ultrams 0.1.0__tar.gz

ultrams-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

ultrams-0.1.0/PKG-INFO

@@ -0,0 +1,67 @@
+Metadata-Version: 2.4
+Name: ultrams
+Version: 0.1.0
+Summary: UltraMS spectrum representations and model inference
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/Dsadd4/UltraMS
+Project-URL: Source, https://github.com/Dsadd4/UltraMS
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy<3,>=1.24
+Requires-Dist: torch>=2.1
+Requires-Dist: transformers<5,>=4.33
+Provides-Extra: hub
+Requires-Dist: huggingface-hub<2,>=0.23; extra == "hub"
+Dynamic: license-file
+
+# UltraMS
+
+Encode an MS/MS spectrum in three lines. Fine-tune the same encoder with your own labels.
+
+```bash
+pip install "ultrams[hub]"
+```
+
+```python
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only")
+embedding = model.encode(
+    mz=[100.1, 121.1, 150.0], intensity=[20, 100, 35], precursor_mz=301.2
+).embedding  # NumPy vector
+```
+
+## Fine-tune
+
+`encode_tensor` keeps gradients. Add a task head and train the encoder and head together:
+
+```python
+import torch
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only").train()
+head = torch.nn.Linear(model.embedding_dim, 1)
+optimizer = torch.optim.AdamW([*model.parameters(), *head.parameters()], lr=1e-5)
+
+for mz, intensity, precursor_mz, target in training_rows:
+    prediction = head(model.encode_tensor(mz, intensity, precursor_mz=precursor_mz))
+    loss = torch.nn.functional.mse_loss(prediction.flatten(), torch.tensor([target], dtype=torch.float32))
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+```
+
+`training_rows` contains your labelled spectra. For MoNA or UltraAtlas fine-tuning, replace the model ID in the same code.
+
+## Checkpoints
+
+| Model | Hugging Face | Use |
+| --- | --- | --- |
+| RT-only D11 | [UltraMS-RT-only](https://huggingface.co/dsadd4/UltraMS-RT-only) | General MS/MS representation |
+| MoNA contrastive | [UltraMS-MoNA](https://huggingface.co/dsadd4/UltraMS-MoNA) | MoNA spectrum similarity |
+| UltraAtlas contrastive | [UltraMS-UltraAtlas](https://huggingface.co/dsadd4/UltraMS-UltraAtlas) | Figure 5 Atlas application |
+
+All three can also be loaded from a downloaded `model.pt` with `UltraMS.from_checkpoint(path)`.
+
+The current pure Ae3 pretraining and RT/ion-mode adaptation source is in [training/](training/).

ultrams-0.1.0/README.md

@@ -0,0 +1,50 @@
+# UltraMS
+
+Encode an MS/MS spectrum in three lines. Fine-tune the same encoder with your own labels.
+
+```bash
+pip install "ultrams[hub]"
+```
+
+```python
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only")
+embedding = model.encode(
+    mz=[100.1, 121.1, 150.0], intensity=[20, 100, 35], precursor_mz=301.2
+).embedding  # NumPy vector
+```
+
+## Fine-tune
+
+`encode_tensor` keeps gradients. Add a task head and train the encoder and head together:
+
+```python
+import torch
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only").train()
+head = torch.nn.Linear(model.embedding_dim, 1)
+optimizer = torch.optim.AdamW([*model.parameters(), *head.parameters()], lr=1e-5)
+
+for mz, intensity, precursor_mz, target in training_rows:
+    prediction = head(model.encode_tensor(mz, intensity, precursor_mz=precursor_mz))
+    loss = torch.nn.functional.mse_loss(prediction.flatten(), torch.tensor([target], dtype=torch.float32))
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+```
+
+`training_rows` contains your labelled spectra. For MoNA or UltraAtlas fine-tuning, replace the model ID in the same code.
+
+## Checkpoints
+
+| Model | Hugging Face | Use |
+| --- | --- | --- |
+| RT-only D11 | [UltraMS-RT-only](https://huggingface.co/dsadd4/UltraMS-RT-only) | General MS/MS representation |
+| MoNA contrastive | [UltraMS-MoNA](https://huggingface.co/dsadd4/UltraMS-MoNA) | MoNA spectrum similarity |
+| UltraAtlas contrastive | [UltraMS-UltraAtlas](https://huggingface.co/dsadd4/UltraMS-UltraAtlas) | Figure 5 Atlas application |
+
+All three can also be loaded from a downloaded `model.pt` with `UltraMS.from_checkpoint(path)`.
+
+The current pure Ae3 pretraining and RT/ion-mode adaptation source is in [training/](training/).

