demo

app.py

@@ -0,0 +1,115 @@
+# 使用gradio开发QA的可视化demo
+
+import gradio as gr
+from transformers import pipeline, AutoTokenizer, AutoModelForQuestionAnswering, BigBirdForQuestionAnswering, BigBirdConfig, PreTrainedModel, BigBirdTokenizer
+import torch
+from torch import nn
+from transformers.models.big_bird.modeling_big_bird import BigBirdOutput, BigBirdIntermediate
+
+class BigBirdNullHead(nn.Module):
+    """Head for question answering tasks."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.intermediate = BigBirdIntermediate(config)
+        self.output = BigBirdOutput(config)
+        self.qa_outputs = nn.Linear(config.hidden_size, 2)
+
+    def forward(self, encoder_output):
+        hidden_states = self.dropout(encoder_output)
+        hidden_states = self.intermediate(hidden_states)
+        hidden_states = self.output(hidden_states, encoder_output)
+        logits = self.qa_outputs(hidden_states)
+        return logits
+
+
+model_path = '/data1/chenzq/demo/checkpoint-epoch-best'
+
+class BigBirdForQuestionAnsweringWithNull(PreTrainedModel):
+    def __init__(self, config, model_id):
+        super().__init__(config)
+        self.bertqa = BigBirdForQuestionAnswering.from_pretrained(model_id,
+            config=self.config, add_pooling_layer=True)
+        self.null_classifier = BigBirdNullHead(self.bertqa.config)
+        self.contrastive_mlp = nn.Sequential(
+            nn.Linear(self.bertqa.config.hidden_size, self.bertqa.config.hidden_size),
+        )
+
+    def forward(self, **kwargs):
+        if self.training:
+            null_labels = kwargs['is_impossible']
+            del kwargs['is_impossible']
+            outputs = self.bertqa(**kwargs)
+            pooler_output = outputs.pooler_output
+            null_logits = self.null_classifier(pooler_output)
+            loss_fct = nn.CrossEntropyLoss()
+            null_loss = loss_fct(null_logits, null_labels)
+
+
+            outputs.loss = outputs.loss + null_loss 
+
+            return outputs.to_tuple()
+        else:
+            outputs = self.bertqa(**kwargs)
+            pooler_output = outputs.pooler_output
+            null_logits = self.null_classifier(pooler_output)
+
+            return (outputs.start_logits, outputs.end_logits, null_logits)
+model_id = 'vasudevgupta/bigbird-roberta-natural-questions'
+config = BigBirdConfig.from_pretrained(model_id)
+model = BigBirdForQuestionAnsweringWithNull(config, model_id)
+model.to('cuda')
+model.eval()
+
+model.load_state_dict(torch.load(model_path+'/pytorch_model.bin', map_location='cuda')) # map_location是指定加载到哪个设备
+
+tokenizer = BigBirdTokenizer.from_pretrained(model_path)
+
+def main(question, context):
+    # 编码输入
+    text = question + " [SEP] " + context
+    inputs = tokenizer(text, max_length=4096, truncation=True, return_tensors="pt")
+    inputs.to('cuda')
+    # 预测答案
+    outputs = model(**inputs)
+    start_scores = outputs[0]
+    end_scores = outputs[1]
+    null_scores = outputs[2]
+    # 解码答案
+    is_impossible = null_scores.argmax().item()
+    if is_impossible:
+        return "No Answer"
+    else:
+        answer_start = torch.argmax(start_scores)
+        answer_end = torch.argmax(end_scores) + 1
+        answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(inputs['input_ids'][0][answer_start:answer_end]))
+        return answer
+
+  
+
+with gr.Blocks() as demo:
+
+  gr.Markdown("""# Question Answerer!""")
+  with gr.Row():
+    with gr.Column():
+    #   options = gr.inputs.Radio(["vasudevgupta/bigbird-roberta-natural-questions", "vasudevgupta/bigbird-roberta-natural-questions"], label="Model")
+      text1 = gr.Textbox(
+            label="Question",
+            lines=1,
+            value="Who does Cristiano Ronaldo play for?",
+        )
+      text2 = gr.Textbox(
+            label="Context",
+            lines=3,
+            value="Cristiano Ronaldo is a player for Manchester United",
+        )
+      output = gr.Textbox()
+      b1 = gr.Button("Ask Question!")
+      b1.click(main, inputs=[text1, text2], outputs=output)
+#   gr.Markdown("""#### powered by [Tassle](https://bit.ly/3LXMklV)""")
+
+
+if __name__ == "__main__":
+
+    demo.launch(share=True)

checkpoint-epoch-best/config.json

@@ -0,0 +1,33 @@
+{
+  "_name_or_path": "vasudevgupta/bigbird-roberta-natural-questions",
+  "architectures": [
+    "BigBirdForQuestionAnsweringWithNull"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "attention_type": "block_sparse",
+  "block_size": 64,
+  "bos_token_id": 1,
+  "classifier_dropout": null,
+  "eos_token_id": 2,
+  "hidden_act": "gelu_fast",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 4096,
+  "model_type": "big_bird",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "num_random_blocks": 3,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "rescale_embeddings": false,
+  "sep_token_id": 66,
+  "torch_dtype": "float32",
+  "transformers_version": "4.18.0",
+  "type_vocab_size": 2,
+  "use_bias": true,
+  "use_cache": true,
+  "vocab_size": 50358
+}

checkpoint-epoch-best/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f56923a86a3827a575c0ba614808b6a7b58cf76a631f0e8f5529cc73583973a6
+size 550147981

checkpoint-epoch-best/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "sep_token": {"content": "[SEP]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "cls_token": {"content": "[CLS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "mask_token": {"content": "[MASK]", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true}}

checkpoint-epoch-best/spiece.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdc81e1fc9d42e0c08b86d5b280d05d7c5e9747c4231c648f2b56b8e1d893c82
+size 845731

checkpoint-epoch-best/tokenizer_config.json

@@ -0,0 +1 @@
+{"bos_token": {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "unk_token": {"content": "<unk>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "pad_token": {"content": "<pad>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sep_token": {"content": "[SEP]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "mask_token": {"content": "[MASK]", "single_word": false, "lstrip": true, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "cls_token": {"content": "[CLS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sp_model_kwargs": {}, "model_max_length": 4096, "name_or_path": "vasudevgupta/bigbird-roberta-natural-questions", "special_tokens_map_file": "/home/vasu/.cache/huggingface/transformers/400be7e354ea6eb77319bcc7fa34899ec9fa2e3aff0fa677f6eb7e45a01b1548.75b358ecb30fa6b001d9d87bfde336c02d9123e7a8f5b90cc890d0f6efc3d4a3", "tokenizer_file": null, "tokenizer_class": "BigBirdTokenizer"}