{ "environment": { "python": "3.12.14", "torch": "2.14.1+cpu", "numpy": "2.3.5" }, "A": { "identity": "DN2 local CPU read-only reevaluation of DN1 cloud checkpoints; no retraining", "selection_rule": "first dynamic greedy failure in stored test order, zero-based index", "test_index": 25, "source": [ "b", "d", "a", "d", "e" ], "target": [ "e", "d", "a", "d", "b", "" ], "greedy": [ "d", "e", "d", "a", "b", "" ], "first_divergence_one_based": 1, "teacher_forced": [ { "step_one_based": 1, "input_prefix": [ "" ], "target": "e", "argmax": "d", "p_target": 0.32724180817604065, "nll": 1.117055892944336 }, { "step_one_based": 2, "input_prefix": [ "", "e" ], "target": "d", "argmax": "d", "p_target": 0.9901096820831299, "nll": 0.009939581155776978 }, { "step_one_based": 3, "input_prefix": [ "", "e", "d" ], "target": "a", "argmax": "a", "p_target": 0.914951741695404, "nll": 0.08888394385576248 }, { "step_one_based": 4, "input_prefix": [ "", "e", "d", "a" ], "target": "d", "argmax": "d", "p_target": 0.9920902252197266, "nll": 0.007941221818327904 }, { "step_one_based": 5, "input_prefix": [ "", "e", "d", "a", "d" ], "target": "b", "argmax": "b", "p_target": 0.9820629358291626, "nll": 0.01809985563158989 }, { "step_one_based": 6, "input_prefix": [ "", "e", "d", "a", "d", "b" ], "target": "", "argmax": "", "p_target": 0.9625928997993469, "nll": 0.038124725222587585 } ], "free_trace": [ { "step_one_based": 1, "input_prefix": [ "" ], "chosen": "d", "p_chosen": 0.6556633710861206, "p_reference_here": 0.32724177837371826, "reference": "e", "source_weights": [ 0.20000000298023224, 0.20000000298023224, 0.20000000298023224, 0.20000000298023224, 0.20000000298023224 ] }, { "step_one_based": 2, "input_prefix": [ "", "d" ], "chosen": "e", "p_chosen": 0.999370276927948, "p_reference_here": 0.00023608427727594972, "reference": "d", "source_weights": [ 0.006220019422471523, 0.04293301701545715, 0.15454530715942383, 0.38063162565231323, 0.41567009687423706 ] }, { "step_one_based": 3, "input_prefix": [ "", "d", "e" ], "chosen": "d", "p_chosen": 0.7937102913856506, "p_reference_here": 0.13423039019107819, "reference": "a", "source_weights": [ 0.04258306324481964, 0.14054341614246368, 0.30694887042045593, 0.36840659379959106, 0.14151804149150848 ] }, { "step_one_based": 4, "input_prefix": [ "", "d", "e", "d" ], "chosen": "a", "p_chosen": 0.9110057353973389, "p_reference_here": 0.001530774519778788, "reference": "d", "source_weights": [ 0.18651393055915833, 0.3696633577346802, 0.3350057899951935, 0.1043139323592186, 0.004503081552684307 ] }, { "step_one_based": 5, "input_prefix": [ "", "d", "e", "d", "a" ], "chosen": "b", "p_chosen": 0.9976014494895935, "p_reference_here": 0.9976014494895935, "reference": "b", "source_weights": [ 0.8021423816680908, 0.15363380312919617, 0.040485356003046036, 0.0036131259985268116, 0.00012530222011264414 ] }, { "step_one_based": 6, "input_prefix": [ "", "d", "e", "d", "a", "b" ], "chosen": "", "p_chosen": 