w
gcw_uQ09W7jl/deepcpgdna-hou2016-hepg2
Model Introduction
Files & versions
Pull Requests
Discussion
Analysis

multimolecule/deepcpgdna-hou2016-hepg2 on Ascend NPU

模型与任务

  • 模型:multimolecule/deepcpgdna-hou2016-hepg2
  • Revision:6671f9ba39a3e3e2604986e691aabde45b2953c4
  • 适配路线:regulatory-sequence
  • 设备:npu:0

环境依赖

{
  "torch": "2.9.0+cpu",
  "torch_npu_available": true,
  "device_name": "Ascend910_9362",
  "device_count": 2,
  "torch_npu": "2.9.0.post1+gitee7ba04",
  "multimolecule": "unknown",
  "transformers": "5.9.0"
}

安装:pip install -r requirements.txt

NPU 推理

默认执行:python inference.py

真实输入

{
  "num_sequences": 6,
  "sequence_length_fixed": 1001,
  "alphabet": "ACGT (N zero channel)",
  "replace_U_with_T": true,
  "center_CpG": "positions 500-501 = CG (0-indexed)",
  "strand": "5'->3' (no RC ensemble)",
  "sequences": [
    {
      "name": "repeat_ACGT_CGcenter",
      "length": 1001,
      "sha256": "9f25ecf7a6019cd1e958851fba282cec33d596b4fdade382435eee7d3290bd9d",
      "preview": "ACGTACGTACGTACGTACGTACGTACGTAC",
      "center": "GTCGGT"
    },
    {
      "name": "lowGC_0.35_CGcenter",
      "length": 1001,
      "sha256": "12d58fdbfef21a857e8514a7804dfee89371172cf306d5e9d98745223567aa46",
      "preview": "GCTAGTATAAAGTTATTGTTACCTACTATA",
      "center": "GTCGGA"
    },
    {
      "name": "midGC_0.50_CGcenter",
      "length": 1001,
      "sha256": "cd884ebaf4561c2bc72c68f143f30f1c8b75d58b03410dad5362d7a9d4d5c6bf",
      "preview": "ACTGATTTATCTAAGGTATTTCCTACTTTC",
      "center": "CGCGAG"
    },
    {
      "name": "highGC_0.65_CGcenter",
      "length": 1001,
      "sha256": "29533479a6f90187bde152bd429968a9516bef7cae0133c1fa0a1dd3afb78f0c",
      "preview": "GCGCGCCGGTAGGTCTCAACCTCCCAATAA",
      "center": "ATCGAG"
    },
    {
      "name": "TP53_window_CGcenter",
      "length": 1001,
      "sha256": "28bcfa33565700daf17e6b2474b047619d655e9adb9ba736b5cfcf40a50bd034",
      "preview": "ACGTACGTACGTACGTACGTACGTACGTAC",
      "center": "GCCGTG"
    },
    {
      "name": "N_masked_control",
      "length": 1001,
      "sha256": "f88ad847adc50ff19f4c5d2440b53d076fcf21d76956d06cfa576de36f9ec5f6",
      "preview": "ACANTANNNTACGCGTGAATGGACAATGGG",
      "center": "ATCGGC"
    }
  ],
  "tokenizer": "DnaTokenizer vocab A0 C1 G2 T3 N4, pad N, model_max_length 1001",
  "output_cells": 6,
  "cell_labels": [
    "HepG21",
    "HepG22",
    "HepG23",
    "HepG24",
    "HepG25",
    "HepG26"
  ]
}

