Create hf_model_adapter.py

hf_model_adapter.py

@@ -0,0 +1,36 @@
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+
+class HFLocalModelAdapter:
+    """
+    Minimal Hugging Face model adapter for text generation.
+    """
+
+    def __init__(self, model_name="stabilityai/stablelm-3b-4e1t", device=None):
+        self.model_name = model_name
+        self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
+        print(f"Loading {model_name} on {self.device} ...")
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name,
+            torch_dtype=torch.float16 if "cuda" in self.device else torch.float32,
+            low_cpu_mem_usage=True,
+            device_map="auto" if "cuda" in self.device else None,
+        )
+        self.model.to(self.device)
+        print("Model loaded.")
+
+    def generate(self, prompt, max_new_tokens=250, temperature=0.7, top_p=0.95):
+        inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
+        out = self.model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            do_sample=True,
+            temperature=temperature,
+            top_p=top_p,
+            pad_token_id=self.tokenizer.eos_token_id,
+        )
+        decoded = self.tokenizer.decode(out[0], skip_special_tokens=True)
+        if decoded.startswith(prompt):
+            return decoded[len(prompt):].strip()
+        return decoded.strip()