| { |
| "architecture": { |
| "attention_variant": "MHA", |
| "family": "image_text_to_text", |
| "is_moe": false, |
| "mixer": "attention", |
| "positional": "learned", |
| "tie_word_embeddings": true, |
| "towers": [ |
| { |
| "attention_variant": "MHA", |
| "config_class": "OPTConfig", |
| "hidden_size": 768, |
| "name": "text", |
| "num_attention_heads": 12, |
| "num_hidden_layers": 12, |
| "positional": "learned" |
| }, |
| { |
| "attention_variant": "MHA", |
| "config_class": "Blip2VisionConfig", |
| "hidden_size": 1408, |
| "name": "vision", |
| "num_attention_heads": 16, |
| "num_hidden_layers": 39 |
| } |
| ], |
| "view": "multimodal" |
| }, |
| "capabilities": { |
| "attention_backends": [ |
| "eager", |
| "sdpa", |
| "flex_attention" |
| ], |
| "attention_patterns": [ |
| "causal" |
| ], |
| "attention_schedule": null, |
| "task_heads": [ |
| "image_text_to_text", |
| "multimodal_lm", |
| "visual_question_answering", |
| "zero_shot_image_classification" |
| ], |
| "tensor_parallel": false |
| }, |
| "components": [ |
| { |
| "children": [ |
| "vision_model", |
| "qformer", |
| "language_projection", |
| "language_model" |
| ], |
| "class_name": "Blip2Model", |
| "id": "model", |
| "kind": "model", |
| "path_pattern": "model" |
| }, |
| { |
| "children": [ |
| "vision_model.embeddings", |
| "vision_model.encoder", |
| "vision_model.post_layernorm" |
| ], |
| "class_name": "Blip2VisionModel", |
| "id": "vision_model", |
| "kind": "model", |
| "path_pattern": "model.vision_model" |
| }, |
| { |
| "attributes": { |
| "embedding_dim": "config.hidden_size" |
| }, |
| "children": [ |
| "vision_model.embeddings.patch_embedding" |
| ], |
| "class_name": "Blip2VisionEmbeddings", |
| "id": "vision_model.embeddings", |
| "kind": "embedding", |
| "path_pattern": "model.vision_model.embeddings" |
| }, |
| { |
| "class_name": "Conv2d", |
| "id": "vision_model.embeddings.patch_embedding", |
| "kind": "convolution", |
| "path_pattern": "model.vision_model.embeddings.patch_embedding" |
| }, |
| { |
| "children": [ |
| "encoder_layers" |
| ], |
| "class_name": "Blip2Encoder", |
| "id": "vision_model.encoder", |
| "kind": "encoder", |
| "path_pattern": "model.vision_model.encoder" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "vision_model.post_layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.vision_model.post_layernorm" |
| }, |
| { |
| "children": [ |
| "qformer.layernorm", |
| "qformer.encoder" |
| ], |
| "class_name": "Blip2QFormerModel", |
| "id": "qformer", |
| "kind": "model", |
| "path_pattern": "model.qformer" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "qformer.layernorm", |
| "kind": "normalization", |
| "path_pattern": "model.qformer.layernorm" |
| }, |
| { |
| "children": [ |
| "encoder_layer_2s" |
| ], |
| "class_name": "Blip2QFormerEncoder", |
| "id": "qformer.encoder", |
| "kind": "encoder", |
| "path_pattern": "model.qformer.encoder" |
| }, |
| { |
| "attributes": { |
| "in_features": 768, |
| "out_features": 768 |
| }, |
| "class_name": "Linear", |
| "id": "language_projection", |
| "kind": "projection", |
| "path_pattern": "model.language_projection" |
| }, |
| { |
| "children": [ |
| "language_model.model" |
| ], |
| "class_name": "OPTForCausalLM", |
| "id": "language_model", |
| "kind": "model", |
| "path_pattern": "model.language_model" |
| }, |
| { |
| "children": [ |
| "language_model.model.decoder" |
| ], |
| "class_name": "OPTModel", |
| "id": "language_model.model", |
| "kind": "model", |
| "path_pattern": "model.language_model.model" |
| }, |
| { |
| "children": [ |
| "language_model.model.decoder.embed_tokens", |
| "language_model.model.decoder.embed_positions", |
| "language_model.model.decoder.final_layer_norm", |
| "decoder_layers" |
| ], |
| "class_name": "OPTDecoder", |
| "id": "language_model.model.decoder", |
| "kind": "decoder", |
| "path_pattern": "model.language_model.model.decoder" |
| }, |
| { |
| "attributes": { |
| "embedding_dim": "config.hidden_size", |
| "num_embeddings": "config.vocab_size", |
| "tied_lm_head": true |
| }, |
| "class_name": "Embedding", |
| "id": "language_model.model.decoder.embed_tokens", |
