Publish experimental Stage 2 model built on Hugging Face Jobs

1_Pooling/config.json

@@ -0,0 +1,10 @@
+{
+  "word_embedding_dimension": 768,
+  "pooling_mode_cls_token": true,
+  "pooling_mode_mean_tokens": false,
+  "pooling_mode_max_tokens": false,
+  "pooling_mode_mean_sqrt_len_tokens": false,
+  "pooling_mode_weightedmean_tokens": false,
+  "pooling_mode_lasttoken": false,
+  "include_prompt": true
+}

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 LevelCapTech (fine-tuning modifications)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

LICENSE-APACHE-2.0

@@ -0,0 +1,201 @@
+                                 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.

README.md

@@ -1,3 +1,102 @@
 ---
-license: apache-2.0
+language:
+- en
+license: mit
+library_name: sentence-transformers
+base_model: nomic-ai/CodeRankEmbed
+pipeline_tag: feature-extraction
+tags:
+- python
+- fault-localization
+- code-retrieval
+- experimental
+- safetensors
 ---
+
+# FailureLens: experimental fault-localization encoder
+
+This is a **trained experimental model**, rebuilt on Hugging Face Jobs from
+the FailureLens Stage 2 recipe. Given a Python traceback and candidate function
+source texts, it supplies embeddings for cosine-similarity ranking.
+
+**Independent held-out evaluation (Gate C) is pending. No improvement over the
+shipped baseline or the base encoder has been established. This release is for
+experimentation, not evidence of production readiness or agent adoption.**
+
+## Use
+
+Install PyTorch 2.11.0, transformers 5.22.0, sentence-transformers 6.2.0,
+safetensors 0.8.0 and einops 0.8.1. The model contains NomicBERT custom Python
+code from the pinned base, with a documented transformers-5 attention-mask fix.
+Review that code before enabling `trust_remote_code`. This example pins the
+model revision and uses only safetensors weights.
+
+```python
+from huggingface_hub import HfApi
+from sentence_transformers import SentenceTransformer
+
+revision = HfApi().model_info("LevelCapTech/failurelens").sha
+model = SentenceTransformer(
+    "LevelCapTech/failurelens",
+    revision=revision,
+    trust_remote_code=True,
+    device="cpu",
+    model_kwargs={"use_safetensors": True},
+)
+error_log = "ZeroDivisionError: division by zero"
+candidates = [
+    "def average(values): return sum(values) / len(values)",
+    "def greet(name): return 'Hello ' + name",
+]
+# Stage 2 uses the stripped traceback directly, without a query prefix.
+query = model.encode([error_log.strip()], normalize_embeddings=True)
+code = model.encode(candidates, normalize_embeddings=True)
+scores = (query @ code.T)[0]
+ranked_indices = sorted(range(len(candidates)), key=lambda i: (-float(scores[i]), i))
+print([(i, float(scores[i])) for i in ranked_indices])
+```
+
+Scores are similarities, not probabilities. Callers supply the candidate pool;
+this encoder does not locate files, invent line numbers, or generate fixes.
+The current FailureLens library's generic `from_pretrained` path has a different
+query transform and a restricted custom-code allowlist. The standalone example
+above is the supported interface for this experimental release.
+
+## Training and limitations
+
+- Base: `nomic-ai/CodeRankEmbed` at `3c4b60807d71f79b43f3c4363786d9493691f8b1` (MIT).
+- 136,731,648 parameters; only the final encoder block (9,440,256 parameters)
+  is fine-tuned, with per-triplet InfoNCE and cosine similarity.
+- 12 private, sanitized synthetic examples: 6 train from `pallets/click` and
+  6 validation from `python-attrs/attrs`; source repositories are disjoint.
+- Seed 42, 3 epochs, batch size 2, AdamW, learning rate 2e-5, weight decay 0.01,
+  temperature 0.07, cosine learning-rate schedule. CLS pooling is retained.
+- Effective model sequence limit: 8192 tokens.
+  The legacy config's `max_length=512` is not applied by the training function;
+  the effective limit and observed feature lengths are recorded in provenance.
+- Built on Tesla T4. This cloud
+  rebuild has its own checksum; it is not claimed bit-identical to local v3.
+- Validation losses only reflect this tiny synthetic set. No natural-bug,
+  multilingual, latency-budget, or independent accuracy claim is made.
+- Baseline retrieval can omit the actual cause; reranking cannot recover an
+  omitted function. Python traceback support in the library remains limited.
+
+## Reproducibility and distribution
+
+The Job fetched pinned public source and base weights, trained, assembled and
+tested the model, and uploaded the weights directly to the Hub. The submitting
+connection sent only this build script and approximately 6 KB of compressed
+training features as an encrypted Job secret. Private records, features,
+optimizer checkpoints and held-out labels are not distributed here.
+
+- Job: https://huggingface.co/jobs/LevelCapTech/6ae7d7a8c78863f9c62ff0e8
+- Training source: https://github.com/LevelCapTech/FailureLens/tree/2985bfcfbf5bf79bc3b3b0589284f6a4ca481970
+- Weights SHA-256: `8e8fb971f8c1dce22ae29ea4f682cb8387790ad608e06d3800b62d3413a0a9e4`
+- Data SHA-256: `74909f51b588c311de798bee29b748cfc1ead65fd2948053c8f3a1c729e465c1`
+- Exact dependencies, training history, code hashes and smoke checks:
+  [failurelens_provenance.json](failurelens_provenance.json).
+
+The weights are distributed under MIT, as declared by the base model.
+Upstream code retains its copyright headers and license notices; see
+[LICENSE](LICENSE) and [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).
+FailureLens's repository source and build script are Apache-2.0.