真实推理结果

{
  "output_shape": [
    6,
    6
  ],
  "dtype": "float32",
  "logits_per_sequence": [
    {
      "name": "repeat_ACGT_CGcenter",
      "logits": [
        -8.750944137573242,
        -8.705138206481934,
        -9.55453109741211,
        -9.112507820129395,
        -8.840112686157227,
        -8.944999694824219
      ],
      "probs": [
        0.00015828672621864825,
        0.00016570459411013871,
        7.087435369612649e-05,
        0.00011026564607163891,
        0.0001447854592697695,
        0.00013037050666753203
      ],
      "logits_mean": -8.984705924987793,
      "probs_mean": 0.0001300478761550039
    },
    {
      "name": "lowGC_0.35_CGcenter",
      "logits": [
        0.1754152774810791,
        0.11300782859325409,
        0.18009355664253235,
        0.013127422891557217,
        0.11527351289987564,
        0.08715878427028656
      ],
      "probs": [
        0.5437417030334473,
        0.5282219052314758,
        0.5449020862579346,
        0.503281831741333,
        0.5287865400314331,
        0.5217759013175964
      ],
      "logits_mean": 0.11401274055242538,
      "probs_mean": 0.5284516215324402
    },
    {
      "name": "midGC_0.50_CGcenter",
      "logits": [
        -1.7769941091537476,
        -1.583351492881775,
        -1.7161356210708618,
        -1.7162753343582153,
        -1.647099256515503,
        -1.707887053489685
      ],
      "probs": [
        0.14467468857765198,
        0.17032136023044586,
        0.15236958861351013,
        0.15235154330730438,
        0.16150137782096863,
        0.15343797206878662
      ],
      "logits_mean": -1.6912904977798462,
      "probs_mean": 0.15577609837055206
    },
    {
      "name": "highGC_0.65_CGcenter",
      "logits": [
        -1.8742077350616455,
        -2.008403778076172,
        -1.859586477279663,
        -2.0113961696624756,
        -1.8843387365341187,
        -1.9702095985412598
      ],
      "probs": [
        0.13305559754371643,
        0.11832339316606522,
        0.13475126028060913,
        0.11801157146692276,
        0.13189131021499634,
        0.12236636877059937
      ],
      "logits_mean": -1.9346904754638672,
      "probs_mean": 0.1263999193906784
    },
    {
      "name": "TP53_window_CGcenter",
      "logits": [
        -6.454075813293457,
        -6.446222305297852,
        -6.965205669403076,
        -6.654971122741699,
        -6.415478229522705,
        -6.598630428314209
      ],
      "probs": [
        0.0015716195339336991,
        0.0015839912230148911,
        0.0009432781371288002,
        0.0012859494891017675,
        0.001633365172892809,
        0.0013603792758658528
      ],
      "logits_mean": -6.589097499847412,
      "probs_mean": 0.0013964303070679307
    },
    {
      "name": "N_masked_control",
      "logits": [
        -0.47227951884269714,
        -0.42052558064460754,
        -0.49105775356292725,
        -0.6112858057022095,
        -0.4110545814037323,
        -0.5385561585426331
      ],
      "probs": [
        0.384076863527298,
        0.39639097452163696,
        0.3796444237232208,
        0.3517659604549408,
        0.39865928888320923,
        0.368523508310318
      ],
      "logits_mean": -0.49079322814941406,
      "probs_mean": 0.37984347343444824
    }
  ],
  "logits_full_first": [
    -8.750944137573242,
    -8.705138206481934,
    -9.55453109741211,
    -9.112507820129395,
    -8.840112686157227,
    -8.944999694824219
  ],
  "probs_full_first": [
    0.00015828672621864825,
    0.00016570459411013871,
    7.087435369612649e-05,
    0.00011026564607163891,
    0.0001447854592697695,
    0.00013037050666753203
  ],
  "device_proof": "first_param:npu:0, logits:npu:0, input:npu:0"
}

CPU-NPU 一致性

{
  "atol": 0.01,
  "rtol": 0.01,
  "max_abs_diff": 0.000431060791015625,
  "mean_abs_diff": 0.00016596207084755102,
  "max_abs_diff_probs": 2.6851892471313477e-05,
  "mean_abs_diff_probs": 8.184304533642717e-06,
  "passed": true,
  "pearson_logits": 0.9999999997909007,
  "pearson_probs": 0.9999999991848292,
  "shape": [
    6,
    6
  ]
}

性能

{
  "first_run_ms": 1.3310527428984642,
  "second_run_ms": 1.2416010722517967,
  "avg_ms": 1.2192672584205866,
  "min_ms": 1.192430965602398,
  "max_ms": 1.2541920877993107,
  "p50_ms": 1.2142062187194824,
  "p90_ms": 1.2428601738065481,
  "throughput_seq_per_s": 4920.988371140446,
  "throughput_bases_per_s": 4925909.359511587,
  "batch_size": 6,
  "sequence_length": 1001,
  "dtype": "float32",
  "peak_memory_mb": 46.533203125
}

自验证截图

Agent workflow

NPU device call

Model result

三张图片均由 xterm.js 根据本次真实日志生成。

已知限制

  • 仅对本次记录的模型 revision、输入契约、dtype 与 Ascend 环境完成验证。

标签

#NPU #Ascend #Ascend910