| "kind": "embedding", |
| "path_pattern": "model.language_model.model.decoder.embed_tokens" |
| }, |
| { |
| "attributes": { |
| "max_position_embeddings": "config.max_position_embeddings", |
| "scheme": "learned" |
| }, |
| "class_name": "OPTLearnedPositionalEmbedding", |
| "id": "language_model.model.decoder.embed_positions", |
| "kind": "position", |
| "path_pattern": "model.language_model.model.decoder.embed_positions" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "language_model.model.decoder.final_layer_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.model.decoder.final_layer_norm" |
| } |
| ], |
| "config": { |
| "class_name": "Blip2Config", |
| "model_type": "blip-2", |
| "module": "transformers.models.blip_2.configuration_blip_2", |
| "referenced_fields": {}, |
| "salient_fields": { |
| "is_encoder_decoder": false |
| } |
| }, |
| "edges": [ |
| { |
| "kind": "data", |
| "source": "vision_model", |
| "target": "qformer" |
| }, |
| { |
| "kind": "data", |
| "source": "vision_model.embeddings", |
| "target": "vision_model.encoder" |
| }, |
| { |
| "kind": "data", |
| "source": "vision_model.encoder", |
| "target": "vision_model.post_layernorm" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer.self_attn", |
| "target": "encoder_layer.layer_norm1" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer.layer_norm1", |
| "target": "encoder_layer.mlp" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer.mlp", |
| "target": "encoder_layer.layer_norm2" |
| }, |
| { |
| "kind": "data", |
| "source": "qformer.layernorm", |
| "target": "qformer.encoder" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer_2.attention", |
| "target": "encoder_layer_2.crossattention" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer_2.crossattention", |
| "target": "encoder_layer_2.intermediate_query" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer_2.crossattention.attention", |
| "target": "encoder_layer_2.crossattention.output" |
| }, |
| { |
| "kind": "data", |
| "source": "encoder_layer_2.attention", |
| "target": "encoder_layer_2.intermediate_query" |
| }, |
| { |
| "kind": "data", |
| "source": "language_model.model.decoder.embed_tokens", |
| "target": "language_model.model.decoder.final_layer_norm" |
| }, |
| { |
| "kind": "data", |
| "source": "language_model.model.decoder.final_layer_norm", |
| "target": "decoder_layers" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer.self_attn" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.attention" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.attention.attention" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.crossattention" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.crossattention.attention" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.crossattention.output" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "encoder_layer_2.crossattention.output.LayerNorm" |
| }, |
| { |
| "kind": "mask", |
| "source": "input:attention_mask", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "residual", |
| "source": "encoder_layer", |
| "target": "encoder_layer.self_attn" |
| }, |
| { |
| "kind": "residual", |
| "source": "encoder_layer", |
| "target": "encoder_layer.mlp" |
| }, |
| { |
| "kind": "residual", |
| "source": "encoder_layer_2", |
| "target": "encoder_layer_2.attention" |
| }, |
| { |
| "kind": "residual", |
| "source": "encoder_layer_2", |
| "target": "encoder_layer_2.crossattention" |
| }, |
| { |
| "kind": "residual", |
| "source": "encoder_layer_2", |
| "target": "encoder_layer_2.intermediate_query" |
| }, |
| { |
| "kind": "residual", |
| "source": "decoder_layer", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "position", |
| "source": "language_model.model.decoder.embed_positions", |
| "target": "decoder_layer.self_attn" |
| }, |
| { |
| "kind": "cross_attention", |
| "source": "vision_model.encoder", |
| "target": "encoder_layer_2.crossattention" |
| }, |
| { |
| "kind": "cross_attention", |
| "source": "vision_model.encoder", |
| "target": "encoder_layer_2.crossattention.attention" |
| }, |
| { |
| "kind": "cross_attention", |