THIRD_PARTY_NOTICES.md

@@ -0,0 +1,107 @@
+# Third-party notices
+
+Base model: nomic-ai/CodeRankEmbed @ 3c4b60807d71f79b43f3c4363786d9493691f8b1; the upstream model card declares MIT.
+Original model-code copyright headers are preserved.
+
+
+## Upstream notice: pallets/click
+
+Source: https://raw.githubusercontent.com/pallets/click/2247b35ea1c47c727d7a06e51fa280e12a863ff6/LICENSE.txt
+
+```text
+Copyright 2014 Pallets
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+1.  Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+2.  Redistributions in binary form must reproduce the above copyright
+    notice, this list of conditions and the following disclaimer in the
+    documentation and/or other materials provided with the distribution.
+
+3.  Neither the name of the copyright holder nor the names of its
+    contributors may be used to endorse or promote products derived from
+    this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
+PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+```
+
+## Upstream notice: python-attrs/attrs
+
+Source: https://raw.githubusercontent.com/python-attrs/attrs/c8bb9bfaa2843e7844b5bb13bed4370b51feec4f/LICENSE
+
+```text
+The MIT License (MIT)
+
+Copyright (c) 2015 Hynek Schlawack and the attrs contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+```
+
+## Upstream notice: Dao-AILab/flash-attention
+
+Source: https://raw.githubusercontent.com/Dao-AILab/flash-attention/main/LICENSE
+
+```text
+BSD 3-Clause License
+
+Copyright (c) 2022, the respective contributors, as shown by the AUTHORS file.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+```

config.json

@@ -0,0 +1,57 @@
+{
+  "_name_or_path": "nomic-uiuc/CodeEmbed",
+  "activation_function": "swiglu",
+  "architectures": [
+    "NomicBertModel"
+  ],
+  "attn_pdrop": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_hf_nomic_bert.NomicBertConfig",
+    "AutoModel": "modeling_hf_nomic_bert.NomicBertModel"
+  },
+  "bos_token_id": null,
+  "causal": false,
+  "dense_seq_output": true,
+  "embd_pdrop": 0.1,
+  "eos_token_id": null,
+  "fused_bias_fc": true,
+  "fused_dropout_add_ln": true,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-12,
+  "max_trained_positions": 2048,
+  "mlp_fc1_bias": false,
+  "mlp_fc2_bias": false,
+  "model_type": "nomic_bert",
+  "n_embd": 768,
+  "n_head": 12,
+  "n_inner": 3072,
+  "n_layer": 12,
+  "n_positions": 8192,
+  "pad_vocab_size_multiple": 64,
+  "parallel_block": false,
+  "parallel_block_tied_norm": false,
+  "prenorm": false,
+  "qkv_proj_bias": false,
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "rotary_emb_base": 1000,
+  "rotary_emb_fraction": 1.0,
+  "rotary_emb_interleaved": false,
+  "rotary_emb_scale_base": null,
+  "rotary_scaling_factor": null,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.45.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "use_flash_attn": true,
+  "use_rms_norm": false,
+  "use_xentropy": true,
+  "vocab_size": 30528
+}

config_sentence_transformers.json

@@ -0,0 +1,11 @@
+{
+  "__version__": {
+    "sentence_transformers": "2.7.0.dev0",
+    "transformers": "4.39.3",
+    "pytorch": "2.1.0+cu121"
+  },
+  "prompts": {
+    "query": "Represent this query for searching relevant code: "
+  },
+  "default_prompt_name": null
+}

