Upload predict.py with huggingface_hub

predict.py

@@ -185,6 +185,8 @@ def get_conversation_template(model_path: str) -> Conversation:
     """Get the default conversation template."""
     if "aquila-v1" in model_path:
         return get_conv_template("aquila-v1")
+    elif "aquila-v2" in model_path:
+        return get_conv_template("aquila-v2")
     elif "aquila-chat" in model_path:
         return get_conv_template("aquila-chat")
     elif "aquila-legacy" in model_path:
@@ -252,6 +254,21 @@ register_conv_template(
     )
 )
 
+register_conv_template(
+    Conversation(
+        name="aquila-v2",
+        system_message="A chat between a curious human and an artificial intelligence assistant. "
+        "The assistant gives helpful, detailed, and polite answers to the human's questions.\n\n",
+        roles=("<|startofpiece|>", "<|endofpiece|>", ""),
+        messages=(),
+        offset=0,
+        sep_style=SeparatorStyle.NO_COLON_TWO,
+        sep="",
+        sep2="</s>",
+        stop_str=["</s>", "<|endoftext|>", "<|startofpiece|>", "<|endofpiece|>"],
+    )
+)
+
 
 if __name__ == "__main__":
     print("aquila template:")
@@ -294,6 +311,17 @@ if __name__ == "__main__":
 
     print("\n")
 
+    print("aquila-v2 template:")
+    conv = get_conv_template("aquila-v2")
+    conv.append_message(conv.roles[0], "Hello!")
+    conv.append_message(conv.roles[1], "Hi!")
+    conv.append_message(conv.roles[0], "How are you?")
+    conv.append_message(conv.roles[1], None)
+    print(conv.get_prompt())
+
+    print("\n")
+
+
 def set_random_seed(seed):
     """Set random seed for reproducability."""
     if seed is not None and seed > 0:
@@ -330,9 +358,9 @@ def covert_prompt_to_input_ids_with_history(text, history, tokenizer, max_token,
     return example
 
 def predict(model, text, tokenizer=None,
-            max_gen_len=200, top_p=0.95,
-            seed=1234, topk=100,
-            temperature=0.9, 
+            max_gen_len=200, top_p=0.9,
+            seed=123, topk=15,
+            temperature=1.0, 
             sft=True, convo_template = "",
             device = "cuda",
             model_name="AquilaChat2-7B",
@@ -346,8 +374,8 @@ def predict(model, text, tokenizer=None,
     
     template_map = {"AquilaChat2-7B": "aquila-v1",
                     "AquilaChat2-34B": "aquila-legacy",
-                    "AquilaChat2-7B-16K": "aquila",
                     "AquilaChat2-70B-Expr": "aquila-v2",
+                    "AquilaChat2-7B-16K": "aquila",
                     "AquilaChat2-34B-16K": "aquila"}
     if not convo_template:
         convo_template=template_map.get(model_name, "aquila-chat")
@@ -357,7 +385,7 @@ def predict(model, text, tokenizer=None,
         topk = 1
         temperature = 1.0
     if sft:
-        tokens = covert_prompt_to_input_ids_with_history(text, history=history, tokenizer=tokenizer, max_token=2048, convo_template=convo_template)
+        tokens = covert_prompt_to_input_ids_with_history(text, history=history, tokenizer=tokenizer, max_token=20480, convo_template=convo_template)
         tokens = torch.tensor(tokens)[None,].to(device)
     else :
         tokens = tokenizer.encode_plus(text)["input_ids"]