Add a generate() function for RWKV

modules/RWKV.py

@@ -31,5 +31,14 @@ class RWKVModel:
         result.model = pipeline
         return result
 
-    def generate(self, context, **kwargs):
-        return self.model.generate(context, **kwargs)
+    def generate(self, context, token_count=20, temperature=1, top_p=1, alpha_frequency=0.25, alpha_presence=0.25, token_ban=[0], token_stop=[], callback=None):
+        args = PIPELINE_ARGS(
+            temperature = temperature,
+            top_p = top_p,
+            alpha_frequency = 0.25, # Frequency Penalty (as in GPT-3)
+            alpha_presence = 0.25, # Presence Penalty (as in GPT-3)
+            token_ban = [0], # ban the generation of some tokens
+            token_stop = []
+        )
+
+        return self.model.generate(context, token_count=token_count, args=args, callback=callback)

modules/text_generation.py

@@ -85,19 +85,13 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
         torch.cuda.empty_cache()
 
     if shared.is_RWKV:
-        args = PIPELINE_ARGS(temperature = temperature, top_p = top_p,
-                             alpha_frequency = 0.25, # Frequency Penalty (as in GPT-3)
-                             alpha_presence = 0.25, # Presence Penalty (as in GPT-3)
-                             token_ban = [0], # ban the generation of some tokens
-                             token_stop = []) # stop generation whenever you see any token here
-
         if shared.args.no_stream:
-            reply = question + shared.model.generate(question, token_count=max_new_tokens, args=args, callback=None)
+            reply = question + shared.model.generate(question, token_count=max_new_tokens, temperature=temperature)
             yield formatted_outputs(reply, None)
             return formatted_outputs(reply, None)
         else:
             for i in range(max_new_tokens//8):
-                reply = question + shared.model.generate(question, token_count=8, args=args, callback=None)
+                reply = question + shared.model.generate(question, token_count=8, temperature=temperature)
                 yield formatted_outputs(reply, None)
                 question = reply
             return formatted_outputs(reply, None)