configuration_hf_nomic_bert.py

@@ -0,0 +1,56 @@
+from transformers import GPT2Config
+
+
+class NomicBertConfig(GPT2Config):
+    model_type = "nomic_bert"
+
+    def __init__(
+        self,
+        prenorm=False,
+        parallel_block=False,
+        parallel_block_tied_norm=False,
+        rotary_emb_fraction=0.0,
+        fused_dropout_add_ln=False,
+        fused_bias_fc=False,
+        use_flash_attn=False,
+        use_xentropy=False,
+        qkv_proj_bias=True,
+        rotary_emb_base=1000,
+        rotary_emb_scale_base=None,
+        rotary_emb_interleaved=False,
+        mlp_fc1_bias=True,
+        mlp_fc2_bias=True,
+        use_rms_norm=False,
+        causal=False,
+        type_vocab_size=2,
+        dense_seq_output=True,
+        pad_vocab_size_multiple=1,
+        tie_word_embeddings=True,
+        rotary_scaling_factor=1.0,
+        max_trained_positions=2048,
+        **kwargs,
+    ):
+        self.prenorm = prenorm
+        self.parallel_block = parallel_block
+        self.parallel_block_tied_norm = parallel_block_tied_norm
+        self.rotary_emb_fraction = rotary_emb_fraction
+        self.tie_word_embeddings = tie_word_embeddings
+        self.fused_dropout_add_ln = fused_dropout_add_ln
+        self.fused_bias_fc = fused_bias_fc
+        self.use_flash_attn = use_flash_attn
+        self.use_xentropy = use_xentropy
+        self.qkv_proj_bias = qkv_proj_bias
+        self.rotary_emb_base = rotary_emb_base
+        self.rotary_emb_scale_base = rotary_emb_scale_base
+        self.rotary_emb_interleaved = rotary_emb_interleaved
+        self.mlp_fc1_bias = mlp_fc1_bias
+        self.mlp_fc2_bias = mlp_fc2_bias
+        self.use_rms_norm = use_rms_norm
+        self.causal = causal
+        self.type_vocab_size = type_vocab_size
+        self.dense_seq_output = dense_seq_output
+        self.pad_vocab_size_multiple = pad_vocab_size_multiple
+        self.rotary_scaling_factor = rotary_scaling_factor
+        self.max_trained_positions = max_trained_positions
+
+        super().__init__(**kwargs)