ultrams-0.1.0/pyproject.toml

@@ -0,0 +1,26 @@
+[build-system]
+requires = ["setuptools>=77"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "ultrams"
+version = "0.1.0"
+description = "UltraMS spectrum representations and model inference"
+readme = "README.md"
+requires-python = ">=3.10"
+license = "Apache-2.0"
+dependencies = [
+  "numpy>=1.24,<3",
+  "torch>=2.1",
+  "transformers>=4.33,<5",
+]
+
+[project.urls]
+Homepage = "https://github.com/Dsadd4/UltraMS"
+Source = "https://github.com/Dsadd4/UltraMS"
+
+[project.optional-dependencies]
+hub = ["huggingface-hub>=0.23,<2"]
+
+[tool.setuptools.packages.find]
+where = ["src"]

ultrams-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

ultrams-0.1.0/src/ultrams/__init__.py

@@ -0,0 +1,6 @@
+"""UltraMS spectrum encoder."""
+
+from .inference import UltraMS, SpectrumEmbedding
+
+__all__ = ["UltraMS", "SpectrumEmbedding"]
+

ultrams-0.1.0/src/ultrams/_architecture.py

@@ -0,0 +1,251 @@
+"""Inference architecture with the state-dict names of the UltraMS v9 trainer."""
+
+from __future__ import annotations
+
+from collections.abc import Mapping
+from typing import Any
+
+import torch
+from torch import nn
+from transformers import BertConfig, BertModel
+
+
+MODEL_DEFAULTS: dict[str, Any] = {
+    "d_model": 1024,
+    "num_heads": 16,
+    "num_layers": 16,
+    "dim_feedforward": 4096,
+    "dropout": 0.1,
+    "max_peaks": 150,
+    "mz_range": (0, 2000),
+    "mz_bin_size": 0.1,
+    "int_bin_size": 0.1,
+    "mask_ratio": 0.15,
+    "mz_codebook_dim": 512,
+    "int_codebook_dim": 512,
+    "coarse_num": 2000,
+    "teacher_forcing_ratio": 0.0,
+    "difficulty_weight_alpha": 2.0,
+    "int_loss_weight": 0.1,
+}
+
+
+def model_config(saved: Mapping[str, Any] | None) -> dict[str, Any]:
+    result = dict(MODEL_DEFAULTS)
+    if isinstance(saved, Mapping):
+        for source in (saved, saved.get("model")):
+            if isinstance(source, Mapping):
+                for key in result:
+                    if key in source:
+                        result[key] = source[key]
+    result["mz_range"] = tuple(result["mz_range"])
+    return result
+
+
+class SoftCodebookLookup(nn.Module):
+    def __init__(
+        self,
+        num_codes: int,
+        d_out: int,
+        val_min: float = 0.0,
+        val_max: float = 1.0,
+        alpha_init: float = 0.1,
+        circular: bool = False,
+    ) -> None:
+        super().__init__()
+        self.circular = circular
+        self.val_range = val_max - val_min
+        self.codebook = nn.Embedding(num_codes, d_out)
+        self.register_buffer("centers", torch.linspace(val_min, val_max, num_codes))
+        self.alpha = nn.Parameter(torch.full((num_codes,), alpha_init))
+
+    def forward(self, values: torch.Tensor) -> torch.Tensor:
+        diff = values.unsqueeze(-1) - self.centers
+        if self.circular:
+            abs_diff = diff.abs()
+            diff = torch.min(abs_diff, self.val_range - abs_diff)
+        weights = torch.softmax(-diff.pow(2) / (self.alpha.abs() + 1e-6), dim=-1)
+        return weights @ self.codebook.weight
+
+
+class MultiResolutionMzCodebook(nn.Module):
+    def __init__(
+        self,
+        d_out: int,
+        mz_max: float = 2000.0,
+        l0_num: int = 200,
+        l1_num: int = 2000,
+        l2_num: int = 200,
+        l3_num: int = 200,
+    ) -> None:
+        super().__init__()
+        d_each = d_out // 4
+        d_last = d_out - 3 * d_each
+        self.level0 = SoftCodebookLookup(l0_num, d_each, 0.0, mz_max, 100.0)
+        self.level1 = SoftCodebookLookup(l1_num, d_each, 0.0, mz_max, 1.0)
