data16w with emotion

README.md

@@ -9,4 +9,30 @@ metrics:
 - rouge
 ---
 ## 模型
-- 基于中文MengziT5的新闻评论生成模型
+- 基于中文[MengziT5](https://huggingface.co/Langboat/mengzi-t5-base)的新闻评论生成模型
+- 数据集来源于论文[《Coherent Comment Generation for Chinese Articles with a Graph-to-Sequence Model》](https://github.com/lancopku/Graph-to-seq-comment-generation)
+
+## 生成评论
+- 在线API只能生成一种评论，模型通过设置model.generate()参数是可以生成多种评论的
+
+```Python
+
+t5_tokenizer = T5Tokenizer.from_pretrained("Langboat/mengzi-t5-base")
+
+model = T5ForConditionalGeneration.from_pretrained("wawaup/MengziT5-Comment")
+
+def generate_comment(input_ids,cnt_num):
+    outputs = model.generate(input_ids,
+                            max_length=128,
+                            do_sample=True,
+                            temperature=0.9,
+                            early_stopping=True,
+                            repetition_penalty=10.0,
+                            top_p=0.5,
+                            num_return_sequences=cnt_num)
+    print(outputs) 
+    preds_cleaned = [t5_tokenizer.decode(ids, skip_special_tokens=True, 
+                            clean_up_tokenization_spaces=True) for ids in outputs]
+    print(preds_cleaned)
+    return preds_cleaned
+```

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "./ckpts/data7w/checkpoint-15000",
+  "_name_or_path": "./ckpts/data8w/checkpoint-30000",
   "architectures": [
     "T5ForConditionalGeneration"
   ],

optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:debd08a8934f406734874ddae41a070fa79392a790d9b0b1bfdfb12483ba1c8f
+oid sha256:64050a7d3cca2e99eb1cebb9e8dd3b08d7e5635b8057a755ad031132ad1e754b
 size 1980801667

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea2a63eb3c349409aa5feca6c295484505753788b48d50f06c1d540fa5ba0a6e
+oid sha256:a0175df517a4511cb9ca17ad07b5aec9d7bab45832e8073e6eb2c9a51154788b
 size 990413657

rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec4fa2cce7927ff90e848326a7c5a6caacc85931725c736aebf60a16a8afc209
+oid sha256:45112dceaf00de83a4ec5456ff955cba53c32bc4bde209d0a92309921b5d5306
 size 14657

scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:124e9dd4835893ac44031b91135b83c01999e98dfe14413d60538837437f6765
+oid sha256:a9e1f7dd49eec2b5c97573622413e8996b9b6e76c7b4dbf86a59c9458d7816ad
 size 623

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5898971c771f6b937939dbb6fdffe250035e6409ba605078700d4c3b33524015
-size 3183
+oid sha256:82eb62b59f4f30a3619a2d4a9b3e311bf2204eedc85619b0b5d9d7f1ba4e2f38
+size 3119