Uploading tokenizer_robustness_completion_farsi_number_romanization subset

README.md

@@ -36,6 +36,10 @@ configs:
   data_files:
   - split: test
     path: tokenizer_robustness_completion_farsi_keyboard_proximity_errors/test-*
+- config_name: tokenizer_robustness_completion_farsi_number_romanization
+  data_files:
+  - split: test
+    path: tokenizer_robustness_completion_farsi_number_romanization/test-*
 dataset_info:
 - config_name: tokenizer_robustness_completion_farsi_arabic_keyboard_for_farsi
   features:
@@ -905,6 +909,130 @@ dataset_info:
     num_examples: 40
   download_size: 40791
   dataset_size: 23457
+- config_name: tokenizer_robustness_completion_farsi_number_romanization
+  features:
+  - name: question
+    dtype: string
+  - name: choices
+    list: string
+  - name: answer
+    dtype: int64
+  - name: answer_label
+    dtype: string
+  - name: split
+    dtype: string
+  - name: subcategories
+    dtype: string
+  - name: category
+    dtype: string
+  - name: lang
+    dtype: string
+  - name: second_lang
+    dtype: string
+  - name: notes
+    dtype: string
+  - name: id
+    dtype: string
+  - name: set_id
+    dtype: string
+  - name: variation_id
+    dtype: string
+  - name: vanilla_cos_sim_to_canonical
+    struct:
+    - name: CohereLabs/aya-expanse-8b
+      dtype: float64
+    - name: Qwen/Qwen3-8B
+      dtype: float64
+    - name: bigscience/bloom
+      dtype: float64
+    - name: common-pile/comma-v0.1-1t
+      dtype: float64
+    - name: facebook/xglm-564M
+      dtype: float64
+    - name: google-bert/bert-base-multilingual-cased
+      dtype: float64
+    - name: google/byt5-small
+      dtype: float64
+    - name: google/gemma-2-2b
+      dtype: float64
+    - name: gpt2
+      dtype: float64
+    - name: meta-llama/Llama-3.2-1B
+      dtype: float64
+    - name: microsoft/Phi-3-mini-4k-instruct
+      dtype: float64
+    - name: mistralai/tekken
+      dtype: float64
+    - name: tiktoken/gpt-4o
+      dtype: float64
+    - name: tokenmonster/englishcode-32000-consistent-v1
+      dtype: float64
+  - name: trimmed_cos_sim_to_canonical
+    struct:
+    - name: CohereLabs/aya-expanse-8b
+      dtype: float64
+    - name: Qwen/Qwen3-8B
+      dtype: float64
+    - name: bigscience/bloom
+      dtype: float64
+    - name: common-pile/comma-v0.1-1t
+      dtype: float64
+    - name: facebook/xglm-564M
+      dtype: float64
+    - name: google-bert/bert-base-multilingual-cased
+      dtype: float64
+    - name: google/byt5-small
+      dtype: float64
+    - name: google/gemma-2-2b
+      dtype: float64
+    - name: gpt2
+      dtype: float64
+    - name: meta-llama/Llama-3.2-1B
+      dtype: float64
+    - name: microsoft/Phi-3-mini-4k-instruct
+      dtype: float64
+    - name: mistralai/tekken
+      dtype: float64
+    - name: tiktoken/gpt-4o
+      dtype: float64
+    - name: tokenmonster/englishcode-32000-consistent-v1
+      dtype: float64
+  - name: token_counts
+    struct:
+    - name: CohereLabs/aya-expanse-8b
+      dtype: int64
+    - name: Qwen/Qwen3-8B
+      dtype: int64
+    - name: bigscience/bloom
+      dtype: int64
+    - name: common-pile/comma-v0.1-1t
+      dtype: int64
+    - name: facebook/xglm-564M
+      dtype: int64
+    - name: google-bert/bert-base-multilingual-cased
+      dtype: int64
+    - name: google/byt5-small
+      dtype: int64
+    - name: google/gemma-2-2b
+      dtype: int64
+    - name: gpt2
+      dtype: int64
+    - name: meta-llama/Llama-3.2-1B
+      dtype: int64
+    - name: microsoft/Phi-3-mini-4k-instruct
+      dtype: int64
+    - name: mistralai/tekken
+      dtype: int64
+    - name: tiktoken/gpt-4o
+      dtype: int64
+    - name: tokenmonster/englishcode-32000-consistent-v1
+      dtype: int64
+  splits:
+  - name: test
+    num_bytes: 7500
+    num_examples: 13
+  download_size: 30813
+  dataset_size: 7500
 ---
 
 # Dataset Card for Tokenization Robustness

tokenizer_robustness_completion_farsi_number_romanization/test-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78969a6c77f53688d30c68d12fd9a727270f85aceda5ebe7da20be257c94d473
+size 30813