| "source": "vision_model.encoder", |
| "target": "encoder_layer_2.crossattention.output" |
| }, |
| { |
| "kind": "cross_attention", |
| "source": "vision_model.encoder", |
| "target": "encoder_layer_2.crossattention.output.LayerNorm" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer.self_attn", |
| "target": "encoder_layer.self_attn.qkv" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer.self_attn", |
| "target": "encoder_layer.self_attn.projection" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer.mlp", |
| "target": "encoder_layer.mlp.fc1" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer.mlp.fc1", |
| "target": "encoder_layer.mlp.fc2" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer_2.attention.attention", |
| "target": "encoder_layer_2.attention.attention.query" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer_2.attention.attention", |
| "target": "encoder_layer_2.attention.attention.key" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer_2.attention.attention", |
| "target": "encoder_layer_2.attention.attention.value" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "encoder_layer_2.intermediate_query", |
| "target": "encoder_layer_2.intermediate_query.dense" |
| }, |
| { |
| "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", |
| "target": "decoder_layer.self_attn.q_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.k_proj", |
| "target": "decoder_layer.self_attn.out_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.v_proj", |
| "target": "decoder_layer.self_attn.out_proj" |
| }, |
| { |
| "kind": "data", |
| "provenance": "intra_module", |
| "source": "decoder_layer.self_attn.q_proj", |
| "target": "decoder_layer.self_attn.out_proj" |
| } |
| ], |
| "extends": null, |
| "model_type": "blip-2", |
| "provenance": { |
| "config_class": "Blip2Config", |
| "config_module": "transformers.models.blip_2.configuration_blip_2", |
| "model_class": "Blip2Model", |
| "model_module": "transformers.models.blip_2.modeling_blip_2" |
| }, |
| "repeats": [ |
| { |
| "body": "encoder_layer", |
| "container_path_pattern": "model.vision_model.encoder.layers", |
| "count": 39, |
| "count_expr": "39", |
| "count_source": "module_tree", |
| "id": "encoder_layers", |
| "index_symbol": "i", |
| "item_path_pattern": "model.vision_model.encoder.layers.{i}", |
| "kind": "symbolic_repeat", |
| "provenance": { |
| "class_name": "Blip2EncoderLayer", |
| "container_path_pattern": "model.vision_model.encoder.layers", |
| "item_path_pattern": "model.vision_model.encoder.layers.{i}", |
| "source": "module_tree_repeat_collapse" |
| }, |
| "repeated_class_name": "Blip2EncoderLayer" |
| }, |
| { |
| "body": "encoder_layer_2", |
| "container_path_pattern": "model.qformer.encoder.layer", |
| "count": 12, |
| "count_expr": "12", |
| "count_source": "module_tree", |
| "id": "encoder_layer_2s", |
| "index_symbol": "i", |
| "item_path_pattern": "model.qformer.encoder.layer.{i}", |
| "kind": "symbolic_repeat", |
| "provenance": { |
| "class_name": "Blip2QFormerLayer", |
| "container_path_pattern": "model.qformer.encoder.layer", |
| "item_path_pattern": "model.qformer.encoder.layer.{i}", |
| "source": "module_tree_repeat_collapse" |
| }, |
| "repeated_class_name": "Blip2QFormerLayer" |
| }, |
| { |
| "body": "decoder_layer", |
| "container_path_pattern": "model.language_model.model.decoder.layers", |
| "count": 12, |
| "count_expr": "12", |
| "count_source": "module_tree", |
| "id": "decoder_layers", |
| "index_symbol": "i", |
| "item_path_pattern": "model.language_model.model.decoder.layers.{i}", |
| "kind": "symbolic_repeat", |
| "provenance": { |
| "class_name": "OPTDecoderLayer", |
| "container_path_pattern": "model.language_model.model.decoder.layers", |
| "item_path_pattern": "model.language_model.model.decoder.layers.{i}", |
| "source": "module_tree_repeat_collapse" |
| }, |
| "repeated_class_name": "OPTDecoderLayer" |
| } |
| ], |
| "schema_version": "architecture-template-v0", |
| "templates": [ |
| { |
| "children": [ |
| "encoder_layer.self_attn", |
| "encoder_layer.layer_norm1", |
| "encoder_layer.mlp", |
| "encoder_layer.layer_norm2" |
| ], |