+        self.level2 = SoftCodebookLookup(l2_num, d_each, 0.0, 1.0, 0.01, True)
+        self.level3 = SoftCodebookLookup(l3_num, d_last, 0.0, 0.1, 0.001, True)
+
+    def forward(self, mz: torch.Tensor) -> torch.Tensor:
+        return torch.cat(
+            (
+                self.level0(mz),
+                self.level1(mz),
+                self.level2(torch.fmod(mz.abs(), 1.0)),
+                self.level3(torch.fmod(mz.abs(), 0.1)),
+            ),
+            dim=-1,
+        )
+
+
+class CodebookIntensityEmbedding(nn.Module):
+    def __init__(self, d_out: int, num_codes: int = 100, alpha_init: float = 0.1):
+        super().__init__()
+        self.lookup = SoftCodebookLookup(num_codes, d_out, 0.0, 1.0, alpha_init)
+
+    def forward(self, intensity: torch.Tensor) -> torch.Tensor:
+        return self.lookup(intensity)
+
+
+class CodebookPeakEncoder(nn.Module):
+    def __init__(
+        self,
+        d_model: int,
+        mz_dim: int = 512,
+        int_dim: int = 512,
+        int_num_codes: int = 100,
+        mz_max: float = 2000.0,
+        coarse_num: int = 2000,
+    ) -> None:
+        super().__init__()
+        self.mz_dim = mz_dim
+        self.int_dim = int_dim
+        self.mz_codebook = MultiResolutionMzCodebook(
+            mz_dim, mz_max=mz_max, l1_num=coarse_num
+        )
+        self.int_codebook = CodebookIntensityEmbedding(int_dim, int_num_codes)
+        self.mz_type_emb = nn.Parameter(torch.zeros(mz_dim))
+        self.int_type_emb = nn.Parameter(torch.zeros(int_dim))
+        self.proj = nn.Linear(mz_dim + int_dim, d_model)
+
+    def forward(self, peaks: torch.Tensor) -> torch.Tensor:
+        mz = self.mz_codebook(peaks[..., 0]) + self.mz_type_emb
+        intensity = self.int_codebook(peaks[..., 1]) + self.int_type_emb
+        return self.proj(torch.cat((mz, intensity), dim=-1))
+
+
+class UltraMSBackbone(nn.Module):
+    """The full pretraining model layout, with an inference-only forward path."""
+
+    def __init__(self, config: Mapping[str, Any]) -> None:
+        super().__init__()
+        self.max_peaks = int(config["max_peaks"])
+        self.mz_range = tuple(config["mz_range"])
+        self.mz_bin_size = float(config["mz_bin_size"])
+        self.mz_num_bins = int((self.mz_range[1] - self.mz_range[0]) / self.mz_bin_size) + 1
+        self.int_bin_size = float(config["int_bin_size"])
+        self.int_num_bins = int(1.0 / self.int_bin_size)
+        self.ms_start_idx = 2
+        self.mask_ratio = float(config["mask_ratio"])
+        self.int_loss_weight = float(config.get("int_loss_weight", 1.0))
+        self.teacher_forcing_ratio = 0.0
+        self.difficulty_weight_alpha = float(config.get("difficulty_weight_alpha", 2.0))
+
+        d_model = int(config["d_model"])
+        mz_dim = int(config["mz_codebook_dim"])
+        self.peak_encoder = CodebookPeakEncoder(
+            d_model=d_model,
+            mz_dim=mz_dim,
+            int_dim=int(config["int_codebook_dim"]),
+            mz_max=float(self.mz_range[1]),
+            coarse_num=int(config.get("coarse_num", 2000)),
+        )
+        self.cls_emb = nn.Parameter(torch.zeros(1, 1, d_model))
+        self.precursor_type_emb = nn.Parameter(torch.zeros(1, 1, d_model))
+        max_seq_len = self.max_peaks + 2
+        self.pos_emb = nn.Embedding(max_seq_len, d_model)
+        mz_precision = 1.0
+        vocab_size = 13 + len(torch.arange(*self.mz_range, mz_precision))
+        bert_config = BertConfig(
+            vocab_size=vocab_size,
+            hidden_size=d_model,
+            num_hidden_layers=int(config["num_layers"]),
