Upload trained model

hanuman_loader.py

@@ -0,0 +1,81 @@
+"""
+Custom Hanuman Model Loader for Hugging Face
+
+This script allows you to load and use the custom Hanuman model from Hugging Face.
+Make sure to include 'modeling.py' in your repository for this to work.
+
+Usage:
+    from hanuman_loader import HanumanModel
+    model = HanumanModel.from_pretrained("your-username/your-model-repo")
+    tokenizer = AutoTokenizer.from_pretrained("your-username/your-model-repo")
+
+    prompt = "สวัสดี"
+    inputs = tokenizer(prompt, return_tensors="pt")
+    outputs = model.generate(inputs["input_ids"], max_new_tokens=50)
+    print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+"""
+
+import torch
+from transformers import AutoTokenizer
+from modeling import Hanuman
+import json
+import os
+
+
+class HanumanModel:
+    """Custom loader for Hanuman model compatible with Hugging Face Hub."""
+
+    @classmethod
+    def from_pretrained(cls, repo_id, map_location="cpu"):
+        """
+        Load Hanuman model from Hugging Face repository.
+
+        Args:
+            repo_id (str): Hugging Face repository ID (e.g., "username/model-name")
+            map_location (str): Device to load model on
+
+        Returns:
+            HanumanModel: Loaded model instance
+        """
+        # Download tokenizer
+        tokenizer = AutoTokenizer.from_pretrained(repo_id)
+
+        # Download config
+        config_path = os.path.join(repo_id, "config.json")  # This will be downloaded by HF
+        with open(config_path, "r", encoding="utf-8") as f:
+            cfg = json.load(f)
+
+        # Instantiate model
+        model = Hanuman(
+            vocab_size=cfg["vocab_size"],
+            n_positions=cfg["n_positions"],
+            n_embd=cfg["n_embd"],
+            n_layer=cfg["n_layer"],
+            n_head=cfg["n_head"],
+            use_think_head=True  # Assuming your model uses this
+        )
+
+        # Load weights
+        model_path = os.path.join(repo_id, "pytorch_model.bin")
+        state = torch.load(model_path, map_location=map_location)
+        model.load_state_dict(state)
+
+        return cls(model, tokenizer)
+
+    def __init__(self, model, tokenizer):
+        self.model = model
+        self.tokenizer = tokenizer
+
+    def generate(self, input_ids, max_new_tokens=50, temperature=1.0, top_k=50, top_p=0.95):
+        """Generate text using the model."""
+        return self.model.generate(
+            input_ids,
+            max_new_tokens=max_new_tokens,
+            temperature=temperature,
+            top_k=top_k,
+            top_p=top_p
+        )
+
+    def __call__(self, input_ids, **kwargs):
+        """Forward pass through the model."""
+        return self.model(input_ids, **kwargs)