Update app.py

app.py

@@ -1,9 +1,84 @@
 import torch
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, PreTrainedModel, PretrainedConfig
 from huggingface_hub import hf_hub_download
 import json
 
+# Define the model architecture
+class SmolLM2Config(PretrainedConfig):
+    model_type = "smollm2"
+    
+    def __init__(
+        self,
+        vocab_size=49152,
+        hidden_size=576,
+        intermediate_size=1536,
+        num_hidden_layers=30,
+        num_attention_heads=9,
+        num_key_value_heads=3,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=0,
+        tie_word_embeddings=True,
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.max_position_embeddings = max_position_embeddings
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs
+        )
+
+# Register the model architecture
+from transformers import AutoConfig
+AutoConfig.register("smollm2", SmolLM2Config)
+
+class SmolLM2ForCausalLM(PreTrainedModel):
+    config_class = SmolLM2Config
+    
+    def __init__(self, config):
+        super().__init__(config)
+        self.config = config
+        
+        # Load the model weights directly from the checkpoint
+        self.model = AutoModelForCausalLM.from_pretrained(
+            "meta-llama/Llama-2-7b-hf",
+            config=config,
+            torch_dtype=torch.float16,
+            low_cpu_mem_usage=True
+        )
+        
+    def forward(self, input_ids=None, attention_mask=None, labels=None, **kwargs):
+        return self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            labels=labels,
+            **kwargs
+        )
+    
+    def prepare_inputs_for_generation(self, input_ids, **kwargs):
+        return self.model.prepare_inputs_for_generation(input_ids, **kwargs)
+
+# Register the model
+AutoModelForCausalLM.register(SmolLM2Config, SmolLM2ForCausalLM)
+
 # Cache for model and tokenizer
 MODEL = None
 TOKENIZER = None
@@ -16,8 +91,12 @@ def initialize():
         model_id = "jatingocodeo/SmolLM2"
         
         try:
-            # Download model files from HF Hub
+            # Download and load config
+            print("Loading config...")
             config_path = hf_hub_download(repo_id=model_id, filename="config.json")
+            with open(config_path, 'r') as f:
+                config_dict = json.load(f)
+            config = SmolLM2Config(**config_dict)
             
             # Load tokenizer
             print("Loading tokenizer...")
@@ -33,8 +112,9 @@ def initialize():
             
             # Load model
             print("Loading model...")
-            MODEL = AutoModelForCausalLM.from_pretrained(
+            MODEL = SmolLM2ForCausalLM.from_pretrained(
                 model_id,
+                config=config,
                 torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
                 trust_remote_code=True,
                 low_cpu_mem_usage=True