From 012546666b10339420d18472fb025413a1b82804 Mon Sep 17 00:00:00 2001
From: Kevin Wang <wangnu_043@126.com>
Date: Tue, 26 Sep 2023 14:31:15 +0800
Subject: [PATCH] [SIG] add WikiText-2&103 (#397)

* fix conflict

* add eval_cfg
---
 .../datasets/wikitext/wikitext_103_raw_ppl.py |  4 ++
 .../wikitext/wikitext_103_raw_ppl_752e2a.py   | 39 +++++++++++++++++++
 .../datasets/wikitext/wikitext_2_raw_ppl.py   |  4 ++
 .../wikitext/wikitext_2_raw_ppl_752e2a.py     | 39 +++++++++++++++++++
 4 files changed, 86 insertions(+)
 create mode 100644 configs/datasets/wikitext/wikitext_103_raw_ppl.py
 create mode 100644 configs/datasets/wikitext/wikitext_103_raw_ppl_752e2a.py
 create mode 100644 configs/datasets/wikitext/wikitext_2_raw_ppl.py
 create mode 100644 configs/datasets/wikitext/wikitext_2_raw_ppl_752e2a.py

diff --git a/configs/datasets/wikitext/wikitext_103_raw_ppl.py b/configs/datasets/wikitext/wikitext_103_raw_ppl.py
new file mode 100644
index 00000000..9a386cf0
--- /dev/null
+++ b/configs/datasets/wikitext/wikitext_103_raw_ppl.py
@@ -0,0 +1,4 @@
+from mmengine.config import read_base
+
+with read_base():
+    from .wikitext_103_raw_ppl_752e2a import wikitext_103_raw_datasets  # noqa: F401, F403
diff --git a/configs/datasets/wikitext/wikitext_103_raw_ppl_752e2a.py b/configs/datasets/wikitext/wikitext_103_raw_ppl_752e2a.py
new file mode 100644
index 00000000..02dea1e9
--- /dev/null
+++ b/configs/datasets/wikitext/wikitext_103_raw_ppl_752e2a.py
@@ -0,0 +1,39 @@
+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import GenInferencer
+from opencompass.openicl.icl_evaluator import AccEvaluator
+from opencompass.datasets import HFDataset
+
+
+wikitext_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template={0: '{text}'}
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer), # need a new ppl inferencer
+)
+
+wikitext_eval_cfg = dict(evaluator=dict(type=AccEvaluator), )
+
+wikitext_103_raw_datasets = []
+for _split in ["validation", "test"]:
+
+    wikitext_reader_cfg = dict(
+        input_columns=["text"],
+        output_column=None,
+        train_split="train",
+        test_split=_split,
+    )
+
+    wikitext_103_raw_datasets.append(
+        dict(
+            abbr=f"wikitext-103-raw-{_split}",
+            type=HFDataset,
+            path="wikitext",
+            name="wikitext-103-raw-v1",
+            reader_cfg=wikitext_reader_cfg,
+            infer_cfg=wikitext_infer_cfg,
+            eval_cfg=wikitext_eval_cfg,
+        )
+    )
diff --git a/configs/datasets/wikitext/wikitext_2_raw_ppl.py b/configs/datasets/wikitext/wikitext_2_raw_ppl.py
new file mode 100644
index 00000000..18549e79
--- /dev/null
+++ b/configs/datasets/wikitext/wikitext_2_raw_ppl.py
@@ -0,0 +1,4 @@
+from mmengine.config import read_base
+
+with read_base():
+    from .wikitext_2_raw_ppl_752e2a import wikitext_2_raw_datasets  # noqa: F401, F403
diff --git a/configs/datasets/wikitext/wikitext_2_raw_ppl_752e2a.py b/configs/datasets/wikitext/wikitext_2_raw_ppl_752e2a.py
new file mode 100644
index 00000000..41b3b872
--- /dev/null
+++ b/configs/datasets/wikitext/wikitext_2_raw_ppl_752e2a.py
@@ -0,0 +1,39 @@
+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import GenInferencer
+from opencompass.openicl.icl_evaluator import AccEvaluator
+from opencompass.datasets import HFDataset
+
+
+wikitext_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template={0: '{text}'}
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer), # need a new ppl inferencer
+)
+
+wikitext_eval_cfg = dict(evaluator=dict(type=AccEvaluator), )
+
+wikitext_2_raw_datasets = []
+for _split in ["validation", "test"]:
+
+    wikitext_reader_cfg = dict(
+        input_columns=["text"],
+        output_column=None,
+        train_split="train",
+        test_split=_split,
+    )
+
+    wikitext_2_raw_datasets.append(
+        dict(
+            abbr=f"wikitext-2-raw-{_split}",
+            type=HFDataset,
+            path="wikitext",
+            name="wikitext-2-raw-v1",
+            reader_cfg=wikitext_reader_cfg,
+            infer_cfg=wikitext_infer_cfg,
+            eval_cfg=wikitext_eval_cfg,
+        )
+    )