Update more dump.

chatglm/modeling_chatglm.py

@@ -558,9 +558,10 @@ class ChatGLMForConditionalGeneration(nn.Module):
 
         outputs = self.sample(
             input_ids,
-            pad_token_id=generation_config.pad_token_id,
+            generation_config.pad_token_id,
-            eos_token_id=generation_config.eos_token_id,
+            generation_config.eos_token_id,
-            output_hidden_states=generation_config.output_hidden_states,
+            generation_config.output_hidden_states,
+            tokenizer,
         )
 
         outputs = outputs.tolist()[0][len(inputs["input_ids"][0]) : -1]
@@ -574,6 +575,7 @@ class ChatGLMForConditionalGeneration(nn.Module):
         pad_token_id: Optional[int] = None,
         eos_token_id: Optional[Union[int, List[int]]] = None,
         output_hidden_states: Optional[bool] = None,
+        tokenizer=None,
     ):
         if isinstance(eos_token_id, int):
             eos_token_id = [eos_token_id]
@@ -582,7 +584,7 @@ class ChatGLMForConditionalGeneration(nn.Module):
         isFinished = torch.zeros(
             input_ids.shape[0], dtype=torch.long, device=input_ids.device
         )
-        token_count = 0
+        # token_count = 0
         while True:
             input_ids_in = input_ids
             batch_size, seq_length = input_ids_in.shape
@@ -599,11 +601,13 @@ class ChatGLMForConditionalGeneration(nn.Module):
             )
             next_token_logits = logits[:, -1, :]
             probs = nn.functional.softmax(next_token_logits, dim=-1)
+            next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
 
-            # show.DumpTensorToImage(next_token_logits[0], "generated/next_tokens"+str(token_count)+".png")
+            # response = tokenizer.decode(next_tokens)
+            # name = "generated/next_tokens" + str(token_count) + "_" + response + "_.png"
+            # show.DumpTensorToImage(next_token_logits[0], name)
             # token_count = token_count + 1
 
-            next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
             # finished sentences should add a padding token to next
             pad_token = pad_token_id * isFinished
             next_tokens = next_tokens * (1 - isFinished) + pad_token

demo.py

@@ -8,7 +8,7 @@ from tools import show
 
 from transformers import AutoConfig
 
-seed = 1234
+seed = 4321
 torch.manual_seed(seed)
 torch.cuda.manual_seed_all(seed)
 
@@ -45,7 +45,7 @@ glm = glm.eval()
 query = "你好"
 response, history = glm.chat(tokenizer, query, history=[])
 print(response)
-if response[1:] != " 你好👋！我是人工智能助手 ChatGLM3-6B，很高兴见到你，欢迎问我任何问题":
+if response[1:] != " 你好！有什么可以帮助您的吗":
     raise ()
 
 # query = "colin"

test_tokenizer.py

@@ -0,0 +1,37 @@
+import json
+import torch
+from tools import show
+
+from chatglm import ChatGLMTokenizer
+
+pretrained_model_name_or_path = "../ZhipuAI/chatglm3-6b"
+
+
+tokenizer_config_file = "./chatglm/tokenizer_config.json"
+if tokenizer_config_file is not None:
+    with open(tokenizer_config_file, encoding="utf-8") as tokenizer_config_handle:
+        init_kwargs = json.load(tokenizer_config_handle)
+    init_kwargs.pop("tokenizer_class", None)
+    init_kwargs.pop("tokenizer_file", None)
+    saved_init_inputs = init_kwargs.pop("init_inputs", ())
+    init_inputs = saved_init_inputs
+init_kwargs["vocab_file"] = "./chatglm/tokenizer.model"
+init_kwargs["added_tokens_file"] = None
+init_kwargs["special_tokens_map_file"] = None
+init_kwargs["tokenizer_file"] = None
+init_kwargs["name_or_path"] = pretrained_model_name_or_path
+tokenizer = ChatGLMTokenizer(*init_inputs, **init_kwargs)
+
+
+aa = tokenizer.build_chat_input("骉")
+ab = tokenizer.encode("骉")
+a = tokenizer.decode([236,173,140])
+ 
+
+
+token = []
+for i in range(64798):
+    token.append(str(i) + " : " + tokenizer.decode(i))
+show.DumpListToFile(token, "generated/token.log")
+
+

tools/show.py

@@ -41,8 +41,17 @@ def DumpTensorToLog(tensor, name="log"):
         f.writelines("%s" % d + os.linesep)
     f.close()
 
+
 def DumpTensorToFile(tensor, name="tensor.pt"):
     torch.save(tensor.cpu(), name)
 
+
 def LoadTensorToFile(name="tensor.pt"):
     return torch.load(name)
+
+
+def DumpListToFile(list, name="list"):
+    f = open(name, "w")
+    for d in list:
+        f.writelines("%s" % d + os.linesep)
+    f.close()