| { | |
| "soft_cap_gib": 2.96, | |
| "fontes": { | |
| "gpt2a": { | |
| "repo": "onnx-community/gpt2-alpaca-gpt4-ONNX", | |
| "kind": "text-instruction", | |
| "d_src": 768, | |
| "K": 32, | |
| "K_xi": 8, | |
| "adapters": 4, | |
| "notas": "GPT-2 124M afinado em Alpaca-GPT4 (epoca <= 2023/24)", | |
| "download_bytes": 252785964, | |
| "download_h": [ | |
| 241.1, | |
| "MiB" | |
| ], | |
| "banco_decomposto_estimado_mib": 3.06 | |
| }, | |
| "gpt2g": { | |
| "repo": "onnx-community/gpt2-ONNX-GQA", | |
| "kind": "text-gqa", | |
| "d_src": 768, | |
| "K": 32, | |
| "K_xi": 8, | |
| "adapters": 4, | |
| "notas": "GPT-2 com atencao agrupada (GQA) exportada ONNX/OGA", | |
| "download_bytes": 501458554, | |
| "download_h": [ | |
| 478.2, | |
| "MiB" | |
| ], | |
| "banco_decomposto_estimado_mib": 3.06 | |
| }, | |
| "dinov2": { | |
| "repo": "onnx-community/dinov2-small-with-attentions", | |
| "kind": "vision", | |
| "d_src": 384, | |
| "K": 48, | |
| "K_xi": 8, | |
| "adapters": 4, | |
| "notas": "ViT-S/14 DINOv2 (autoatencao auto-supervisionada, 2023/24)", | |
| "download_bytes": 46029900, | |
| "download_h": [ | |
| 43.9, | |
| "MiB" | |
| ], | |
| "banco_decomposto_estimado_mib": 1.54 | |
| } | |
| }, | |
| "veredito": { | |
| "cabem_juntos": true, | |
| "n_fontes": 3, | |
| "pico_ram_estimado_gib": 1.1, | |
| "justificativa": "runtime usa memmap (page cache evictavel) + bancos int4 (MBs); decomposicao um-por-vez (P-SRC)" | |
| } | |
| } |