Create evo_plugin_example.py

evo_plugin_example.py

@@ -0,0 +1,48 @@
+"""
+evo_plugin_example.py
+Step 8: Example text generator plugin (for immediate testing).
+(Objective)
+- Provides the same interface your real Evo plugin will expose.
+- Uses a tiny HF model (distilgpt2) so you can test generative mode now.
+- Replace this later with `evo_plugin.py` wrapping your EvoDecoder/Evo QA.
+
+Real Evo instructions:
+- Create a file `evo_plugin.py` with a `load_model()` that returns an object
+  exposing `generate(prompt: str, max_new_tokens: int, temperature: float) -> str`.
+- The app will auto-prefer `evo_plugin.py` if it exists.
+"""
+
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+
+class _HFGenerator:
+    def __init__(self, model_name: str = "distilgpt2"):
+        self.device = torch.device("cpu")
+        self.tok = AutoTokenizer.from_pretrained(model_name)
+        self.model = AutoModelForCausalLM.from_pretrained(model_name).to(self.device)
+        # GPT-2 models have no pad token; set to eos to avoid warnings
+        if self.tok.pad_token_id is None and self.tok.eos_token_id is not None:
+            self.tok.pad_token_id = self.tok.eos_token_id
+
+    @torch.no_grad()
+    def generate(self, prompt: str, max_new_tokens: int = 200, temperature: float = 0.4) -> str:
+        inputs = self.tok(prompt, return_tensors="pt").to(self.device)
+        out = self.model.generate(
+            **inputs,
+            max_new_tokens=int(max_new_tokens),
+            do_sample=temperature > 0.0,
+            temperature=float(max(0.01, temperature)),
+            top_p=0.95,
+            pad_token_id=self.tok.pad_token_id,
+        )
+        text = self.tok.decode(out[0], skip_special_tokens=True)
+        # Return only the completion after the prompt to reduce echo
+        return text[len(prompt):].strip()
+
+
+def load_model():
+    """
+    (Objective) Entry-point used by evo_inference.py.
+    """
+    return _HFGenerator()