operators:




- id: hf_source_numina
  config:
    type: hf_source
    dataset: AI-MO/NuminaMath-CoT
    split: train
- id: select_amc_aime
  config:
    type: function
    function: data_strategies.commons.select_rows
    function_config:
      conditions:
        source: amc_aime
  input_ids:
    - hf_source_numina
- id: select_aops
  config:
    type: function
    function: data_strategies.commons.select_rows
    function_config:
      conditions:
        source: aops_forum
  input_ids:
    - hf_source_numina
- id: rename_task_column_numina
  config:
    type: function
    function: data_strategies.commons.force_rename_columns
    function_config:
      column_maps:
        problem: question
  input_ids:
  - select_amc_aime
  - select_aops

- id: hf_source_bad_seed_math
  config:
    type: load_preexisting
    framework_name: seed_math_math_instruct
- id: rename_task_column
  config:
    type: function
    function: data_strategies.commons.force_rename_columns
    function_config:
      column_maps:
        instruction_seed: question
  input_ids:
  - hf_source_bad_seed_math

- id: calc_embeddings_positive
  config:
    type: gpu_function
    num_gpus: 1
    num_shards: 1
    sharded: true
    function: data_strategies.commons.calc_sentence_transformers_embedding
    function_config:
      text_column: question
      model_name: intfloat/multilingual-e5-large-instruct
  input_ids:
    - rename_task_column_numina

- id: force_merge_shards_positive
  config:
    type: function
    function: data_strategies.commons.uniform_sample_fixed
    function_config:
      num_samples: 1_000_000
  input_ids:
  - calc_embeddings_positive

- id: calc_embeddings_negative
  config:
    type: gpu_function
    num_gpus: 1
    num_shards: 1
    sharded: true
    function: data_strategies.commons.calc_sentence_transformers_embedding
    function_config:
      text_column: question
      model_name: intfloat/multilingual-e5-large-instruct
  input_ids:
    - rename_task_column

- id: force_merge_shards_negative
  config:
    type: function
    function: data_strategies.commons.uniform_sample_fixed
    function_config:
      num_samples: 1_000_000
  input_ids:
  - calc_embeddings_negative

- id: seed_math_load_in
  config:
    type: load_preexisting
    framework_name: seed_math_load_in
- id: get_tiger_lab_math
  config:
    type: function
    function: data_strategies.commons.select_rows
    function_config:
      conditions:
        source: "tiger_lab_math"
  input_ids:
    - seed_math_load_in
- id: sample_tiger_lab_math_before
  config:
    type: function
    function: data_strategies.commons.uniform_sample_fixed
    function_config:
      num_samples: 75_000
  input_ids:
  - get_tiger_lab_math
- id: calc_embeddings_tiger_lab_math
  config:
    type: gpu_function
    num_gpus: 1
    num_shards: 1
    sharded: true
    function: data_strategies.commons.calc_sentence_transformers_embedding
    function_config:
      text_column: instruction_seed
      model_name: intfloat/multilingual-e5-large-instruct
  input_ids:
    - sample_tiger_lab_math_before

- id: force_merge_shards_actual
  config:
    type: function
    function: data_strategies.commons.uniform_sample_fixed
    function_config:
      num_samples: 1_000_000
  input_ids:
  - calc_embeddings_tiger_lab_math

- id: calc_embedding_difference_tiger_lab_math
  config:
    type: high_memory_function
    function: data_strategies.InstructionFiltering.EmbeddingFilter.utils.calc_embedding_score_mean
    function_config:
      embedding_column: embeddings
      positive_embedding_column: embeddings
      negative_embedding_column: embeddings
    input_dataset_map:
      dataset:  force_merge_shards_actual
      negative_dataset: force_merge_shards_negative
      positive_dataset: force_merge_shards_positive
  input_ids:
    - force_merge_shards_actual
    - force_merge_shards_negative
    - force_merge_shards_positive
# # - id: select_to_be_used_tiger_lab_math
# #   config:
# #     type: function
# #     function: data_strategies.InstructionFiltering.AskLLM.utils.filter_rank
# #     function_config:
# #       top_percentage: 0.10
# #       rating_column: difference_score
# #   input_ids:
# #     - calc_embedding_difference_tiger_lab_math
# # - id: sample_dataset_tiger_lab_math
# #   config:
# #     type: function
# #     function: data_strategies.commons.uniform_sample_fixed
# #     function_config:
# #       num_samples: 4_000
# #   input_ids:
# #   - select_to_be_used_tiger_lab_math