0.8554282188415527, "p_reference_here": 0.8554282188415527, "reference": "", "source_weights": [ 0.948374330997467, 0.047530658543109894, 0.0036404321435838938, 0.00043080587056465447, 2.366264016018249e-05 ] } ], "beam": { "1": { "tokens": [ "d", "e", "d", "a", "b", "" ], "log_probability": -0.9055351097485982, "ended": true, "correct": false, "trace": [ [ { "tokens": [ "d" ], "logp": -0.4221077263355255 } ], [ { "tokens": [ "d", "e" ], "logp": -0.4227376682101749 } ], [ { "tokens": [ "d", "e", "d" ], "logp": -0.6537744504748844 } ], [ { "tokens": [ "d", "e", "d", "a" ], "logp": -0.7469805655418895 } ], [ { "tokens": [ "d", "e", "d", "a", "b" ], "logp": -0.7493820119998418 } ], [ { "tokens": [ "d", "e", "d", "a", "b", "" ], "logp": -0.9055351097485982 } ] ] }, "2": { "tokens": [ "d", "e", "d", "a", "b", "" ], "log_probability": -0.9055351097485982, "ended": true, "correct": false, "trace": [ [ { "tokens": [ "d" ], "logp": -0.4221077263355255 }, { "tokens": [ "e" ], "logp": -1.117055892944336 } ], [ { "tokens": [ "d", "e" ], "logp": -0.4227376682101749 }, { "tokens": [ "e", "d" ], "logp": -1.126995474100113 } ], [ { "tokens": [ "d", "e", "d" ], "logp": -0.6537744504748844 }, { "tokens": [ "e", "d", "a" ], "logp": -1.2158794179558754 } ], [ { "tokens": [ "d", "e", "d", "a" ], "logp": -0.7469805655418895 }, { "tokens": [ "e", "d", "a", "d" ], "logp": -1.2238206397742033 } ], [ { "tokens": [ "d", "e", "d", "a", "b" ], "logp": -0.7493820119998418 }, { "tokens": [ "e", "d", "a", "d", "b" ], "logp": -1.2419204954057932 } ], [ { "tokens": [ "d", "e", "d", "a", "b", "" ], "logp": -0.9055351097485982 }, { "tokens": [ "e", "d", "a", "d", "b", "" ], "logp": -1.2800452206283808 } ] ] }, "4": { "tokens": [ "d", "e", "d", "a", "b", "" ], "log_probability": -0.9055351097485982, "ended": true, "correct": false, "trace": [ [ { "tokens": [ "d" ], "logp": -0.4221077263355255 }, { "tokens": [ "e" ], "logp": -1.117055892944336 }, { "tokens": [ "a" ], "logp": -5.11196756362915 }, { "tokens": [ "b" ], "logp": -5.19541072845459 } ], [ { "tokens": [ "d", "e" ], "logp": -0.4227376682101749 }, { "tokens": [ "e", "d" ], "logp": -1.126995474100113 }, { "tokens": [ "a", "e" ], "logp": -5.125791520811617 }, { "tokens": [ "b", "d" ], "logp": -5.505416691303253 } ], [ { "tokens": [ "d", "e", "d" ], "logp": -0.6537744504748844 }, { "tokens": [ "e", "d", "a" ], "logp": -1.2158794179558754 }, { "tokens": [ "d", "e", "a" ], "logp": -2.430935452634003 }, { "tokens": [ "d", "e", "e" ], "logp": -3.2737570506869815 } ], [ { "tokens": [ "d", "e", "d", "a" ], "logp": -0.7469805655418895 }, { "tokens": [ "e", "d", "a", "d" ], "logp": -1.2238206397742033 }, { "tokens": [ "d", "e", "a", "d" ], "logp": -2.4461979343905114 }, { "tokens": [ "d", "e", "d", "b" ], "logp": -3.1256064874469303 } ], [ { "tokens": [ "d", "e", "d", "a", "b" ], "logp": -0.7493820119998418 }, { "tokens": [ "e", "d", "a", "d", "b" ], "logp": -1.2419204954057932 }, { "tokens": [ "d", "e", "a", "d", "b" ], "logp": -2.8997711911215447 }, { "tokens": [ "d", "e", "d", "b", "a" ], "logp": -3.175014032924082 } ], [ { "tokens": [ "d", "e", "d", "a", "b", "" ], "logp": -0.9055351097485982 }, { "tokens": [ "e", "d", "a", "d", "b", "" ], "logp": -1.2800452206283808 }, { "tokens": [ "d", "e", "d", "b", "a", "" ], "logp": -3.2083430056809448 }, { "tokens": [ "d", "e", "d", "a", "b", "b" ], "logp": -3.2280161310336553 } ], [ { "tokens": [ "d", "e", "d", "a", "b", "" ], "logp": -0.9055351097485982 }, { "tokens": [ "e", "d", "a", "d", "b", "" ], "logp": -1.2800452206283808 }, { "tokens": [ "d", "e", "d", "b", "a", "" ], "logp": -3.2083430056809448 }, { "tokens": [ "d", "e", "d", "a", "b", "b", "" ], "logp": -3.233292219170835 } ] ] } }, "metrics": { "fixed": { "teacher_forced_nll": 0.28789767007979133, "teacher_forced_token_accuracy": 0.9097222222222222, "free_generation_exact_match": 0.5766666666666667, "valid_tokens": 2016 }, "dynamic": { "teacher_forced_nll": 0.10667316591928876, "teacher_forced_token_accuracy": 0.9776785714285714, "free_generation_exact_match": 0.87, "valid_tokens": 2016 } }, "config": { "seed": 20261018, "train": 3000, "validation": 300, "test": 300, "source_lengths": [ 3, 7 ], "embed": 16, "encoder_hidden": 16, "decoder_hidden": 32, "batch": 32, "updates": 500, "checkpoint_interval": 100, "lr": 0.003, "max_seconds_per_model": 600, "source_order": "natural", "target": "reverse(source)+EOS", "initialization": "copy all same-shaped common tensors; separate optimizers", "summary": "masked mean of BiLSTM outputs", "decoder_initial_state": "zeros", "output_score": "sum log p including EOS; PAD/BOS forbidden; no length penalty" }, "checkpoint_sha256": { "fixed": "a09d20398d97c3e68cf4e2c24d54dd4d8476130126d3780d7b44fdb206c24c16", "dynamic": "67ffebe93c8072e23b65a03b3ded8c6eabd18bff5701986f36747842d7634b38" }, "data_sha256": "db6725776262550fb99122455919bbcf8f4a725876a2e58ca6732a7fc5af17c0" }, "B": { "identity": "DN2 local CPU untrained mechanism; all weights random fixed seed 20261016", "prefix": [ "a", "scene" ], "vocab": [ "a", "scene", "river", "person", "bank", "EOS" ], "conditions": [ "A on", "B on", "A off", "B off" ], "probabilities": [ [ 0.28199455882772956, 0.15666509160135741, 0.16143468597624872, 0.1457559679758329, 0.11707047541250226, 0.1370792202063292 ], [ 0.2795800871719861, 0.15695729902048491, 0.16165542859827903, 0.1464786732392372, 0.1170723879114036, 0.13825612405860924 ], [ 0.13819355450039303, 0.20876195279822396, 0.1425203429563665, 0.2034016355764242, 0.14564967209272928, 0.161472842075863 ], [ 0.13819355450039303, 0.20876195279822396, 0.1425203429563665, 0.2034016355764242, 0.14564967209272928, 0.161472842075863 ] ], "max_Z_change": 0.9221700177295178, "max_on_logit_change": 0.0164402921875616, "max_off_logit_change": 0.0, "shapes": { "features": [ 196, 5 ], "Z": [ 196, 8 ], "Q": [ 1, 4 ], "K": [ 196, 4 ], "V": [ 196, 4 ], "weights": [ 1, 196 ], "read": [ 1, 4 ], "branch": [ 8 ], "logits": [ 6 ] }, "images": { "caption.webp": "a4e4f6c3f55328ffa1cec4a9c137a93a30704d704233ece82ab10668bca1a768", "bank.webp": "9702133477d0be9d4870f9b377bc4ff2c4a8c0e2bda79b8b012df602091e4d30" } }, "C": { "identity": "DN2 untrained TinyDecoder mechanism on explicit teaching character encoding, NOT document QA", "document": "09:00 林岑把备用钥匙放在蓝柜；取件人是周禾。