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

multimolecule/deepcpgdna-hou2016-mesc on Ascend NPU

模型与任务

  • 模型:multimolecule/deepcpgdna-hou2016-mesc
  • Revision:a21c2965ee47880c49d923766bbfdc657a707c64
  • 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": [
    "mESC1",
    "mESC2",
    "mESC3",
    "mESC4",
    "mESC5",
    "mESC6"
  ]
}

真实推理结果

{
  "output_shape": [
    6,
    6
  ],
  "dtype": "float32",
  "logits_per_sequence": [
    {
      "name": "repeat_ACGT_CGcenter",
      "logits": [
        -4.198724746704102,
        -4.558064937591553,
        -4.576184272766113,
        -4.023329734802246,
        -4.5284199714660645,
        -4.295491695404053
      ],
      "probs": [
        0.014792606234550476,
        0.010373583994805813,
        0.010189211927354336,
        0.017578743398189545,
        0.010682377964258194,
        0.013446593657135963
      ],
      "logits_mean": -4.363369464874268,
      "probs_mean": 0.012843853794038296
    },
    {
      "name": "lowGC_0.35_CGcenter",
      "logits": [
        1.4702013731002808,
        1.4887793064117432,
        1.5990163087844849,
        1.4255402088165283,
        2.195486068725586,
        1.481890320777893
      ],
      "probs": [
        0.8130879998207092,
        0.8158949613571167,
        0.8318808078765869,
        0.8062054514884949,
        0.8998434543609619,
        0.8148579597473145
      ],
      "logits_mean": 1.610152244567871,
      "probs_mean": 0.8302950263023376
    },
    {
      "name": "midGC_0.50_CGcenter",
      "logits": [
        -4.02754020690918,
        -3.943082094192505,
        -4.315902233123779,
        -4.10427713394165,
        -4.176808834075928,
        -3.9217233657836914
      ],
      "probs": [
        0.017506178468465805,
        0.019019605591893196,
        0.013178502209484577,
        0.016234049573540688,
        0.015115423128008842,
        0.01942223682999611
      ],
      "logits_mean": -4.0815558433532715,
      "probs_mean": 0.016745999455451965
    },
    {
      "name": "highGC_0.65_CGcenter",
      "logits": [
        -6.876260757446289,
        -6.39695405960083,
        -6.99578857421875,
        -7.003179550170898,
        -6.25938081741333,
        -6.5548906326293945
      ],
      "probs": [
        0.0010309318313375115,
        0.0016638529486954212,
        0.0009148925310000777,
        0.0009081616881303489,
        0.0019087791442871094,
        0.0014211160596460104
      ],
      "logits_mean": -6.6810760498046875,
      "probs_mean": 0.0013079555938020349
    },
    {
      "name": "TP53_window_CGcenter",
      "logits": [
        -4.11770486831665,
        -4.3633270263671875,
        -4.511991024017334,
        -3.9763240814208984,
        -4.530969619750977,
        -4.146795749664307
      ],
      "probs": [
        0.016020989045500755,
        0.012575780041515827,
        0.010857406072318554,
        0.01840919815003872,
        0.01065546553581953,
        0.015568790957331657
      ],
      "logits_mean": -4.2745184898376465,
      "probs_mean": 0.014014605432748795
    },
    {
      "name": "N_masked_control",
      "logits": [
        0.08321520686149597,
        -0.4281061291694641,
        -0.04598072171211243,
        0.013722382485866547,
        0.6907728314399719,
        -0.03940684720873833
      ],
      "probs": [
        0.52079176902771,
        0.3945786654949188,
        0.48850682377815247,
        0.5034305453300476,
        0.6661388278007507,
        0.4901495575904846
      ],
      "logits_mean": 0.04570278525352478,
      "probs_mean": 0.5105993747711182
    }
  ],
  "logits_full_first": [
    -4.198724746704102,
    -4.558064937591553,
    -4.576184272766113,
    -4.023329734802246,
    -4.5284199714660645,
    -4.295491695404053
  ],
  "probs_full_first": [
    0.014792606234550476,
    0.010373583994805813,
    0.010189211927354336,
    0.017578743398189545,
    0.010682377964258194,
    0.013446593657135963
  ],
  "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": 0.0017614364624023438,
  "mean_abs_diff": 0.0005371582487391101,
  "max_abs_diff_probs": 4.1484832763671875e-05,
  "mean_abs_diff_probs": 1.3109325664117932e-05,
  "passed": true,
  "pearson_logits": 0.999999987997105,
  "pearson_probs": 0.9999999991677251,
  "shape": [
    6,
    6
  ]
}

性能

{
  "first_run_ms": 1.1911517940461636,
  "second_run_ms": 1.0504801757633686,
  "avg_ms": 1.0651994496583939,
  "min_ms": 1.0201898403465748,
  "max_ms": 1.0958807542920113,
  "p50_ms": 1.0689706541597843,
  "p90_ms": 1.0900582652539015,
  "throughput_seq_per_s": 5632.747934599648,
  "throughput_bases_per_s": 5638380.682534249,
  "batch_size": 6,
  "sequence_length": 1001,
  "dtype": "float32",
  "peak_memory_mb": 46.05322265625
}

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