diffusion_baseline / results /summary /rdlm_diagnostics.json
jyliuAI's picture
Add files using upload-large-folder tool
dc72977 verified
Raw History Blame Contribute Delete
8.56 kB
{
"dataset": "lm1b",
"expected_sample_count": 1024,
"model": "rdlm",
"root": "/home/ruizzhao/diffusion_baseline",
"rows": [
{
"cls_text_fraction": 0.0380859375,
"command_batch_per_gpu": "8",
"command_sampling_steps": "1",
"dataset": "lm1b",
"generative_perplexity": 22.702928392437855,
"max_token_length": 128,
"mean_text_chars": 818.8828125,
"mean_token_length": 128,
"median_text_chars": 833.5,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_1/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.01833343505859375,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_1",
"self_bleu": 0.2611895581130025,
"sep_text_fraction": 0.0078125,
"steps": 1,
"token_ids_reason": "upstream_ids_outside_canonical_vocab",
"token_ids_source": "retokenized",
"unigram_entropy": 3.2009783661618685,
"unique_text_count": 1024,
"unique_text_ratio": 1.0,
"warnings": [
"low_ppl_with_low_entropy",
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.14453125,
"command_batch_per_gpu": "8",
"command_sampling_steps": "2",
"dataset": "lm1b",
"generative_perplexity": 43.6756504502907,
"max_token_length": 128,
"mean_text_chars": 819.6103515625,
"mean_token_length": 128,
"median_text_chars": 873.0,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_2/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.050537109375,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_2",
"self_bleu": 0.1562727194071143,
"sep_text_fraction": 0.015625,
"steps": 2,
"token_ids_reason": "upstream_ids_outside_canonical_vocab",
"token_ids_source": "retokenized",
"unigram_entropy": 3.5428684108815003,
"unique_text_count": 1024,
"unique_text_ratio": 1.0,
"warnings": [
"low_ppl_with_low_entropy",
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.3125,
"command_batch_per_gpu": "8",
"command_sampling_steps": "4",
"dataset": "lm1b",
"generative_perplexity": 79.02632500884272,
"max_token_length": 128,
"mean_text_chars": 758.2734375,
"mean_token_length": 128,
"median_text_chars": 843.0,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_4/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.06224822998046875,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_4",
"self_bleu": 0.11636916742993436,
"sep_text_fraction": 0.01171875,
"steps": 4,
"token_ids_reason": "upstream_ids_not_canonical_length",
"token_ids_source": "retokenized",
"unigram_entropy": 3.7381412992665806,
"unique_text_count": 1002,
"unique_text_ratio": 0.978515625,
"warnings": [
"low_ppl_with_low_entropy",
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.7333984375,
"command_batch_per_gpu": "8",
"command_sampling_steps": "8",
"dataset": "lm1b",
"generative_perplexity": 229.3208427701825,
"max_token_length": 128,
"mean_text_chars": 650.83203125,
"mean_token_length": 128,
"median_text_chars": 636.0,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_8/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.0711212158203125,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_8",
"self_bleu": 0.060051441962376335,
"sep_text_fraction": 0.00390625,
"steps": 8,
"token_ids_reason": "upstream_ids_not_canonical_length",
"token_ids_source": "retokenized",
"unigram_entropy": 4.098309110124398,
"unique_text_count": 1006,
"unique_text_ratio": 0.982421875,
"warnings": [
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.9462890625,
"command_batch_per_gpu": "8",
"command_sampling_steps": "16",
"dataset": "lm1b",
"generative_perplexity": 406.05692318214335,
"max_token_length": 128,
"mean_text_chars": 608.2490234375,
"mean_token_length": 128,
"median_text_chars": 622.0,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_16/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.05516815185546875,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_16",
"self_bleu": 0.057739472763989,
"sep_text_fraction": 0.0,
"steps": 16,
"token_ids_reason": "upstream_ids_not_canonical_length",
"token_ids_source": "retokenized",
"unigram_entropy": 4.289774377120292,
"unique_text_count": 1017,
"unique_text_ratio": 0.9931640625,
"warnings": [
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.98828125,
"command_batch_per_gpu": "8",
"command_sampling_steps": "32",
"dataset": "lm1b",
"generative_perplexity": 375.9462579853705,
"max_token_length": 128,
"mean_text_chars": 612.744140625,
"mean_token_length": 128,
"median_text_chars": 623.5,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_32/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.03092193603515625,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_32",
"self_bleu": 0.0706832023382317,
"sep_text_fraction": 0.0,
"steps": 32,
"token_ids_reason": "upstream_ids_not_canonical_length",
"token_ids_source": "retokenized",
"unigram_entropy": 4.333617707089788,
"unique_text_count": 1021,
"unique_text_ratio": 0.9970703125,
"warnings": [
"retokenized_from_text"
]
},
{
"cls_text_fraction": 0.9990234375,
"command_batch_per_gpu": "8",
"command_sampling_steps": "1024",
"dataset": "lm1b",
"generative_perplexity": 274.9406258182183,
"max_token_length": 128,
"mean_text_chars": 620.728515625,
"mean_token_length": 128,
"median_text_chars": 623.0,
"metrics_partial": false,
"metrics_path": "/home/ruizzhao/diffusion_baseline/results/metrics/lm1b/rdlm/steps_1024/metrics.json",
"metrics_sample_count": 1024,
"min_token_length": 128,
"model": "rdlm",
"pad_token_fraction": 0.00969696044921875,
"run_status": "succeeded",
"sample_count": 1024,
"sample_dir": "/home/ruizzhao/diffusion_baseline/results/samples/lm1b/rdlm/steps_1024",
"self_bleu": 0.0945744656424746,
"sep_text_fraction": 0.0,
"steps": 1024,
"token_ids_reason": "upstream_ids_not_canonical_length",
"token_ids_source": "retokenized",
"unigram_entropy": 4.346245654267354,
"unique_text_count": 1024,
"unique_text_ratio": 1.0,
"warnings": [
"retokenized_from_text"
]
}
],
"schema": "dlb-rdlm-diagnostics-v1",
"summary": {
"missing_steps": [],
"ppl_curve_nonmonotonic": true,
"retokenized_steps": [
1,
2,
4,
8,
16,
32,
1024
],
"suspect_steps": [
1,
2,
4
],
"verdict": "low_ppl_points_show_collapse_signals"
}
}