update configs

config.json

@@ -23,7 +23,7 @@
   "n_embd": 4096,
   "n_head": 32,
   "n_layer": 32,
-  "n_positions": 6144,
+  "n_positions": 8192,
   "no_bias": true,
   "onnx_safe": null,
   "padded_vocab_size": 151936,

generation_config.json

@@ -1,8 +1,9 @@
 {
   "chat_format": "chatml",
   "eos_token_id": 151643,
-  "max_new_tokens": 512,
   "pad_token_id": 151643,
+  "max_window_size": 6144,
+  "max_new_tokens": 512,
   "do_sample": true,
   "top_k": 0,
   "top_p": 0.5,

modeling_qwen.py

@@ -994,12 +994,15 @@ class QWenLMHeadModel(QWenPreTrainedModel):
         if stop_words_ids is None:
             stop_words_ids = []
 
+        max_window_size = kwargs.get('max_window_size', None)
+        if max_window_size is None:
+            max_window_size = self.generation_config.max_window_size
         raw_text, context_tokens = make_context(
             tokenizer,
             query,
             history=history,
             system=system,
-            max_window_size=6144,
+            max_window_size=max_window_size,
             chat_format=self.generation_config.chat_format,
         )
 
@@ -1045,12 +1048,15 @@ class QWenLMHeadModel(QWenPreTrainedModel):
         if stop_words_ids is None:
             stop_words_ids = []
 
+        max_window_size = kwargs.get('max_window_size', None)
+        if max_window_size is None:
+            max_window_size = self.generation_config.max_window_size
         raw_text, context_tokens = make_context(
             tokenizer,
             query,
             history=history,
             system=system,
-            max_window_size=6144,
+            max_window_size=max_window_size,
             chat_format=self.generation_config.chat_format,
         )