---
dataset_info:
  features:
  - name: index
    dtype: int64
  - name: modelId
    dtype: string
  - name: config_model_type
    dtype: string
  - name: config_architectures
    dtype: string
  - name: config_vocab_size
    dtype: string
  - name: config_torch_dtype
    dtype: string
  - name: config_transformers_version
    dtype: string
  - name: config_hidden_size
    dtype: float64
  - name: config_intermediate_size
    dtype: float64
  - name: config_num_hidden_layers
    dtype: float64
  - name: config_num_attention_heads
    dtype: string
  - name: config_num_key_value_heads
    dtype: float64
  - name: config_hidden_act
    dtype: string
  - name: config_attention_dropout
    dtype: float64
  - name: config_use_cache
    dtype: string
  - name: config_max_position_embeddings
    dtype: float64
  - name: config_rope_theta
    dtype: float64
  - name: config_rms_norm_eps
    dtype: float64
  - name: config_initializer_range
    dtype: float64
  - name: config_bos_token_id
    dtype: string
  - name: config_eos_token_id
    dtype: string
  - name: config_tie_word_embeddings
    dtype: bool
  - name: config_head_dimension
    dtype: float64
  - name: config_gqa_ratio
    dtype: float64
  - name: config_moe_enabled
    dtype: bool
  - name: config_n_routed_experts
    dtype: float64
  - name: config_num_experts_per_tok
    dtype: float64
  - name: is_llama_family
    dtype: bool
  - name: is_bert_family
    dtype: bool
  - name: is_gpt_family
    dtype: bool
  - name: is_t5_family
    dtype: bool
  - name: is_whisper_family
    dtype: bool
  - name: is_deepseek_family
    dtype: bool
  - name: is_mistral_family
    dtype: bool
  - name: uses_moe
    dtype: bool
  - name: uses_gqa
    dtype: bool
  - name: uses_rope
    dtype: bool
  - name: config_approx_params_billions
    dtype: float64
  - name: size_category
    dtype: string
  - name: context_category
    dtype: string
  splits:
  - name: train
    num_bytes: 34934897
    num_examples: 125271
  download_size: 6129969
  dataset_size: 34934897
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---