| "class_name": "Blip2EncoderLayer", |
| "id": "encoder_layer", |
| "kind": "transformer_block", |
| "path_pattern": "model.vision_model.encoder.layers.{i}" |
| }, |
| { |
| "attributes": { |
| "head_dim": 88, |
| "n_heads": 16, |
| "n_kv_heads": 16, |
| "rope": false, |
| "variant": "MHA" |
| }, |
| "children": [ |
| "encoder_layer.self_attn.qkv", |
| "encoder_layer.self_attn.projection" |
| ], |
| "class_name": "Blip2Attention", |
| "id": "encoder_layer.self_attn", |
| "kind": "attention", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": 4224 |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer.self_attn.qkv", |
| "kind": "projection", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn.qkv" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer.self_attn.projection", |
| "kind": "projection", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn.projection" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "encoder_layer.layer_norm1", |
| "kind": "normalization", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.layer_norm1" |
| }, |
| { |
| "attributes": { |
| "activation": "gelu", |
| "hidden_size": 1408, |
| "intermediate_size": 6144 |
| }, |
| "children": [ |
| "encoder_layer.mlp.fc1", |
| "encoder_layer.mlp.fc2" |
| ], |
| "class_name": "Blip2MLP", |
| "id": "encoder_layer.mlp", |
| "kind": "feed_forward", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.mlp" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.intermediate_size" |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer.mlp.fc1", |
| "kind": "projection", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.mlp.fc1" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.intermediate_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer.mlp.fc2", |
| "kind": "projection", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.mlp.fc2" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "encoder_layer.layer_norm2", |
| "kind": "normalization", |
| "path_pattern": "model.vision_model.encoder.layers.{i}.layer_norm2" |
| }, |
| { |
| "children": [ |
| "encoder_layer_2.attention", |
| "encoder_layer_2.crossattention", |
| "encoder_layer_2.intermediate_query", |
| "encoder_layer_2.output_query" |
| ], |
| "class_name": "Blip2QFormerLayer", |
| "id": "encoder_layer_2", |
| "kind": "transformer_block", |
| "path_pattern": "model.qformer.encoder.layer.{i}" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "children": [ |
| "encoder_layer_2.attention.attention", |
| "encoder_layer_2.attention.output" |
| ], |
| "class_name": "Blip2QFormerAttention", |
| "id": "encoder_layer_2.attention", |
| "kind": "attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "children": [ |
| "encoder_layer_2.attention.attention.query", |
| "encoder_layer_2.attention.attention.key", |
| "encoder_layer_2.attention.attention.value" |
| ], |
| "class_name": "Blip2QFormerMultiHeadAttention", |
| "id": "encoder_layer_2.attention.attention", |
| "kind": "attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.attention" |
| }, |
| { |
| "attributes": { |
| "in_features": 768, |
| "out_features": 768 |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer_2.attention.attention.query", |
| "kind": "projection", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.attention.query" |
| }, |
| { |
| "attributes": { |
| "in_features": 768, |
| "out_features": 768 |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer_2.attention.attention.key", |
| "kind": "projection", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.attention.key" |
| }, |
| { |
| "attributes": { |
| "in_features": 768, |
| "out_features": 768 |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer_2.attention.attention.value", |
| "kind": "projection", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.attention.value" |
| }, |
| { |
| "children": [ |
| "encoder_layer_2.attention.output.LayerNorm" |
| ], |
| "class_name": "Blip2QFormerSelfOutput", |
| "id": "encoder_layer_2.attention.output", |
| "kind": "module", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.output" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "encoder_layer_2.attention.output.LayerNorm", |
