| { |
| "model_name": "embeddinggemma-300m", |
| "architecture": "gemma3-embedding", |
| "tokenizer_repo": "google/embeddinggemma-300m", |
| "parts": { |
| "encoder": "encoder.mlpackage" |
| }, |
| "io_contract": { |
| "inputs": { |
| "input_ids": { |
| "shape": [ |
| 1, |
| 128 |
| ], |
| "dtype": "int32" |
| }, |
| "attention_mask": { |
| "shape": [ |
| 1, |
| 128 |
| ], |
| "dtype": "fp16", |
| "doc": "1.0 for valid tokens, 0.0 for pad" |
| } |
| }, |
| "outputs": { |
| "embedding": { |
| "shape": [ |
| 1, |
| 768 |
| ], |
| "dtype": "fp16", |
| "doc": "L2-normalized; Matryoshka-truncate the last dim" |
| } |
| } |
| }, |
| "hidden_size": 768, |
| "num_hidden_layers": 24, |
| "num_attention_heads": 3, |
| "num_key_value_heads": 1, |
| "head_dim": 256, |
| "intermediate_size": 1152, |
| "vocab_size": 262144, |
| "max_seq_len": 128, |
| "sliding_window": 512, |
| "sliding_window_pattern": 6, |
| "layer_types": [ |
| "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" |
| ], |
| "rms_norm_eps": 1e-06, |
| "rope_theta_global": 1000000.0, |
| "rope_theta_local": 10000.0, |
| "embed_dim": 768, |
| "dense_intermediate_dim": 3072, |
| "pooling": "mean", |
| "normalize": "l2", |
| "matryoshka_dims": [ |
| 768, |
| 512, |
| 256, |
| 128 |
| ], |
| "task_prefixes": { |
| "retrieval_query": "task: search result | query: ", |
| "retrieval_document": "title: none | text: ", |
| "classification": "task: classification | query: ", |
| "clustering": "task: clustering | query: ", |
| "similarity": "task: sentence similarity | query: ", |
| "code_retrieval": "task: code retrieval | query: ", |
| "question_answering": "task: question answering | query: ", |
| "fact_verification": "task: fact checking | query: " |
| }, |
| "quantization": "int8", |
| "compute_units": "CPU_AND_NE" |
| } |
