Upload folder using huggingface_hub

main/QBModelConfig.py

@@ -0,0 +1,11 @@
+from transformers import PretrainedConfig
+import torch
+
+class QBModelConfig(PretrainedConfig):
+    model_type = 'QA-umd-quizbowl'
+
+    def __init__(self, **kwargs):
+        self.torch_dtype = torch.float16
+        super().__init__( **kwargs)
+        
+

main/QBModelWrapperCopy.py

@@ -0,0 +1,20 @@
+from typing import List
+from transformers import PreTrainedModel
+from transformers import PretrainedConfig
+from QBModelConfig import QBModelConfig
+from qbmodel import QuizBowlModel
+
+class QBModelWrapper(PreTrainedModel):
+    config_class= QBModelConfig
+
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.model = QuizBowlModel()
+        self.tfmodel = self.model.guesser
+
+    
+    def forward(self, question, context):
+        output = self.model.guess_and_buzz([question])
+        return output[0]

main/QBpipeline.py

@@ -0,0 +1,60 @@
+from transformers import Pipeline
+from transformers.utils import ModelOutput
+
+from transformers import PreTrainedModel, Pipeline
+from typing import Any, Dict, List
+
+class QApipeline(Pipeline):
+    def __init__(
+        self,
+        model: PreTrainedModel,
+        **kwargs
+    ):
+        super().__init__(
+            model=model,
+            **kwargs
+        )
+
+        print("in __init__")
+
+    def __call__( self, question: str,  context: str, **kwargs) -> Dict[str, Any]:
+        inputs = {
+            "question": question,
+            "context": context
+        }
+
+        outputs = self.model(**inputs)
+
+        answer = self._process_output(outputs)
+
+        print("in __call___")
+
+        return {"answer": answer}
+
+    def _process_output(self, outputs: Any) -> str:
+
+        print("in process outputs")
+
+        format =  {'guess': outputs[0], 'confidence': int(outputs[1])}
+        return format
+
+    
+    def _sanitize_parameters(self, **kwargs):
+        print("in sanatize params")
+
+        return {}, {}, {}
+
+    def preprocess(self, inputs):
+        print("in preprocess")
+
+        return inputs
+
+    def postprocess(self, outputs):
+        print("in postprocess")
+        format =  {'guess': outputs[0], 'confidence': float(outputs[1])}
+        return format
+    
+    def _forward(self, input_tensors, **forward_parameters: Dict) -> ModelOutput:
+        print("in _forward")
+        return super()._forward(input_tensors, **forward_parameters)
+    

main/config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "QBModelWrapper"
+  ],
+  "auto_map": {
+    "AutoConfig": "QBModelConfig.QBModelConfig",
+    "AutoModelForQuestionAnswering": "QBModelWrapperCopy.QBModelWrapper"
+  },
+  "custom_pipelines": {
+    "qa-pipeline-qb": {
+      "impl": "QBpipeline.QApipeline",
+      "pt": [
+        "AutoModelForQuestionAnswering"
+      ],
+      "tf": [
+        "TFAutoModelForQuestionAnswering"
+      ]
+    }
+  },
+  "model_type": "QA-umd-quizbowl",
+  "torch_dtype": "float16",
+  "transformers_version": "4.40.1"
+}

main/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4c64ac2a95e8002d6b3f02bc84e6dc4e980d5e592e6ad7897361ed2ad1462e0
+size 888

test-huggingface

@@ -37,6 +37,7 @@ PIPELINE_REGISTRY.register_pipeline(
 
 qa_pipe = pipeline("qa-pipeline-qb", model=qb_model)
 #qa_pipe.push_to_hub("new-attempt-pipeline-2", safe_serialization=False)
+qa_pipe.save_pretrained("main", safe_serialization=False)
 
 result = qa_pipe(question="This star in the solar system has 8 planets", context="Context for the question")
 print(result["answer"])

tfidf.py

@@ -8,7 +8,6 @@ import os
 from nltk.corpus import stopwords
 from nltk.tokenize import word_tokenize
 from nltk.stem import WordNetLemmatizer
-from models import processed_tfidf_wiki_page_text_model
 
 
 class TfidfWikiGuesser:
@@ -18,7 +17,6 @@ class TfidfWikiGuesser:
         self.titles = None 
         self.vectorizer = None 
         self.lemmatizer = WordNetLemmatizer()
-        mod_file = processed_tfidf_wiki_page_text_model
         model_file = "processed_tfidf_wiki_page_text_model.pkl" # <--- has best acc so far (using wiki_page_text.json from gdrive folder)
         #model_file = "processed_tfidf_wiki_16_model.pkl"
         # full_model_path = model_file