+            num_attention_heads=int(config["num_heads"]),
+            intermediate_size=int(config["dim_feedforward"]),
+            hidden_dropout_prob=float(config["dropout"]),
+            attention_probs_dropout_prob=float(config["dropout"]),
+            max_position_embeddings=max(max_seq_len * 2, 512),
+            layer_norm_eps=1e-6,
+            is_decoder=False,
+            add_cross_attention=False,
+        )
+        self.encoder = BertModel(bert_config, add_pooling_layer=False)
+        self.encoder.embeddings.word_embeddings.requires_grad_(False)
+        self.dropout = nn.Dropout(float(config["dropout"]))
+        self.mz_level0_bins = 200
+        self.mz_level1_bins = 10
+        self.mz_level2_bins = 10
+        self.register_buffer("level0_centers", torch.linspace(5.0, 1995.0, 200))
+        self.register_buffer("level1_offsets", torch.linspace(0.5, 9.5, 10))
+        self.register_buffer("level2_offsets", torch.linspace(0.05, 0.95, 10))
+        self.mz_level0_head = nn.Linear(d_model, self.mz_level0_bins)
+        self.mz_level1_head = nn.Sequential(
+            nn.Linear(d_model + mz_dim, d_model // 2),
+            nn.GELU(),
+            nn.Linear(d_model // 2, self.mz_level1_bins),
+        )
+        self.mz_level2_head = nn.Sequential(
+            nn.Linear(d_model + mz_dim, d_model // 2),
+            nn.GELU(),
+            nn.Linear(d_model // 2, self.mz_level2_bins),
+        )
+        self.int_predictor = nn.Linear(d_model, self.int_num_bins)
+
+    def encode(
+        self, spectra: torch.Tensor, attention_mask: torch.Tensor, precursor_mz: torch.Tensor
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        batch_size = spectra.shape[0]
+        peak_embeddings = self.peak_encoder(spectra)
+        cls_embedding = self.cls_emb.expand(batch_size, 1, -1)
+        precursor_input = torch.stack(
+            (precursor_mz, torch.full_like(precursor_mz, 1.1)), dim=-1
+        ).unsqueeze(1)
+        precursor_embedding = self.peak_encoder(precursor_input) + self.precursor_type_emb
+        embeddings = torch.cat((cls_embedding, precursor_embedding, peak_embeddings), dim=1)
+        positions = torch.arange(embeddings.shape[1], device=spectra.device).unsqueeze(0)
+        embeddings = self.dropout(embeddings + self.pos_emb(positions))
+        prefix_attention = torch.ones(
+            batch_size, 2, dtype=attention_mask.dtype, device=spectra.device
+        )
+        full_attention = torch.cat((prefix_attention, attention_mask), dim=1)
+        hidden = self.encoder(
+            inputs_embeds=embeddings, attention_mask=full_attention
+        ).last_hidden_state
+        return hidden, hidden[:, 0, :]
+
+
+class _Head(nn.Module):
+    def __init__(self, d_model: int) -> None:
+        super().__init__()
+        self.head = nn.Sequential(
+            nn.Linear(d_model, d_model // 2),
+            nn.GELU(),
+            nn.LayerNorm(d_model // 2),
+            nn.Linear(d_model // 2, 1),
+        )
+
+    def forward(self, cls_embedding: torch.Tensor) -> torch.Tensor:
+        return self.head(cls_embedding).squeeze(-1)
+
+
+class UltraMSWithHeads(nn.Module):
+    def __init__(self, config: Mapping[str, Any]) -> None:
+        super().__init__()
+        self.base = UltraMSBackbone(config)
+        self.rt_head = _Head(int(config["d_model"]))
+        self.pol_head = _Head(int(config["d_model"]))
+
+    def encode(
+        self, spectra: torch.Tensor, attention_mask: torch.Tensor, precursor_mz: torch.Tensor
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        return self.base.encode(spectra, attention_mask, precursor_mz)