\n09:10 陈默说红柜空着；这句话只说明红柜，没有改动钥匙。\n09:20 林岑通知：把那把备用钥匙改放红柜。\n09:30 周禾回复：我还没取；“那把”指备用钥匙。\n09:40 林岑撤回09:20的改放通知：它仍在蓝柜，取件人不变。\n问题：现在周禾应去哪个柜子取备用钥匙？", "moved_document": "09:40 林岑撤回09:20的改放通知：它仍在蓝柜，取件人不变。\n09:00 林岑把备用钥匙放在蓝柜；取件人是周禾。\n09:10 陈默说红柜空着；这句话只说明红柜，没有改动钥匙。\n09:20 林岑通知：把那把备用钥匙改放红柜。\n09:30 周禾回复：我还没取；“那把”指备用钥匙。\n问题：现在周禾应去哪个柜子取备用钥匙？", "question": "现在周禾应去哪个柜子取备用钥匙？", "human_rule_answer": "蓝柜", "encoding": "one Unicode code point -> one sorted-vocabulary ID; whitespace/newline/punctuation count; no BOS/template added", "vocab": [ "\n", " ", "0", "1", "2", "3", "4", "9", ":", "“", "”", "。", "不", "个", "人", "仍", "件", "动", "匙", "去", "取", "变", "句", "只", "周", "哪", "回", "在", "备", "复", "子", "它", "岑", "应", "我", "把", "指", "撤", "改", "放", "明", "是", "有", "林", "柜", "没", "现", "用", "的", "着", "知", "禾", "空", "红", "绿", "蓝", "话", "说", "还", "这", "通", "那", "钥", "问", "陈", "题", "默", "，", "：", "；", "？" ], "token_ids": [ 2, 7, 8, 2, 2, 1, 43, 32, 35, 28, 47, 62, 18, 39, 27, 55, 44, 69, 20, 16, 14, 41, 24, 51, 11, 0, 2, 7, 8, 3, 2, 1, 64, 66, 57, 53, 44, 52, 49, 69, 59, 22, 56, 23, 57, 40, 53, 44, 67, 45, 42, 38, 17, 62, 18, 11, 0, 2, 7, 8, 4, 2, 1, 43, 32, 60, 50, 68, 35, 61, 35, 28, 47, 62, 18, 38, 39, 53, 44, 11, 0, 2, 7, 8, 5, 2, 1, 24, 51, 26, 29, 68, 34, 58, 45, 20, 69, 9, 61, 35, 10, 36, 28, 47, 62, 18, 11, 0, 2, 7, 8, 6, 2, 1, 43, 32, 37, 26, 2, 7, 8, 4, 2, 48, 38, 39, 60, 50, 68, 31, 15, 27, 55, 44, 67, 20, 16, 14, 12, 21, 11, 0, 63, 65, 68, 46, 27, 24, 51, 33, 19, 25, 13, 44, 30, 20, 28, 47, 62, 18, 70 ], "token_count": 161, "vocab_size": 71, "prefill_tokens": 141, "spans": [ { "event": "09:00", "start_zero_based": 0, "end_exclusive": 25 }, { "event": "09:10", "start_zero_based": 26, "end_exclusive": 56 }, { "event": "09:20", "start_zero_based": 57, "end_exclusive": 80 }, { "event": "09:30", "start_zero_based": 81, "end_exclusive": 107 }, { "event": "09:40", "start_zero_based": 108, "end_exclusive": 141 } ], "moved_spans": [ { "event": "09:00", "start_zero_based": 34, "end_exclusive": 59 }, { "event": "09:10", "start_zero_based": 60, "end_exclusive": 90 }, { "event": "09:20", "start_zero_based": 91, "end_exclusive": 114 }, { "event": "09:30", "start_zero_based": 115, "end_exclusive": 141 }, { "event": "09:40", "start_zero_based": 0, "end_exclusive": 33 } ], "moved_token_count": 161, "full_incremental_max_error": 1.8041124150158794e-15, "changed_prefix_stale_cache_max_error": 0.015632048194583525, "cached_bytes": 41216, "model": { "layers": 2, "heads": 2, "width": 8, "dtype": "float64", "trained": false, "seed": 20261021 }, "task_prediction": "not evaluated; human answer is not a model prediction" } }