EdwardoSunny's picture
Upload data.js with huggingface_hub
3d06378 verified
Raw
History Blame Contribute Delete
12.2 kB
window.DASH = {
"generated_utc": "2026-07-20 21:49 UTC",
"provenance": "suite numbers recomputed from data/**.jsonl at build time; NESTFUL IBM-scorer metrics + frontier context transcribed from run logs (EXPERIMENT_LOG.md)",
"countdown": {
"qwen": {
"baseline": {
"file": "data/countdown/qwen_unconstrained_n256.jsonl",
"acc": 0.8047,
"n": 256
},
"mask": {
"file": "data/countdown/qwen_mask_n256.jsonl",
"acc": 0.832,
"n": 256,
"flips": {
"lost": 0,
"gained": 7,
"n": 256
}
},
"mask_dfa": {
"file": "data/countdown/qwen_maskdfa_n256.jsonl",
"acc": 0.832,
"n": 256,
"flips": {
"lost": 0,
"gained": 7,
"n": 256
}
},
"hmm_w02": {
"file": "data/countdown/qwen_hmm_w02_n256.jsonl",
"acc": 0.8164,
"n": 256,
"flips": {
"lost": 4,
"gained": 7,
"n": 256
}
},
"mask_bon": {
"file": "data/countdown/qwen_maskbon_n256.jsonl",
"acc": 0.8359,
"n": 256,
"flips": {
"lost": 0,
"gained": 8,
"n": 256
}
},
"free_bon4": {
"file": "data/countdown/qwen_bon4_n256.jsonl",
"acc": 0.832,
"n": 256,
"flips": {
"lost": 1,
"gained": 8,
"n": 256
}
}
},
"llama": {
"baseline": {
"file": "data/countdown/llama_unconstrained_n256.jsonl",
"acc": 0.1328,
"n": 256
},
"mask": {
"file": "data/countdown/llama_mask_n256.jsonl",
"acc": 0.1758,
"n": 256,
"flips": {
"lost": 0,
"gained": 11,
"n": 256
}
}
},
"frontier": {
"claude-opus-4-7": {
"acc": 0.9688,
"n": 256
},
"claude-opus-4-8": {
"acc": 0.9219,
"n": 256
},
"claude-sonnet-5": {
"acc": 0.9453,
"n": 256
},
"gpt-5_6-luna": {
"acc": 0.9414,
"n": 256
},
"gpt-5_6-sol": {
"acc": 0.9805,
"n": 256
},
"gpt-5_6-terra": {
"acc": 0.9609,
"n": 256
}
}
},
"nested_arith": {
"qwen": {
"baseline": {
"file": "data/nested_arith/qwen_unconstrained_n384.jsonl",
"acc": 0.3958,
"n": 384,
"chain": {
"acc": 0.3958,
"n": 384,
"chain_valid": 0.3073,
"faithful": 0.2995
}
},
"mask": {
"file": "data/nested_arith/qwen_mask_n384.jsonl",
"acc": 0.3828,
"n": 384,
"flips": {
"lost": 7,
"gained": 2,
"n": 384
},
"chain": {
"acc": 0.3828,
"n": 384,
"chain_valid": 0.4323,
"faithful": 0.3802
}
},
"mask_dfa": {
"file": "data/nested_arith/qwen_mask_dfa_n384.jsonl",
"acc": 0.3177,
"n": 384,
"flips": {
"lost": 52,
"gained": 22,
"n": 384
}
},
"hmm_w05": {
"file": "data/nested_arith/qwen_hmm_h256_w05_n384.jsonl",
"acc": 0.151,
"n": 384,
"flips": {
"lost": 108,
"gained": 14,
"n": 384
},
"chain": {
"acc": 0.151,
"n": 384,
"chain_valid": 0.9844,
"faithful": 0.151
}
}
},
"llama": {
"baseline": {
"file": "data/nested_arith/llama_unconstrained_n384.jsonl",
"acc": 0.1953,
"n": 384,
"chain": {
"acc": 0.1953,
"n": 384,
"chain_valid": 0.5365,
"faithful": 0.1901
}
},
"mask": {
"file": "data/nested_arith/llama_mask_n384.jsonl",
"acc": 0.2005,
"n": 384,
"flips": {
"lost": 1,
"gained": 3,
"n": 384
},
"chain": {
"acc": 0.2005,
"n": 384,
"chain_valid": 0.7448,
"faithful": 0.2005
}
}
},
"lambda_grid_train": {
"02": {
"acc": 0.3203,
"n": 128,
"chain_valid": 0.9297,
"faithful": 0.3203
},
"05": {
"acc": 0.3359,
"n": 128,
"chain_valid": 0.9844,
"faithful": 0.3359
},
"10": {
"acc": 0.0,
"n": 128,
"chain_valid": 1.0,
"faithful": 0.0
}
},
"frontier": {
"claude-opus-4-7": {
"acc": 0.7448,
"n": 384
},
"claude-opus-4-8": {
"acc": 0.7578,
"n": 384
},
"claude-sonnet-5": {
"acc": 0.5885,
"n": 384
},
