add task gqa-ru

lmms_eval/tasks/gqa_ru/gqa_ru.yaml

@@ -0,0 +1,29 @@
+dataset_path: deepvk/GQA-ru
+dataset_name: testdev_balanced_instructions
+dataset_kwargs:
+  token: True
+task: "gqa-ru"
+test_split: testdev
+output_type: generate_until
+doc_to_visual: !function utils.gqa_doc_to_visual
+doc_to_text: !function utils.gqa_doc_to_text
+doc_to_target: "answer"
+generation_kwargs:
+  max_new_tokens: 16
+  temperature: 0
+  top_p: 1.0
+  num_beams: 1
+  do_sample: false
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true
+metadata:
+  - version: 0.0
+
+model_specific_prompt_kwargs:
+  default:
+    pre_prompt: ""
+    post_prompt: "\nОтветь одним словом."

lmms_eval/tasks/gqa_ru/utils.py

@@ -0,0 +1,23 @@
+from datasets import load_dataset
+
+GQA_RAW_IMAGE_DATASET = None
+GQA_ID2IMAGE = None
+
+
+def gqa_doc_to_visual(doc):
+    global GQA_RAW_IMAGE_DATASET
+    global GQA_ID2IMAGE
+    if GQA_RAW_IMAGE_DATASET is None:
+        GQA_RAW_IMAGE_DATASET = load_dataset("deepvk/GQA-ru", "testdev_balanced_images", split="testdev", token=True)
+        GQA_ID2IMAGE = {}
+        for row in GQA_RAW_IMAGE_DATASET:
+            GQA_ID2IMAGE[row["id"]] = row["image"].convert("RGB")
+    image = GQA_ID2IMAGE[doc["imageId"]]
+    return [image]
+
+
+def gqa_doc_to_text(doc, model_specific_prompt_kwargs):
+    question = doc["question"]
+    pre_prompt = model_specific_prompt_kwargs["pre_prompt"]
+    post_prompt = model_specific_prompt_kwargs["post_prompt"]
+    return f"{pre_prompt}{question}{post_prompt}"