ultrams-0.1.0/src/ultrams/inference.py

@@ -0,0 +1,261 @@
+"""Load an UltraMS checkpoint and encode one MS/MS spectrum."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from itertools import chain
+from pathlib import Path
+from typing import Any
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+
+from ._architecture import UltraMSBackbone, UltraMSWithHeads, model_config
+
+
+@dataclass(frozen=True)
+class SpectrumEmbedding:
+    cls: np.ndarray
+    peak_intensity_weighted: np.ndarray
+    fusion: np.ndarray
+    projection: np.ndarray | None = None
+    rt_seconds: float | None = None
+
+    @property
+    def embedding(self) -> np.ndarray:
+        """The model-specific representation used for similarity search."""
+        return self.projection if self.projection is not None else self.cls
+
+
+def _prepare_spectrum(
+    mz: Any, intensity: Any, max_peaks: int
+) -> tuple[np.ndarray, np.ndarray]:
+    mz_array = np.asarray(mz, dtype=np.float32)
+    intensity_array = np.asarray(intensity, dtype=np.float32)
+    if mz_array.ndim != 1 or intensity_array.ndim != 1 or len(mz_array) != len(intensity_array):
+        raise ValueError("mz and intensity must be one-dimensional arrays of equal length")
+    if not np.isfinite(mz_array).all() or not np.isfinite(intensity_array).all():
+        raise ValueError("mz and intensity must contain finite values")
+    valid = mz_array > 0
+    mz_array = mz_array[valid]
+    intensity_array = intensity_array[valid]
+    if len(mz_array) < 3:
+        raise ValueError("a spectrum needs at least three peaks with positive m/z")
+    maximum = intensity_array.max()
+    if maximum > 0:
+        intensity_array = intensity_array / maximum
+    intensity_array = np.clip(intensity_array, 0, 1)
+    if len(mz_array) > max_peaks:
+        selected = np.argsort(intensity_array)[-max_peaks:]
+        selected = np.sort(selected)
+        mz_array = mz_array[selected]
+        intensity_array = intensity_array[selected]
+    order = np.argsort(mz_array)
+    peaks = np.stack((mz_array[order], intensity_array[order]), axis=-1).astype(np.float32)
+    mask = np.ones(len(peaks), dtype=np.int64)
+    return peaks, mask
+
+
+def _checkpoint_state(
+    payload: Any,
+) -> tuple[dict[str, torch.Tensor], dict[str, Any], int, str, torch.nn.Module | None]:
+    if not isinstance(payload, dict):
+        raise ValueError("checkpoint must be a mapping")
+    model_family = "base"
+    projection = None
+    if "encoder_state_dict" in payload:
+        state = payload["encoder_state_dict"]
+        if "proj_state_dict" in payload:
+            model_family = "atlas"
+            projection = _atlas_projection(payload["proj_state_dict"])
+        elif "projection_state_dict" in payload:
+            model_family = "mona"
+            projection = _mona_projection(payload["projection_state_dict"])
+        else:
+            raise ValueError("encoder checkpoint has no projection state")
+    else:
+        state = payload.get("model_state_dict")
+        if state is None:
+            state = payload.get("base_state_dict")
+    if not isinstance(state, dict) or not state:
+        raise ValueError("checkpoint has no supported model state")
+    if all(key.startswith("module.") for key in state):
+        state = {key[len("module."):]: value for key, value in state.items()}
+    if not all(isinstance(key, str) and isinstance(value, torch.Tensor) for key, value in state.items()):
+        raise ValueError("checkpoint state must contain named tensors")
+    saved_config = payload.get("config")
+    config = model_config(saved_config)
+    first_key = "base.pos_emb.weight" if any(key.startswith("base.") for key in state) else "pos_emb.weight"
+    if first_key in state:
+        config["max_peaks"] = int(state[first_key].shape[0]) - 2
+    input_max_peaks = config["max_peaks"]
+    if model_family == "mona" and isinstance(saved_config, dict):
+        input_max_peaks = int(saved_config.get("max_peaks", 100))
+    return state, config, input_max_peaks, model_family, projection
+
+
+def _mona_projection(state: Any) -> torch.nn.Module:
+    if not isinstance(state, dict):
+        raise ValueError("MoNA checkpoint has no projection weights")
+    if set(state) == {"weight", "bias"}:
+        size = int(state["weight"].shape[1])
+        projection: torch.nn.Module = torch.nn.Linear(size, size)
+    elif "1.weight" in state and "4.weight" in state:
+        size = int(state["1.weight"].shape[1])
+        hidden = int(state["1.weight"].shape[0])
+        projection = torch.nn.Sequential(
+            torch.nn.LayerNorm(size),
+            torch.nn.Linear(size, hidden),
+            torch.nn.GELU(),
+            torch.nn.Dropout(0.0),
+            torch.nn.Linear(hidden, size),
+        )
+    else:
+        raise ValueError("unrecognized MoNA projection layout")
+    projection.load_state_dict(state, strict=True)
+    return projection.eval()
+
+
+def _atlas_projection(state: Any) -> torch.nn.Module:
+    if not isinstance(state, dict) or "net.3.weight" not in state:
+        raise ValueError("unrecognized UltraAtlas projection layout")
+    hidden = int(state["net.1.weight"].shape[0])
+    output = int(state["net.3.weight"].shape[0])
+    projection = torch.nn.Sequential(
+        torch.nn.LayerNorm(hidden),
+        torch.nn.Linear(hidden, hidden),
+        torch.nn.GELU(),
+        torch.nn.Linear(hidden, output),
+    )
+    named_state = {key.removeprefix("net."): value for key, value in state.items()}
+    projection.load_state_dict(named_state, strict=True)
+    return projection.eval()
+
+
+class UltraMS:
+    """Inference wrapper for the pretraining and RT/polarity model layouts."""
+
+    def __init__(
+        self,
+        model: UltraMSBackbone | UltraMSWithHeads,
+        config: dict[str, Any],
+        input_max_peaks: int,
+        model_family: str,
+        projection: torch.nn.Module | None,
+    ):
+        self.model = model.eval()
+        self.config = config
+        self.input_max_peaks = input_max_peaks
+        self.model_family = model_family
+        self.projection = projection
+
+    @property
+    def embedding_dim(self) -> int:
+        if self.projection is not None:
+            for layer in reversed(list(self.projection.modules())):
+                if isinstance(layer, torch.nn.Linear):
+                    return layer.out_features
+        return int(self.config["d_model"])
+
+    def parameters(self):
+        """Parameters of the encoder and its optional contrastive projection."""
+        return chain(
+            self.model.parameters(),
+            self.projection.parameters() if self.projection is not None else (),
+        )
+
+    def train(self) -> "UltraMS":
+        self.model.train()
+        if self.projection is not None:
+            self.projection.train()
+        return self
+
+    def eval(self) -> "UltraMS":
+        self.model.eval()
+        if self.projection is not None:
+            self.projection.eval()
+        return self
+
+    def encode_tensor(
+        self, mz: Any, intensity: Any, *, precursor_mz: float
+    ) -> torch.Tensor:
+        """Differentiable model embedding of one spectrum, shape ``(1, embedding_dim)``."""
+        peaks, attention = _prepare_spectrum(mz, intensity, self.input_max_peaks)
+        if not np.isfinite(precursor_mz) or precursor_mz <= 0:
+            raise ValueError("precursor_mz must be a positive finite number")
+        device = next(self.model.parameters()).device