"gpt-5_6-luna": {
"acc": 0.7786,
"n": 384
},
"gpt-5_6-sol": {
"acc": 0.9219,
"n": 384
},
"gpt-5_6-terra": {
"acc": 0.8724,
"n": 384
}
}
},
"planetarium": {
"qwen": {
"baseline": {
"file": "data/planetarium/qwen_unconstrained_n256.jsonl",
"acc": 0.3346,
"n": 254
},
"mask": {
"file": "data/planetarium/qwen_mask_n256.jsonl",
"acc": 0.332,
"n": 256,
"flips": {
"lost": 0,
"gained": 0,
"n": 254
}
},
"mask_dfa": {
"file": "data/planetarium/qwen_mask_dfa_n256.jsonl",
"acc": 0.332,
"n": 256,
"flips": {
"lost": 0,
"gained": 0,
"n": 254
}
}
},
"llama": {
"baseline": {
"file": "data/planetarium/llama_unconstrained_n256.jsonl",
"acc": 0.0117,
"n": 256
},
"mask": {
"file": "data/planetarium/llama_mask_n256.jsonl",
"acc": 0.0156,
"n": 256,
"flips": {
"lost": 0,
"gained": 1,
"n": 256
}
}
},
"frontier": {
"claude-opus-4-7": {
"acc": 0.6914,
"n": 256
},
"claude-opus-4-8": {
"acc": 0.707,
"n": 256
},
"claude-sonnet-5": {
"acc": 0.7617,
"n": 256
},
"gpt-5_6-luna": {
"acc": 0.6523,
"n": 256
},
"gpt-5_6-sol": {
"acc": 0.7305,
"n": 256
},
"gpt-5_6-terra": {
"acc": 0.6523,
"n": 256
}
}
},
"racket": {
"qwen": {
"baseline": {
"file": "data/racket/qwen_unconstrained_n161.jsonl",
"acc": 0.2857,
"n": 161
},
"mask": {
"file": "data/racket/qwen_mask_n161.jsonl",
"acc": 0.3602,
"n": 161,
"flips": {
"lost": 0,
"gained": 12,
"n": 161
}
}
},
"llama": {
"baseline": {
"file": "data/racket/llama_unconstrained_n161.jsonl",
"acc": 0.1925,
"n": 161
},
"mask": {
"file": "data/racket/llama_mask_n161.jsonl",
"acc": 0.1988,
"n": 161,
"flips": {
"lost": 2,
"gained": 3,
"n": 161
}
}
},
"frontier": {
"claude-opus-4-7": {
"acc": 0.913,
"n": 161
},
"claude-opus-4-8": {
"acc": 0.9379,
"n": 161
},
"claude-sonnet-5": {
"acc": 0.8447,
"n": 161
},
"gpt-5_6-luna": {
"acc": 0.5901,
"n": 161
},
"gpt-5_6-sol": {
"acc": 0.913,
"n": 161
},
"gpt-5_6-terra": {
"acc": 0.4348,
"n": 161
}
}
},
"bbeh_dyck": {
"qwen": {
"baseline": {
"file": "data/bbeh_dyck/qwen_unconstrained_n200.jsonl",
"acc": 0.205,
"n": 200
},
"mask": {
"file": "data/bbeh_dyck/qwen_mask_n200.jsonl",
"acc": 0.3,
"n": 200,
"flips": {
"lost": 4,
"gained": 23,
"n": 200
}
}
},
"llama": {
"baseline": {
"file": "data/bbeh_dyck/llama_unconstrained_n200.jsonl",
"acc": 0.025,
"n": 200
},
"mask": {
"file": "data/bbeh_dyck/llama_mask_n200.jsonl",
"acc": 0.005,
"n": 200,
"flips": {
"lost": 4,
"gained": 0,
"n": 200
}
}
},
"frontier": {
"claude-opus-4-7": {
"acc": 0.7,
"n": 200
},
"claude-opus-4-8": {
"acc": 0.345,
"n": 200
},
"claude-sonnet-5": {
"acc": 0.225,
"n": 200
},
"gpt-5_6-luna": {
"acc": 0.555,
"n": 200
},
"gpt-5_6-sol": {
"acc": 0.84,
"n": 200
},
"gpt-5_6-terra": {
"acc": 0.725,
"n": 200
}
}
},
"structural": {
"countdown/qwen": {
"n": 256,
"fails": 50,
"parse_reject": 17,
"incomplete": 19,
"semantic": 14
},
"countdown/llama": {
"n": 256,
"fails": 222,
"parse_reject": 64,
"incomplete": 148,
"semantic": 10
},
"nested_arith/qwen": {
"n": 384,
"fails": 232,
"parse_reject": 167,
"incomplete": 62,
"semantic": 3
},
"nested_arith/llama": {
"n": 384,
"fails": 309,
"parse_reject": 178,
"incomplete": 40,
"semantic": 91
},
"planetarium/qwen": {
"n": 254,
"fails": 169,
"parse_reject": 2,
"incomplete": 2,
"semantic": 165
},
"planetarium/llama": {
"n": 256,
"fails": 253,
"parse_reject": 2,
"incomplete": 69,
"semantic": 182
}
},
"dyck_kl": {
"config": {
"quick": false,
