Download 4b/decision.json from ollaya-dev/arbiter: direct link, hf CLI and curl.
- Browser
- Download file 3.01 kB
-
https://huggingface.co/ollaya-dev/arbiter/resolve/main/4b/decision.json
- Command line
-
hf download hf://ollaya-dev/arbiter/4b/decision.json
-
curl -L -o decision.json https://huggingface.co/ollaya-dev/arbiter/resolve/main/4b/decision.json
3.01 kB
| { | |
| "engine": "onnx", | |
| "family": "arbiter", | |
| "layout": "arbiter-fixed-v1", | |
| "upstream": { | |
| "repo": "hiteshluke/arbiter-4b", | |
| "revision": "0c44271c59f89758e3cae17b032e98a9140093e9", | |
| "base": "unsloth/gemma-3-4b-it", | |
| "base_revision": "bf46152c47f5dd20b896357cb51abc4c03b8ee8c", | |
| "code": { | |
| "repo": "https://github.com/CodekinsTech/arbiter", | |
| "commit": "e1cb30fd3019cf91d2631a2d1004abd858b8e4fa", | |
| "script": "training_v3_6/train.py" | |
| }, | |
| "lora": { | |
| "r": 16, | |
| "alpha": 32, | |
| "scaling": 2.0, | |
| "merged": false | |
| } | |
| }, | |
| "contract": { | |
| "inputs": { | |
| "input_ids": { | |
| "dtype": "int64", | |
| "shape": [ | |
| "rows", | |
| "seq" | |
| ], | |
| "note": "one row per question; seq a multiple of 64; right-pad with any id (pad)" | |
| }, | |
| "last_pos": { | |
| "dtype": "int64", | |
| "shape": [ | |
| "rows" | |
| ], | |
| "note": "position of the row's last token" | |
| } | |
| }, | |
| "outputs": { | |
| "scores": { | |
| "dtype": "float32", | |
| "shape": [ | |
| "rows", | |
| 24 | |
| ], | |
| "note": "raw scores of the head's 24 slots; a question's option logits are the scores at its slots" | |
| } | |
| }, | |
| "seq_multiple": 64, | |
| "positions": "0..seq-1, implicit", | |
| "attention": "causal (sliding layers: the last 1024 positions); no mask input" | |
| }, | |
| "max_row_tokens": 8192, | |
| "special_tokens": { | |
| "bos": 2, | |
| "pad": 0 | |
| }, | |
| "num_slots": 24, | |
| "slots": { | |
| "noul": [ | |
| 1, | |
| 0 | |
| ], | |
| "choice": [ | |
| 2, | |
| 3, | |
| 4, | |
| 5, | |
| 6, | |
| 7, | |
| 8, | |
| 9, | |
| 10, | |
| 11, | |
| 12, | |
| 13, | |
| 14, | |
| 15, | |
| 16, | |
| 17 | |
| ], | |
| "score": [ | |
| 18, | |
| 19, | |
| 20, | |
| 21, | |
| 22, | |
| 23 | |
| ] | |
| }, | |
| "max_options": 16, | |
| "score_levels": 6, | |
| "verbalizers": [ | |
| "T", | |
| "F", | |
| "A", | |
| "B", | |
| "C", | |
| "D", | |
| "E", | |
| "F", | |
| "G", | |
| "H", | |
| "I", | |
| "J", | |
| "K", | |
| "L", | |
| "M", | |
| "N", | |
| "O", | |
| "P", | |
| "0", | |
| "1", | |
| "2", | |
| "3", | |
| "4", | |
| "5" | |
| ], | |
| "templates": { | |
| "row": "[bos] tok(\"State: {render(state)}\\n\\nQuestion: {render(instructions)}\\n\\nOptions:\\n{block}\\n\\nAnswer:\"); the head is read at the last token", | |
| "noul_block": "T. Yes / True\nF. No / False", | |
| "choice_block": "{letter}. {option}, one line per option, letters A..P", | |
| "choice_option": "{name} | {name}: {render(description)}", | |
| "score_block": "0\n1\n2\n3\n4\n5", | |
| "render": "None->'', scalars->Python str(), list->'- item' lines, dict->'key: value' lines, 2-space nesting", | |
| "add_special_tokens": false | |
| }, | |
| "option_logits": { | |
| "noul": "scores[row, [1, 0]] (false = F, true = T)", | |
| "choice": "scores[row, 2 + j] for option j < 16", | |
| "score": "scores[row, 18 + level] for level < 6" | |
| }, | |
| "opset": 20, | |
| "precision": "fp32 compute; base weights and head BF16, widened by Cast (at load, or per forward pass with weights_in_memory bf16); adapter F32", | |
| "weights_in_memory": "bf16" | |
| } | |