+        spectra = torch.as_tensor(peaks, device=device).unsqueeze(0)
+        mask = torch.as_tensor(attention, device=device).unsqueeze(0)
+        precursor = torch.tensor([precursor_mz], dtype=torch.float32, device=device)
+        _, cls = self.model.encode(spectra, mask, precursor)
+        embedding = self.projection(cls) if self.projection is not None else cls
+        return F.normalize(embedding.float(), dim=-1)
+
+    @classmethod
+    def from_checkpoint(
+        cls, checkpoint: str | Path, *, device: str | torch.device = "cpu"
+    ) -> "UltraMS":
+        path = Path(checkpoint).expanduser()
+        payload = torch.load(path, map_location="cpu", weights_only=False)
+        state, config, input_max_peaks, model_family, projection = _checkpoint_state(payload)
+        unified = any(key.startswith("base.") for key in state)
+        model = UltraMSWithHeads(config) if unified else UltraMSBackbone(config)
+        model.load_state_dict(state, strict=True)
+        model.to(torch.device(device))
+        if projection is not None:
+            projection.to(torch.device(device))
+        return cls(model, config, input_max_peaks, model_family, projection)
+
+    @classmethod
+    def from_hub(
+        cls,
+        repo_id: str,
+        *,
+        filename: str = "model.pt",
+        revision: str | None = None,
+        device: str | torch.device = "cpu",
+    ) -> "UltraMS":
+        try:
+            from huggingface_hub import hf_hub_download
+        except ImportError as exc:
+            raise ImportError("Hugging Face downloads require pip install 'ultrams[hub]'") from exc
+        path = hf_hub_download(repo_id=repo_id, filename=filename, revision=revision)
+        return cls.from_checkpoint(path, device=device)
+
+    @torch.inference_mode()
+    def encode(
+        self, mz: Any, intensity: Any, *, precursor_mz: float
+    ) -> SpectrumEmbedding:
+        peaks, attention = _prepare_spectrum(mz, intensity, self.input_max_peaks)
+        if not np.isfinite(precursor_mz) or precursor_mz <= 0:
+            raise ValueError("precursor_mz must be a positive finite number")
+        device = next(self.model.parameters()).device
+        spectra = torch.as_tensor(peaks, device=device).unsqueeze(0)
+        mask = torch.as_tensor(attention, device=device).unsqueeze(0)
+        precursor = torch.tensor([precursor_mz], dtype=torch.float32, device=device)
+        hidden, raw_cls = self.model.encode(spectra, mask, precursor)
+        peak_hidden = hidden[:, 2 : 2 + spectra.shape[1]]
+        weights = spectra[:, :, 1] * mask.float()
+        weighted = (peak_hidden * weights.unsqueeze(-1)).sum(1) / weights.sum(
+            1, keepdim=True
+        ).clamp_min(1e-8)
+        cls = F.normalize(raw_cls.float(), dim=-1)
+        weighted = F.normalize(weighted.float(), dim=-1)
+        projected = None
+        if self.projection is not None:
+            projected = self.projection(raw_cls)
+            if self.model_family == "atlas":
+                projected = F.normalize(projected, dim=-1)
+        fusion_cls = F.normalize(projected.float(), dim=-1) if self.model_family == "mona" else cls
+        fusion = F.normalize(torch.cat((fusion_cls, weighted), dim=-1), dim=-1)
+        rt_seconds = None
+        if isinstance(self.model, UltraMSWithHeads):
+            rt_seconds = float(self.model.rt_head(hidden[:, 0])[0] * 600.0)
+        return SpectrumEmbedding(
+            cls=cls[0].cpu().numpy(),
+            peak_intensity_weighted=weighted[0].cpu().numpy(),
+            fusion=fusion[0].cpu().numpy(),
+            projection=projected[0].cpu().numpy() if projected is not None else None,
+            rt_seconds=rt_seconds,
+        )

