janky seq2seq support (will be reverted)

Author: Vipitis

bigcode_eval/tasks/shadereval.py

@@ -133,6 +133,8 @@ def __init__(self, prompt="minimal"):
             requires_execution=True, #we run shadercode - could that be harmful? (all in the metric)
         )
         self.prompt = prompt # "minimal" or "full". "minimal" is the function header and comments before/after it, "full" is the whole code up untill the function declaration ends
+        # while we develop this dataset, we use a private dataset, so we overwrite the init call here, which silently fails -.- (warn doesn't show up)
+        self.dataset = datasets.load_dataset(path=self.DATASET_PATH, name=self.DATASET_NAME, trust_remote_code=True, use_auth_token=True)
 
     def get_dataset(self):
         # TODO replace with subset once that is set up
@@ -160,6 +162,13 @@ def get_prompt(self, doc):
         # only have one alternative, but could be more?
         model_context += doc["model_ctx"]
         return model_context
+    
+    def get_prompt_encoder(self, doc):
+        """
+        this is needed for seq2seq models, but not availabel by default?
+        """
+        enc_prompt = doc["model_ctx"] + "<extra_id_0>" #magic token to trigger generation for CodeT5p?
+        return enc_prompt
 
     def get_reference(self, doc):
         # TODO: get the reference solution from a sample `doc` from the dataset
@@ -213,6 +222,7 @@ def postprocess_generation(self, generation, idx):
         # from: https://huggingface.co/spaces/Vipitis/ShaderCoder/blob/main/utils/tree_utils.py#L45
         # generation = ShaderCoder.utils.parse_functions(generation)[0].text.decode() #not easily imported...
 
+        print(generation)
 
         # assemble into the full code with just the function replaced
         ref = self.dataset["test"][idx]

bigcode_eval/utils.py

@@ -281,11 +281,11 @@ def complete_code(
                     )
                 else:
                     generated_tokens = model.generate(
-                        decoder_input_ids=inputs,
+                        # decoder_input_ids=inputs,
                         input_ids=batch["ids_encoder"][:, : batch["input_len_encoder"]],
                         num_return_sequences=batch_size,
-                        decoder_start_token_id=tokenizer.pad_token_id,
-                        eos_token_id=tokenizer.eos_token_id,
+                        # decoder_start_token_id=tokenizer.pad_token_id,
+                        # eos_token_id=tokenizer.eos_token_id,
                         **gen_kwargs,
                     )
             else: