| { |
| "architecture": { |
| "attention_variant": "GQA", |
| "family": "image_text_to_text", |
| "is_moe": false, |
| "mixer": "attention", |
| "positional": "rope", |
| "sliding_window": 512, |
| "tie_word_embeddings": true, |
| "towers": [ |
| { |
| "attention_variant": "GQA", |
| "config_class": "Gemma4TextConfig", |
| "hidden_size": 2304, |
| "name": "text", |
| "num_attention_heads": 8, |
| "num_hidden_layers": 30, |
| "positional": "rope" |
| } |
| ], |
| "view": "multimodal" |
| }, |
| "capabilities": { |
| "attention_backends": [ |
| "eager", |
| "sdpa", |
| "flash_attention", |
| "flex_attention" |
| ], |
| "attention_patterns": [ |
| "sliding", |
| "causal" |
| ], |
| "attention_schedule": [ |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention" |
| ], |
| "task_heads": [ |
| "causal_lm", |
| "image_text_to_text", |
| "multimodal_lm", |
| "pretraining" |
| ], |
| "tensor_parallel": true |
| }, |
| "components": [ |
| { |
| "children": [ |
| "language_model" |
| ], |
| "class_name": "Gemma4Model", |
| "id": "model", |
| "kind": "model", |
| "path_pattern": "model" |
| }, |
| { |
| "children": [ |
| "language_model.embed_tokens", |
| "decoder_layers", |
| "language_model.norm", |
| "language_model.rotary_emb", |
| "language_model.embed_tokens_per_layer", |
| "language_model.per_layer_projection_norm" |
| ], |
| "class_name": "Gemma4TextModel", |
| "id": "language_model", |
| "kind": "model", |
| "path_pattern": "model.language_model" |
| }, |
| { |
| "attributes": { |
| "embedding_dim": "config.hidden_size", |
| "num_embeddings": "config.vocab_size", |
| "tied_lm_head": true |
| }, |
| "class_name": "Gemma4TextScaledWordEmbedding", |
| "id": "language_model.embed_tokens", |
| "kind": "embedding", |
| "path_pattern": "model.language_model.embed_tokens" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "language_model.norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.norm" |
| }, |
| { |
| "attributes": { |
| "head_dim": "config.head_dim", |
| "scheme": "rope" |
| }, |
| "class_name": "Gemma4TextRotaryEmbedding", |
| "id": "language_model.rotary_emb", |
| "kind": "position", |
| "path_pattern": "model.language_model.rotary_emb" |
| }, |
| { |
| "attributes": { |
| "embedding_dim": "config.hidden_size", |
| "num_embeddings": "config.vocab_size" |
| }, |
| "class_name": "Gemma4TextScaledWordEmbedding", |
| "id": "language_model.embed_tokens_per_layer", |
| "kind": "embedding", |
| "path_pattern": "model.language_model.embed_tokens_per_layer" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "language_model.per_layer_projection_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.per_layer_projection_norm" |
| } |
| ], |
| "config": { |
| "class_name": "Gemma4Config", |
| "model_type": "gemma4", |
| "module": "transformers.models.gemma4.configuration_gemma4", |
| "referenced_fields": {}, |
| "salient_fields": { |
| "is_encoder_decoder": false, |
| "tie_word_embeddings": true |
| } |
| }, |
| "edges": [ |
| { |
| "kind": "data", |
| "source": "language_model.embed_tokens", |
| "target": "decoder_layers" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layers", |
| "target": "language_model.norm" |
| }, |
| { |
| "kind": "data", |
| "source": "language_model.norm", |
| "target": "language_model.embed_tokens_per_layer" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.self_attn", |
| "target": "decoder_layer.mlp" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.mlp", |
| "target": "decoder_layer.input_layernorm" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.input_layernorm", |
| "target": "decoder_layer.post_attention_layernorm" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.post_attention_layernorm", |
| "target": "decoder_layer.pre_feedforward_layernorm" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.pre_feedforward_layernorm", |
| "target": "decoder_layer.post_feedforward_layernorm" |
| }, |
| { |
| "kind": "data", |
| "source": "decoder_layer.self_attn.k_norm", |
| "target": "decoder_layer.self_attn.v_norm" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "residual", |
| "source": "decoder_layer", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "residual", |
| "source": "decoder_layer", |
| "target": "decoder_layer.mlp" |
| }, |
| { |
| "kind": "position", |
| "source": "language_model.rotary_emb", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn", |
| "target": "decoder_layer.self_attn.q_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn", |
| "target": "decoder_layer.self_attn.k_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn", |
| "target": "decoder_layer.self_attn.v_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.q_proj", |
| "target": "decoder_layer.self_attn.q_norm" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.k_proj", |
| "target": "decoder_layer.self_attn.k_norm" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.q_norm", |
| "target": "decoder_layer.self_attn.o_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.k_norm", |
| "target": "decoder_layer.self_attn.o_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.v_proj", |
| "target": "decoder_layer.self_attn.o_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.mlp", |
| "target": "decoder_layer.mlp.gate_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.mlp", |
| "target": "decoder_layer.mlp.up_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.mlp.gate_proj", |
| "target": "decoder_layer.mlp.down_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.mlp.up_proj", |
| "target": "decoder_layer.mlp.down_proj" |
| } |
| ], |
| "extends": "gemma3", |
| "model_type": "gemma4", |
| "patches": [ |
| { |
| "added": { |
| "attrs": [ |
| "shared_kv_states" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Gemma3nModelOutputWithPast", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4ModelOutputWithPast" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "shared_kv_states" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Gemma3nCausalLMOutputWithPast", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4CausalLMOutputWithPast" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "shared_kv_states" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "BaseModelOutputWithPast", |
| "relation": "new", |
| "target_class": "Gemma4TextModelOutputWithPast" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "attention_mask" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "BaseModelOutputWithPooling", |
| "relation": "new", |
| "target_class": "Gemma4AudioModelOutput" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4ClippableLinear" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "inv_timescales" |
| ], |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioRelPositionalEncoding" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "_convert_to_block", |
| "_extract_block_context", |
| "_rel_shift", |
| "forward" |
| ] |
| }, |
| "component_kind": "attention", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioAttention" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "transformer_block", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioSubSampleConvProjectionLayer" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioSubSampleConvProjection" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "feed_forward", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioFeedForward" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "forward", |
| "left_pad" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Conv1d", |
| "relation": "new", |
| "target_class": "Gemma4AudioCausalConv1d" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioLightConv1d" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "transformer_block", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4AudioLayer" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "_position_embeddings", |
| "forward" |
| ] |
| }, |
| "component_kind": null, |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4VisionPatchEmbedder" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "_avg_pool_by_positions", |
| "forward" |
| ] |
| }, |
| "component_kind": "pooler", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4VisionPooler" |
| }, |
| { |
| "component_kind": "feed_forward", |
| "overridden": { |
| "methods": [ |
| "__init__" |
| ] |
| }, |
| "parent_class": "Gemma3MLP", |
| "relation": "inherits", |
| "target_class": "Gemma4VisionMLP" |
| }, |
| { |
| "component_kind": "position", |
| "overridden": { |
| "methods": [ |
| "compute_default_rope_parameters", |
| "forward" |
| ] |
| }, |
| "parent_class": "LlamaRotaryEmbedding", |
| "parent_model": "llama", |
| "relation": "inherits", |
| "target_class": "Gemma4VisionRotaryEmbedding" |
| }, |
| { |
| "component_kind": "attention", |
| "overridden": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3Attention", |
| "relation": "inherits", |
| "target_class": "Gemma4VisionAttention" |
| }, |
| { |
| "component_kind": "transformer_block", |
| "overridden": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3DecoderLayer", |
| "relation": "inherits", |
