Refactor model ID handling and update requirements.txt

app.py

@@ -1,6 +1,3 @@
-import os
-os.environ["CUDA_LAUNCH_BLOCKING"]="1"
-
 import spaces
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 import torch
@@ -13,9 +10,7 @@ models_available = [
     "MohamedRashad/Arabic-Orpo-Llama-3-8B-Instruct",
     "silma-ai/SILMA-9B-Instruct-v1.0",
     "inceptionai/jais-adapted-7b-chat",
-    # "inceptionai/jais-adapted-13b-chat",
     "inceptionai/jais-family-6p7b-chat",
-    # "inceptionai/jais-family-13b-chat",
     "NousResearch/Meta-Llama-3.1-8B-Instruct",
     # "unsloth/gemma-2-9b-it",
     "NousResearch/Meta-Llama-3-8B-Instruct",
@@ -25,7 +20,6 @@ tokenizer_a, model_a = None, None
 tokenizer_b, model_b = None, None
 torch_dtype = torch.bfloat16
 attn_implementation = "flash_attention_2"
-# attn_implementation = None
 
 def load_model_a(model_id):
     global tokenizer_a, model_a
@@ -40,8 +34,9 @@ def load_model_a(model_id):
             attn_implementation=attn_implementation,
             trust_remote_code=True,
         ).eval()
-    except:
+    except Exception as e:
         print(f"Using default attention implementation in {model_id}")
+        print(f"Error: {e}")
         model_a = AutoModelForCausalLM.from_pretrained(
             model_id,
             torch_dtype=torch_dtype,
@@ -64,7 +59,8 @@ def load_model_b(model_id):
             attn_implementation=attn_implementation,
             trust_remote_code=True,
         ).eval()
-    except:
+    except Exception as e:
+        print(f"Error: {e}")
         print(f"Using default attention implementation in {model_id}")
         model_b = AutoModelForCausalLM.from_pretrained(
             model_id,