config.json

{
  "architectures": [
    "Gemma3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attn_logit_softcapping": null,
  "attn_mechanism": "splash",
  "axis_dims": [
    1,
    -1,
    2,
    1
  ],
  "axis_names": [
    "dp",
    "fsdp",
    "tp",
    "sp"
  ],
  "backend": null,
  "bits": null,
  "blocksize_b": 1,
  "blocksize_k": 128,
  "blocksize_q": 128,
  "bos_token_id": 2,
  "cache_implementation": "hybrid",
  "dcn_axis_dims": null,
  "easy_method": "train",
  "eos_token_id": 1,
  "fcm_max_ratio": 0.0,
  "fcm_min_ratio": 0.0,
  "final_logit_softcapping": null,
  "flash_attention_backward_pass_impl": "triton",
  "freq_max_position_embeddings": 8192,
  "gradient_checkpointing": "",
  "hardware_abstraction": false,
  "head_dim": 256,
  "hidden_activation": "gelu_pytorch_tanh",
  "hidden_size": 2560,
  "initializer_range": 0.02,
  "intermediate_size": 10240,
  "kv_cache_quantization_blocksize": 64,
  "kv_cache_quantization_method": "None",
  "kv_cache_sharding_sequence_axis_name": "sp",
  "mask_max_position_embeddings": 8192,
  "max_position_embeddings": 131072,
  "model_type": "gemma3_text",
  "num_attention_heads": 8,
  "num_hidden_layers": 34,
  "num_key_value_heads": 4,
  "pad_token_id": 0,
  "pallas_k_block_size": 128,
  "pallas_m_block_size": 128,
  "pallas_n_block_size": 128,
  "partition_axis": {
    "attention_dim_axis": null,
    "batch_axis": [
      "fsdp",
      "dp"
    ],
    "bias_head_sequence_axis": null,
    "bias_key_sequence_axis": null,
    "data_parallel_axis": "dp",
    "expert_axis": "ep",
    "expert_gate_axis": null,
    "expert_parallel_axis": "ep",
    "fully_sharded_data_parallel_axis": "fsdp",
    "generation_attention_dim_axis": null,
    "generation_batch_axis": null,
    "generation_head_axis": "tp",
    "generation_key_sequence_axis": "sp",
    "generation_query_sequence_axis": null,
    "head_axis": "tp",
    "hidden_state_axis": "tp",
    "key_sequence_axis": "sp",
    "mlp_intermediate_axis": "tp",
    "query_sequence_axis": "sp",
    "sequence_axis": "sp",
    "sequence_parallel_axis": "sp",
    "tensor_parallel_axis": "tp",
    "vocab_axis": "tp"
  },
  "platform": "jax",
  "precompute_masks": true,
  "pretraining_tp": 1,
  "quantization_blocksize": 64,
  "quantization_method": "None",
  "quantization_pattern": ".*",
  "query_pre_attn_scalar": 256,
  "rms_norm_eps": 1e-06,
  "rope_local_base_freq": 10000.0,
  "rope_scaling": {
    "factor": 8.0,
    "rope_type": "linear"
  },
  "rope_theta": 1000000.0,
  "scan_attention_layers": false,
  "scan_layers": false,
  "scan_mlp_chunk_size": 1024,
  "scan_ring_attention": true,
  "sequence_axis_name": "sp",
  "shard_attention_computation": true,
  "sliding_window": 1024,
  "sliding_window_pattern": 6,
  "transformers_version": "4.50.3",
  "use_cache": true,
  "use_scan_mlp": false,
  "use_sharded_kv_caching": false,
  "use_sharding_constraint": false,
  "vocab_size": 262208
}