| "target_class": "Gemma4VisionEncoderLayer" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "encoder", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4VisionEncoder" |
| }, |
| { |
| "component_kind": "feed_forward", |
| "overridden": { |
| "methods": [ |
| "__init__" |
| ] |
| }, |
| "parent_class": "Gemma3MLP", |
| "relation": "inherits", |
| "target_class": "Gemma4TextMLP" |
| }, |
| { |
| "component_kind": "position", |
| "overridden": { |
| "methods": [ |
| "__init__" |
| ] |
| }, |
| "parent_class": "Gemma3RotaryEmbedding", |
| "relation": "inherits", |
| "target_class": "Gemma4TextRotaryEmbedding" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "attention", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4TextAttention" |
| }, |
| { |
| "component_kind": "moe", |
| "overridden": { |
| "methods": [ |
| "__init__" |
| ] |
| }, |
| "parent_class": "MixtralExperts", |
| "parent_model": "mixtral", |
| "relation": "inherits", |
| "target_class": "Gemma4TextExperts" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "moe", |
| "parent_class": "Module", |
| "relation": "new", |
| "target_class": "Gemma4TextRouter" |
| }, |
| { |
| "component_kind": "transformer_block", |
| "overridden": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3DecoderLayer", |
| "relation": "inherits", |
| "target_class": "Gemma4TextDecoderLayer" |
| }, |
| { |
| "component_kind": "model", |
| "overridden": { |
| "attrs": [ |
| "_can_record_outputs", |
| "_no_split_modules", |
| "input_modalities" |
| ], |
| "methods": [ |
| "_init_weights" |
| ] |
| }, |
| "parent_class": "Gemma3nPreTrainedModel", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4PreTrainedModel" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "_can_record_outputs" |
| ], |
| "methods": [ |
| "get_per_layer_inputs", |
| "project_per_layer_inputs" |
| ] |
| }, |
| "component_kind": "model", |
| "overridden": { |
| "attrs": [ |
| "config" |
| ], |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3TextModel", |
| "relation": "inherits", |
| "target_class": "Gemma4TextModel" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "base_model_prefix" |
| ] |
| }, |
| "component_kind": "lm_head", |
| "overridden": { |
| "methods": [ |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3ForCausalLM", |
| "relation": "inherits", |
| "target_class": "Gemma4ForCausalLM" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "_can_record_outputs", |
| "base_model_prefix", |
| "config", |
| "main_input_name" |
| ], |
| "methods": [ |
| "__init__", |
| "_convert_4d_mask_to_blocked_5d", |
| "forward" |
| ] |
| }, |
| "component_kind": "model", |
| "parent_class": "Gemma4PreTrainedModel", |
| "relation": "new", |
| "target_class": "Gemma4AudioModel" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "_can_record_outputs", |
| "config" |
| ], |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "component_kind": "model", |
| "parent_class": "Gemma4PreTrainedModel", |
| "relation": "new", |
| "target_class": "Gemma4VisionModel" |
| }, |
| { |
| "component_kind": null, |
| "overridden": { |
| "methods": [ |
| "__init__", |
| "forward" |
| ] |
| }, |
| "parent_class": "Gemma3nMultimodalEmbedder", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4MultimodalEmbedder" |
| }, |
| { |
| "added": { |
| "methods": [ |
| "get_video_features" |
| ] |
| }, |
| "component_kind": "model", |
| "overridden": { |
| "methods": [ |
| "__init__", |
| "forward", |
| "get_audio_features", |
| "get_image_features", |
| "get_per_layer_input_embeddings", |
| "get_placeholder_mask", |
| "set_per_layer_input_embeddings" |
| ] |
| }, |
| "parent_class": "Gemma3nModel", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4Model" |
| }, |
| { |
| "added": { |
| "attrs": [ |
| "base_model_prefix" |
| ], |
| "methods": [ |
| "create_masks_for_generate" |
| ] |
| }, |
| "component_kind": "lm_head", |
| "overridden": { |
| "methods": [ |
| "forward", |
| "get_image_features", |
| "get_per_layer_input_embeddings", |
| "prepare_inputs_for_generation", |
| "set_per_layer_input_embeddings" |
| ] |
| }, |
| "parent_class": "Gemma3nForConditionalGeneration", |
| "parent_model": "gemma3n", |
| "relation": "inherits", |
| "target_class": "Gemma4ForConditionalGeneration" |
| } |
| ], |
| "provenance": { |
| "config_class": "Gemma4Config", |
| "config_module": "transformers.models.gemma4.configuration_gemma4", |
| "model_class": "Gemma4Model", |
| "model_module": "transformers.models.gemma4.modeling_gemma4" |
| }, |
| "repeats": [ |
| { |
| "body": "decoder_layer", |
| "container_path_pattern": "model.language_model.layers", |
| "count": 30, |