"hs": [
2,
4,
8,
16,
32
],
"n_samples": 100000,
"em_iters": 100,
"restarts": 2,
"smooth": 0.001,
"lm_seed": 7,
"sample_seed": 1234,
"em_seed": 1000
},
"language": {
"depth": 3,
"max_len_incl_eos": 12,
"n_accepted": 1347,
"nwa_states": 48,
"lm_mass_on_accepted": 0.023153314044920457
},
"distillation": {
"n_samples": 100000,
"n_unique": 33198,
"n_forced_eos": 17791,
"per_h": {
"2": {
"avg_loglik": -13.994782996264572,
"em_iters_used": 100,
"distill_kl_mc": 4.078227244013312
},
"4": {
"avg_loglik": -13.15911380651663,
"em_iters_used": 100,
"distill_kl_mc": 3.2405201988037593
},
"8": {
"avg_loglik": -12.366249485576574,
"em_iters_used": 100,
"distill_kl_mc": 2.4476052910339057
},
"16": {
"avg_loglik": -11.92665970131934,
"em_iters_used": 100,
"distill_kl_mc": 2.0088123023190465
},
"32": {
"avg_loglik": -11.578717748637416,
"em_iters_used": 100,
"distill_kl_mc": 1.6594528339427308
}
}
},
"table": [
{
"policy": "mask-only",
"h": null,
"kl_fwd": 0.5310182411550299,
"kl_rev": 0.5516616319620444,
"sum_p": 1.0,
"guide_mc_kl": null
},
{
"policy": "ctrlg",
"h": 2,
"kl_fwd": 1.9992174110894245,
"kl_rev": 1.0392534771829336,
"sum_p": 1.0000000000000004,
"guide_mc_kl": 4.078227244013312
},
{
"policy": "ctrlg",
"h": 4,
"kl_fwd": 1.7284096916719445,
"kl_rev": 1.0090787228214624,
"sum_p": 0.9999999999999998,
"guide_mc_kl": 3.2405201988037593
},
{
"policy": "ctrlg",
"h": 8,
"kl_fwd": 1.1733715958351294,
"kl_rev": 0.6541745569739358,
"sum_p": 0.9999999999999996,
"guide_mc_kl": 2.4476052910339057
},
{
"policy": "ctrlg",
"h": 16,
"kl_fwd": 0.8072083795865748,
"kl_rev": 0.47696071923941025,
"sum_p": 0.9999999999999997,
"guide_mc_kl": 2.0088123023190465
},
{
"policy": "ctrlg",
"h": 32,
"kl_fwd": 0.4811768698605567,
"kl_rev": 0.33703712896819793,
"sum_p": 1.0000000000000004,
"guide_mc_kl": 1.6594528339427308
},
{
"policy": "oracle",
"h": null,
"kl_fwd": -4.595798209410908e-16,
"kl_rev": 4.595798209410923e-16,
"sum_p": 1.0000000000000007,
"guide_mc_kl": null
}
],
"runtime_s": 652.6919736862183
},
"nestful": {
"provenance": "IBM vendored scorer output, test-200 (train-100 for the grid); job logs 2026-07-16..19, see EXPERIMENT_LOG.md",
"test200": {
"unconstrained": {
"fsm": 0.145,
"partial": 0.183,
"f1_func": 0.703,
"parse_err": 20,
"valid": null
},
"mask_only": {
"fsm": 0.235,
"partial": 0.251,
"f1_func": null,
"parse_err": 1,
"valid": "~178-196/200"
},
"ctrlg_old_guide": {
"fsm": 0.21,
"partial": 0.271,
"f1_func": 0.931,
"parse_err": 6,
"valid": "194/200"
},
"ctrlg_walker_l1": {
"fsm": 0.125,
"partial": 0.211,
"f1_func": 0.745,
"parse_err": 0,
"valid": "200/200"
},
"ctrlg_walker_l01": {
"fsm": 0.23,
"partial": 0.286,
"f1_func": null,
"parse_err": 0,
"valid": "200/200"
}
},
"lambda_grid_train100": {
"mask_only": {
"fsm": 0.22,
"partial": 0.271,
"valid": 99
},
"0.1": {
"fsm": 0.2,
"partial": 0.266,
"valid": 99
},
"0.2": {
"fsm": 0.2,
"partial": 0.264,
"valid": 99
},
"0.5": {
"fsm": 0.17,
"partial": 0.24,
"valid": 100
},
"1.0": {
"fsm": 0.125,
"partial": 0.211,
"valid": 100
}
},
"frontier_no_reasoning": {
"claude-opus-4-7": 0.36,
"claude-sonnet-5": 0.35,
"claude-opus-4-8": 0.33,
"gpt-5.6-sol": 0.315,
"gpt-5.6-terra": 0.3,
"gpt-5.6-luna": 0.255
}
},
"live": {
"name": "nested_arith test-384 \u00b7 full Ctrl-G h256 \u03bb=0.5",
"done": 384,
"total": 384,
"acc": 0.151
}
};