| "kind": "normalization", |
| "path_pattern": "model.qformer.encoder.layer.{i}.attention.output.LayerNorm" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "children": [ |
| "encoder_layer_2.crossattention.attention", |
| "encoder_layer_2.crossattention.output" |
| ], |
| "class_name": "Blip2QFormerAttention", |
| "id": "encoder_layer_2.crossattention", |
| "kind": "cross_attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.crossattention" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "class_name": "Blip2QFormerMultiHeadAttention", |
| "id": "encoder_layer_2.crossattention.attention", |
| "kind": "cross_attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.crossattention.attention" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "children": [ |
| "encoder_layer_2.crossattention.output.LayerNorm" |
| ], |
| "class_name": "Blip2QFormerSelfOutput", |
| "id": "encoder_layer_2.crossattention.output", |
| "kind": "cross_attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.crossattention.output" |
| }, |
| { |
| "attributes": { |
| "rope": false |
| }, |
| "class_name": "LayerNorm", |
| "id": "encoder_layer_2.crossattention.output.LayerNorm", |
| "kind": "cross_attention", |
| "path_pattern": "model.qformer.encoder.layer.{i}.crossattention.output.LayerNorm" |
| }, |
| { |
| "children": [ |
| "encoder_layer_2.intermediate_query.dense" |
| ], |
| "class_name": "Blip2QFormerIntermediate", |
| "id": "encoder_layer_2.intermediate_query", |
| "kind": "feed_forward", |
| "path_pattern": "model.qformer.encoder.layer.{i}.intermediate_query" |
| }, |
| { |
| "attributes": { |
| "in_features": 768, |
| "out_features": 3072 |
| }, |
| "class_name": "Linear", |
| "id": "encoder_layer_2.intermediate_query.dense", |
| "kind": "projection", |
| "path_pattern": "model.qformer.encoder.layer.{i}.intermediate_query.dense" |
| }, |
| { |
| "children": [ |
| "encoder_layer_2.output_query.LayerNorm" |
| ], |
| "class_name": "Blip2QFormerOutput", |
| "id": "encoder_layer_2.output_query", |
| "kind": "module", |
| "path_pattern": "model.qformer.encoder.layer.{i}.output_query" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "encoder_layer_2.output_query.LayerNorm", |
| "kind": "normalization", |
| "path_pattern": "model.qformer.encoder.layer.{i}.output_query.LayerNorm" |
| }, |
| { |
| "children": [ |
| "decoder_layer.self_attn", |
| "decoder_layer.self_attn_layer_norm", |
| "decoder_layer.final_layer_norm" |
| ], |
| "class_name": "OPTDecoderLayer", |
| "id": "decoder_layer", |
| "kind": "transformer_block", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}" |
| }, |
| { |
| "attributes": { |
| "head_dim": 64, |
| "n_heads": 12, |
| "n_kv_heads": 12, |
| "rope": false, |
| "variant": "MHA" |
| }, |
| "children": [ |
| "decoder_layer.self_attn.k_proj", |
| "decoder_layer.self_attn.v_proj", |
| "decoder_layer.self_attn.q_proj", |
| "decoder_layer.self_attn.out_proj" |
| ], |
| "class_name": "OPTAttention", |
| "id": "decoder_layer.self_attn", |
| "kind": "attention", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.k_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn.k_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.v_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn.v_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.q_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn.q_proj" |
| }, |
| { |
| "attributes": { |
| "in_features": "config.hidden_size", |
| "out_features": "config.hidden_size" |
| }, |
| "class_name": "Linear", |
| "id": "decoder_layer.self_attn.out_proj", |
| "kind": "projection", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn.out_proj" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "decoder_layer.self_attn_layer_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.self_attn_layer_norm" |
| }, |
| { |
| "attributes": { |
| "norm_type": "layer" |
| }, |
| "class_name": "LayerNorm", |
| "id": "decoder_layer.final_layer_norm", |
| "kind": "normalization", |
| "path_pattern": "model.language_model.model.decoder.layers.{i}.final_layer_norm" |
| } |
| ] |
| } |
|
|