| "count_expr": "30", |
| "count_source": "module_tree", |
| "id": "decoder_layers", |
| "index_symbol": "i", |
| "item_path_pattern": "model.language_model.layers.{i}", |
| "kind": "symbolic_repeat", |
| "provenance": { |
| "class_name": "Gemma4TextDecoderLayer", |
| "container_path_pattern": "model.language_model.layers", |
| "item_path_pattern": "model.language_model.layers.{i}", |
| "source": "module_tree_repeat_collapse" |
| }, |
| "repeated_class_name": "Gemma4TextDecoderLayer" |
| } |
| ], |
| "schema_version": "architecture-template-v0", |
| "templates": [ |
| { |
| "children": [ |
| "decoder_layer.self_attn", |
| "decoder_layer.mlp", |
| "decoder_layer.input_layernorm", |
| "decoder_layer.post_attention_layernorm", |
| "decoder_layer.pre_feedforward_layernorm", |
| "decoder_layer.post_feedforward_layernorm", |
| "decoder_layer.post_per_layer_input_norm" |
| ], |
| "class_name": "Gemma4TextDecoderLayer", |
| "id": "decoder_layer", |
| "kind": "transformer_block", |
| "path_pattern": "model.language_model.layers.{i}" |
| }, |
| { |
| "attributes": { |
| "head_dim": 256, |
| "n_heads": 8, |
| "n_kv_heads": 4, |
| "qk_norm": true, |
| "rope": true, |
| "sliding_window": 512, |
| "variant": "GQA" |
| }, |
| "children": [ |
| "decoder_layer.self_attn.q_proj", |
| "decoder_layer.self_attn.q_norm", |
| "decoder_layer.self_attn.k_norm", |
| "decoder_layer.self_attn.v_norm", |
| "decoder_layer.self_attn.k_proj", |
| "decoder_layer.self_attn.v_proj", |
| "decoder_layer.self_attn.o_proj" |
| ], |
| "class_name": "Gemma4TextAttention", |
| "id": "decoder_layer.self_attn", |
| "kind": "attention", |
| "path_pattern": "model.language_model.layers.{i}.self_attn" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": 2048 |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.q_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.q_proj" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.self_attn.q_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.q_norm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.self_attn.k_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.k_norm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.self_attn.v_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.v_norm" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": 1024 |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.k_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.k_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": 1024 |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.v_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.v_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": 2048, |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.o_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.self_attn.o_proj" |
| }, |
| { |
| "attributes": { |
| "hidden_size": 2304, |
| "intermediate_size": 9216 |
| }, |
| "children": [ |
| "decoder_layer.mlp.gate_proj", |
| "decoder_layer.mlp.up_proj", |
| "decoder_layer.mlp.down_proj" |
| ], |
| "class_name": "Gemma4TextMLP", |
| "id": "decoder_layer.mlp", |
| "kind": "feed_forward", |
| "path_pattern": "model.language_model.layers.{i}.mlp" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.intermediate_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.mlp.gate_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.mlp.gate_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.intermediate_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.mlp.up_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.mlp.up_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.intermediate_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.mlp.down_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.layers.{i}.mlp.down_proj" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.input_layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.input_layernorm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.post_attention_layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.post_attention_layernorm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.pre_feedforward_layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.pre_feedforward_layernorm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.post_feedforward_layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.post_feedforward_layernorm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "rms" |
| }, |
| "class_name": "Gemma4RMSNorm", |
| "id": "decoder_layer.post_per_layer_input_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.layers.{i}.post_per_layer_input_norm" |
| } |
| ] |
| } |
|
|
