保存评估的sample级别（样本级别）的详细数据，包含完整的token序列、每个token的所有logits/概率\、以及该样本包含的所有Sapn信息，这样 前端就可以加载一个样本，展示争端对话，并高亮显示所有预测信息，同时也能看到哪些部分是标注的span

R1-Onevision-7B/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:865c8d45d60e668ef35d35197c0c50e3b1d35af818c16ae511bc0ca627df5aea
+size 79220788

README.md

@@ -0,0 +1,45 @@
+---
+dataset_info:
+  config_name: R1-Onevision-7B
+  features:
+  - name: tokens
+    list: string
+  - name: input_ids
+    list: int32
+  - name: probs
+    list:
+      list: float64
+  - name: preds
+    list: int64
+  - name: labels
+    list: float64
+  - name: spans
+    list:
+    - name: end
+      dtype: int64
+    - name: label
+      dtype: int64
+    - name: start
+      dtype: int64
+    - name: text
+      dtype: string
+    - name: type
+      dtype: string
+  - name: image
+    dtype: image
+  - name: context_text
+    dtype: string
+  - name: user_content
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 114098193
+    num_examples: 498
+  download_size: 79220788
+  dataset_size: 114098193
+configs:
+- config_name: R1-Onevision-7B
+  data_files:
+  - split: train
+    path: R1-Onevision-7B/train-*
+---