Download system_graph.json from larql/gemma-3-4b-it: direct link, hf CLI and curl.
- Browser
- Download file 20.1 kB
-
https://huggingface.co/larql/gemma-3-4b-it/resolve/main/system_graph.json
- Command line
-
hf download hf://larql/gemma-3-4b-it/system_graph.json
-
curl -L -o system_graph.json https://huggingface.co/larql/gemma-3-4b-it/resolve/main/system_graph.json
20.1 kB
| { | |
| "schema": 6, | |
| "components": [ | |
| { | |
| "id": "target", | |
| "role": "primary_text", | |
| "source_artifact": "gemma-3-4b-it", | |
| "num_layers": 34, | |
| "hidden_size": 2560, | |
| "attention": [ | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "linear", | |
| "theta": 1000000.0, | |
| "factor": 8.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "linear", | |
| "theta": 1000000.0, | |
| "factor": 8.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "linear", | |
| "theta": 1000000.0, | |
| "factor": 8.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "linear", | |
| "theta": 1000000.0, | |
| "factor": 8.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "linear", | |
| "theta": 1000000.0, | |
| "factor": 8.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "sliding", | |
| "window": 1024, | |
| "position": { | |
| "kind": "rope", | |
| "theta": 10000.0 | |
| }, | |
| "geometry": { | |
| "head_dim": 256, | |
| "num_kv_heads": 4 | |
| } | |
| } | |
| ], | |
| "execution": { | |
| "attention": { | |
| "num_q_heads": 8, | |
| "num_kv_heads": 4, | |
| "head_dim": 256, | |
| "score_scale": 0.0625, | |
| "qk_norm_scope": "per_head", | |
| "qk_norm_weight_offset": 1.0, | |
| "parameter_free_qk_norm": { | |
| "q": false, | |
| "k": false, | |
| "v": false | |
| } | |
| }, | |
| "ffn": { | |
| "intermediate_size": 10240, | |
| "activation": "gelu_tanh", | |
| "ffn_type": "gated", | |
| "gate_policy": { | |
| "kind": "gated" | |
| } | |
| }, | |
| "norm": { | |
| "pre": { | |
| "kind": "rms_norm", | |
| "eps": 9.999999974752427e-7, | |
| "weight_offset": 1.0 | |
| }, | |
| "post": { | |
| "kind": "rms_norm", | |
| "eps": 9.999999974752427e-7, | |
| "weight_offset": 1.0 | |
| }, | |
| "final_norm": { | |
| "kind": "rms_norm", | |
| "eps": 9.999999974752427e-7, | |
| "weight_offset": 1.0 | |
| }, | |
| "placement": "pre_post" | |
| }, | |
| "head": { | |
| "vocab_size": 262208, | |
| "embed_scale": 50.596443, | |
| "head_reuses_embedding": true | |
| }, | |
| "residual_topology": { | |
| "kind": "single_stream" | |
| } | |
| } | |
| }, | |
| { | |
| "id": "vision", | |
| "role": "perception", | |
| "source_artifact": "gemma-3-4b-it", | |
| "num_layers": 27, | |
| "hidden_size": 1152, | |
| "attention": [ | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| }, | |
| { | |
| "operator": "softmax", | |
| "span": "full", | |
| "position": { | |
| "kind": "none" | |
| }, | |
| "declared_span": "full_attention" | |
| } | |
| ], | |
| "perception": { | |
| "modality": "image", | |
| "transform": { | |
| "encoder": { | |
| "depth": 27, | |
| "width": 1152, | |
| "num_heads": 16 | |
| } | |
| } | |
| } | |
| } | |
| ], | |
| "objects": [ | |
| { | |
| "id": "target.decoder_stack", | |
| "component": "target", | |
| "kind": "decoder_stack", | |
| "source_bindings": [ | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "language_model.model.layers", | |
| "tensors": 442, | |
| "bytes": 6418016256 | |
| } | |
| ], | |
| "representations": [ | |
| { | |
| "encoding": "BF16", | |
| "fidelity": "canonical" | |
| } | |
| ] | |
| }, | |
| { | |
| "id": "target.embedding", | |
| "component": "target", | |
| "kind": "embedding", | |
| "source_bindings": [ | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "language_model.model.embed_tokens", | |
| "tensors": 1, | |
| "bytes": 1342504960 | |
| } | |
| ], | |
| "representations": [ | |
| { | |
| "encoding": "BF16", | |
| "fidelity": "canonical" | |
| } | |
| ] | |
| }, | |
| { | |
| "id": "target.final_norm", | |
| "component": "target", | |
| "kind": "final_norm", | |
| "source_bindings": [ | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "language_model.model.norm", | |
| "tensors": 1, | |
| "bytes": 5120 | |
| } | |
| ], | |
| "representations": [ | |
| { | |
| "encoding": "BF16", | |
| "fidelity": "canonical" | |
| } | |
| ] | |
| }, | |
| { | |
| "id": "vision.perception_adapter", | |
| "component": "vision", | |
| "kind": "perception_adapter", | |
| "source_bindings": [ | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "multi_modal_projector", | |
| "tensors": 1, | |
| "bytes": 5898240 | |
| }, | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "multi_modal_projector.mm_soft_emb_norm", | |
| "tensors": 1, | |
| "bytes": 2304 | |
| } | |
| ], | |
| "representations": [ | |
| { | |
| "encoding": "BF16", | |
| "fidelity": "canonical" | |
| } | |
| ] | |
| }, | |
| { | |
| "id": "vision.perception_tower", | |
| "component": "vision", | |
| "kind": "perception_tower", | |
| "source_bindings": [ | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "vision_tower.vision_model.embeddings.patch_embedding", | |
| "tensors": 2, | |
| "bytes": 1357056 | |
| }, | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "vision_tower.vision_model.embeddings.position_embedding", | |
| "tensors": 1, | |
| "bytes": 9437184 | |
| }, | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "vision_tower.vision_model.encoder.layers", | |
| "tensors": 432, | |
| "bytes": 822933216 | |
| }, | |
| { | |
| "artifact": "gemma-3-4b-it", | |
| "tensor_prefix": "vision_tower.vision_model.post_layernorm", | |
| "tensors": 2, | |
| "bytes": 4608 | |
| } | |
| ], | |
| "representations": [ | |
| { | |
| "encoding": "BF16", | |
| "fidelity": "canonical" | |
| } | |
| ] | |
| } | |
| ], | |
| "edges": [] | |
| } |