ultrams-0.1.0/src/ultrams.egg-info/PKG-INFO

@@ -0,0 +1,67 @@
+Metadata-Version: 2.4
+Name: ultrams
+Version: 0.1.0
+Summary: UltraMS spectrum representations and model inference
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/Dsadd4/UltraMS
+Project-URL: Source, https://github.com/Dsadd4/UltraMS
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy<3,>=1.24
+Requires-Dist: torch>=2.1
+Requires-Dist: transformers<5,>=4.33
+Provides-Extra: hub
+Requires-Dist: huggingface-hub<2,>=0.23; extra == "hub"
+Dynamic: license-file
+
+# UltraMS
+
+Encode an MS/MS spectrum in three lines. Fine-tune the same encoder with your own labels.
+
+```bash
+pip install "ultrams[hub]"
+```
+
+```python
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only")
+embedding = model.encode(
+    mz=[100.1, 121.1, 150.0], intensity=[20, 100, 35], precursor_mz=301.2
+).embedding  # NumPy vector
+```
+
+## Fine-tune
+
+`encode_tensor` keeps gradients. Add a task head and train the encoder and head together:
+
+```python
+import torch
+from ultrams import UltraMS
+
+model = UltraMS.from_hub("dsadd4/UltraMS-RT-only").train()
+head = torch.nn.Linear(model.embedding_dim, 1)
+optimizer = torch.optim.AdamW([*model.parameters(), *head.parameters()], lr=1e-5)
+
+for mz, intensity, precursor_mz, target in training_rows:
+    prediction = head(model.encode_tensor(mz, intensity, precursor_mz=precursor_mz))
+    loss = torch.nn.functional.mse_loss(prediction.flatten(), torch.tensor([target], dtype=torch.float32))
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+```
+
+`training_rows` contains your labelled spectra. For MoNA or UltraAtlas fine-tuning, replace the model ID in the same code.
+
+## Checkpoints
+
+| Model | Hugging Face | Use |
+| --- | --- | --- |
+| RT-only D11 | [UltraMS-RT-only](https://huggingface.co/dsadd4/UltraMS-RT-only) | General MS/MS representation |
+| MoNA contrastive | [UltraMS-MoNA](https://huggingface.co/dsadd4/UltraMS-MoNA) | MoNA spectrum similarity |
+| UltraAtlas contrastive | [UltraMS-UltraAtlas](https://huggingface.co/dsadd4/UltraMS-UltraAtlas) | Figure 5 Atlas application |
+
+All three can also be loaded from a downloaded `model.pt` with `UltraMS.from_checkpoint(path)`.
+
+The current pure Ae3 pretraining and RT/ion-mode adaptation source is in [training/](training/).

ultrams-0.1.0/src/ultrams.egg-info/SOURCES.txt

@@ -0,0 +1,11 @@
+LICENSE
+README.md
+pyproject.toml
+src/ultrams/__init__.py
+src/ultrams/_architecture.py
+src/ultrams/inference.py
+src/ultrams.egg-info/PKG-INFO
+src/ultrams.egg-info/SOURCES.txt
+src/ultrams.egg-info/dependency_links.txt
+src/ultrams.egg-info/requires.txt
+src/ultrams.egg-info/top_level.txt

ultrams-0.1.0/src/ultrams.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

ultrams-0.1.0/src/ultrams.egg-info/requires.txt

@@ -0,0 +1,6 @@
+numpy<3,>=1.24
+torch>=2.1
+transformers<5,>=4.33
+
+[hub]
+huggingface-hub<2,>=0.23

ultrams-0.1.0/src/ultrams.egg-info/top_level.txt

@@ -0,0 +1 @@
+ultrams