24 @@ fronted_theme = 'Soft'
 
 # Model charge ---------------------------------------------------------
 model_id_bloom  = "proxectonos/Carballo-bloom-1.3B"
-generator_model_bloom = pipeline("text-generation", model=model_id_bloom)
 model_id_carvalho = "Nos-PT/Carvalho_pt-gl-1.3B"
-generator_model_carvalho = pipeline("text-generation", model=model_id_carvalho)
-
-#Quantized Carballo-Cerebras (model less used) to save memory 
 model_id_cerebras  = "proxectonos/Carballo-cerebras-1.3B"
-quantization_config = BitsAndBytesConfig(
-        load_in_4bit=True,
-        bnb_4bit_quant_type="nf4",
-        bnb_4bit_compute_dtype=torch.float16,
-)
 
-model_quantizied = AutoModelForCausalLM.from_pretrained(model_id_cerebras, quantization_config=quantization_config)
-tokenizer = AutoTokenizer.from_pretrained(model_id_cerebras)
-generator_model_cerebras = pipeline("text-generation", model=model_id_cerebras)
+generator_model_gl = pipeline("text-generation", model=model_id_bloom)
+generator_model_pt = pipeline("text-generation", model=model_id_carvalho)
 
 # Generation functions ---------------------------------------------------------
 def get_model(model_selection):
     if model_selection == "Carballo-bloom-1.3B":
-        return generator_model_bloom
+        if generator_model_gl.model.name_or_path != model_id_bloom:
+            generator_model_gl = pipeline("text-generation", model=model_id_bloom)
+        return generator_model_gl
     elif model_selection == "Carballo-cerebras-1.3B":
-        return generator_model_cerebras
+        if generator_model_gl.model.name_or_path != model_id_cerebras:
+            generator_model_gl = pipeline("text-generation", model=model_id_cerebras)
+        return generator_model_gl
     else:
-        return generator_model_carvalho
+        return generator_model_pt
 
 def remove_empty_lines(text):
     lines = text.strip().split("\n")

requirements.txt

@@ -1,4 +1,3 @@
 transformers
 torch
-accelerate
-bitsandbytes
+accelerate