Upload telegram_bot.py with huggingface_hub

telegram_bot.py

@@ -111,30 +111,30 @@ def call_kora_chat(prompt: str, model: str = "kora-chat-0.5b") -> str:
     # 1. Try direct in-memory generation if running within gateway process
     try:
         import sys
-        if "gateway" in sys.modules:
-            gw = sys.modules["gateway"]
-            if hasattr(gw, "CODER_MODEL") and hasattr(gw, "CODER_TOKENIZER"):
-                tok = gw.CODER_TOKENIZER
-                mdl = gw.CODER_MODEL
-                system_prompt = "You are Kora, an intelligent, helpful, and friendly AI assistant built by Kontyra." if "coder" not in model else "You are Kora Coder, an expert AI programming assistant built by Kontyra."
-                messages = [
-                    {"role": "system", "content": system_prompt},
-                    {"role": "user", "content": prompt}
-                ]
-                prompt_str = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-                device = next(mdl.parameters()).device
-                inputs = tok([prompt_str], return_tensors="pt").to(device)
-                import torch
-                with torch.no_grad():
-                    out = mdl.generate(
-                        **inputs,
-                        max_new_tokens=256,
-                        temperature=0.7,
-                        pad_token_id=tok.eos_token_id
-                    )
-                reply = tok.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True).strip()
-                if reply:
-                    return reply
+        gw = sys.modules.get("gateway") or sys.modules.get("__main__")
+        if gw and hasattr(gw, "CODER_MODEL") and hasattr(gw, "CODER_TOKENIZER"):
+            tok = gw.CODER_TOKENIZER
+            mdl = gw.CODER_MODEL
+            is_coder = "coder" in model or "code" in prompt.lower()
+            system_prompt = "You are Kora Coder, an expert AI programming assistant built by Kontyra." if is_coder else "You are Kora, an intelligent, helpful, and friendly AI assistant built by Kontyra."
+            messages = [
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": prompt}
+            ]
+            prompt_str = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+            device = next(mdl.parameters()).device
+            inputs = tok([prompt_str], return_tensors="pt").to(device)
+            import torch
+            with torch.no_grad():
+                out = mdl.generate(
+                    **inputs,
+                    max_new_tokens=300,
+                    temperature=0.3 if is_coder else 0.7,
+                    pad_token_id=tok.eos_token_id
+                )
+            reply = tok.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True).strip()
+            if reply:
+                return reply
     except Exception as err:
         print(f"ℹ️ In-memory inference notice: {err}", flush=True)
 
@@ -202,7 +202,7 @@ def handle_update(update: dict, token: str = None):
         img_url = call_kora_image(prompt)
         send_telegram_photo(chat_id, img_url, f"✨ {prompt}")
     elif text.startswith("/code"):
-        prompt = text.replace("/code", "").strip()
+        prompt = text.replace("/code", "").strip() or "Write a Python function to demonstrate binary search"
         send_telegram_message(chat_id, "💻 Generating code with Kora Coder...")
         reply = call_kora_chat(prompt, model="kora-coder-0.5b")
         send_telegram_message(chat_id, f"```\n{reply}\n```")