failurelens_provenance.json

@@ -0,0 +1,104 @@
+{
+  "build_kind": "hf-jobs-stage2-rebuild",
+  "job_id": "6ae7d7a8c78863f9c62ff0e8",
+  "base_model": "nomic-ai/CodeRankEmbed",
+  "base_model_revision": "3c4b60807d71f79b43f3c4363786d9493691f8b1",
+  "source_revision": "2985bfcfbf5bf79bc3b3b0589284f6a4ca481970",
+  "source_archive_sha256": "c632b33ca539cc6185a4e20bbb68af78cb80a42b3ac9b98c2510efa4316bfc5d",
+  "build_script_sha256": "f5554887939ed1b02b7855bad15198f12aceebf58c9c01a987edb2bd6d2138a5",
+  "artifact_sha256": "8e8fb971f8c1dce22ae29ea4f682cb8387790ad608e06d3800b62d3413a0a9e4",
+  "dataset_sha256": "74909f51b588c311de798bee29b748cfc1ead65fd2948053c8f3a1c729e465c1",
+  "training_features_sha256": "baf5af79eb5c8893f84d852232b6540a84348e264f2251b7c4dbb9998f2cb8a9",
+  "effective_max_seq_length": 8192,
+  "max_observed_training_feature_tokens": 991,
+  "training": {
+    "git_sha": "2985bfcfbf5bf79bc3b3b0589284f6a4ca481970",
+    "config": {
+      "seed": 42,
+      "epochs": 3,
+      "batch_size": 2,
+      "learning_rate": 2e-05,
+      "weight_decay": 0.01,
+      "temperature": 0.07,
+      "max_length": 512,
+      "optimizer": "adamw",
+      "objective": "infonce",
+      "model_id": "nomic-ai/CodeRankEmbed",
+      "model_revision": "3c4b60807d71f79b43f3c4363786d9493691f8b1",
+      "dataset": "failurelens-train-issue17-v2",
+      "dataset_sha256": "74909f51b588c311de798bee29b748cfc1ead65fd2948053c8f3a1c729e465c1",
+      "split": "repo-level-6-6",
+      "train_repo": "pallets/click",
+      "val_repo": "python-attrs/attrs",
+      "trainable_layers": "last-encoder-layer"
+    },
+    "history": {
+      "train_loss": [
+        0.47753843665122986,
+        0.5231968661149343,
+        0.412482092777888
+      ],
+      "val_loss": [
+        0.7336968183517456,
+        0.7239548762639364,
+        0.7208325664202372
+      ]
+    },
+    "n_train": 6,
+    "n_val": 6,
+    "trainable_params": 9440256,
+    "total_params": 136731648,
+    "artifact_sha256": "8e8fb971f8c1dce22ae29ea4f682cb8387790ad608e06d3800b62d3413a0a9e4",
+    "elapsed_seconds": 7.6,
+    "peak_gpu_memory_mib": 885,
+    "run_metadata": {
+      "git_sha": "2985bfcfbf5bf79bc3b3b0589284f6a4ca481970",
+      "model_id": "nomic-ai/CodeRankEmbed",
+      "model_revision": "3c4b60807d71f79b43f3c4363786d9493691f8b1",
+      "dataset_sha256": "74909f51b588c311de798bee29b748cfc1ead65fd2948053c8f3a1c729e465c1",
+      "n_train": 6,
+      "n_val": 6,
+      "trainable_params": 9440256,
+      "total_params": 136731648,
+      "python": "3.12.3",
+      "platform": "Linux-6.12.113-138.209.amzn2023.x86_64-x86_64-with-glibc2.39",
+      "torch_version": "2.11.0+cu128",
+      "device_name": "Tesla T4",
+      "device_memory_mib": 15095
+    }
+  },
+  "dependencies": {
+    "torch": "2.11.0+cu128",
+    "transformers": "5.22.0",
+    "sentence-transformers": "6.2.0",
+    "huggingface_hub": "2.4.0",
+    "safetensors": "0.8.0",
+    "einops": "0.8.1",
+    "pydantic": "2.14.2",
+    "jsonschema": "4.26.0",
+    "typer": "0.27.3"
+  },
+  "verification": {
+    "exact_trained_tensors": true,
+    "changed_tensors": 9,
+    "changed_layers": [
+      11
+    ],
+    "offline_cpu_embeddings": true
+  },
+  "independent_evaluation": "pending; no accuracy or superiority claim",
+  "files": {
+    "1_Pooling/config.json": "d02ebf56344d20f773449b15d0c10ee9a86a9178f3b3c9bfecb8b87d4350ce38",
+    "config.json": "5ff856a41d0f53ef2d74520627d464bd75c2efd8f26f381bd528654895c29b6c",
+    "config_sentence_transformers.json": "06bc4c2dc7098e98669beb67cf21dc1373e44a77c7eecdf70aad6585d87190cf",
+    "configuration_hf_nomic_bert.py": "8632792e922e62ab1a6feaab15baf406e223c0547a21de23ab4f520b2b36d674",
+    "model.safetensors": "8e8fb971f8c1dce22ae29ea4f682cb8387790ad608e06d3800b62d3413a0a9e4",
+    "modeling_hf_nomic_bert.py": "f49c7bb2f834ad9915f98326759a712a7804faa870c038c31ac04cba7a2f77a5",
+    "modules.json": "8f4b264b80206c830bebbdcae377e137925650a433b689343a63bdc9b3145460",
+    "sentence_bert_config.json": "d8ecbf1eb946d17aec64c83e6614e5ec1b19b62498cf72a4ed4ec882c86af07e",
+    "special_tokens_map.json": "5d5b662e421ea9fac075174bb0688ee0d9431699900b90662acd44b2a350503a",
+    "tokenizer.json": "91f1def9b9391fdabe028cd3f3fcc4efd34e5d1f08c3bf2de513ebb5911a1854",
+    "tokenizer_config.json": "7809f768ee3614618b3f1b91dcbfab4f6a9d4b79fb1ad5d17feb65a7c1bb5b7a",
+    "vocab.txt": "07eced375cec144d27c900241f3e339478dec958f92fddbc551f295c992038a3"
+  }
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e8fb971f8c1dce22ae29ea4f682cb8387790ad608e06d3800b62d3413a0a9e4
+size 546938144

modeling_hf_nomic_bert.py

@@ -0,0 +1,1225 @@
+# Copyright (c) 2022, Tri Dao.
+# This BERT implementation is based on our MLPerf 2.0 and MLPerf 2.1 BERT implementation.
+# https://github.com/mlcommons/training_results_v2.0/blob/main/HazyResearch/benchmarks/bert/implementations/pytorch/modeling.py
+# https://github.com/mlcommons/training_results_v2.1/blob/main/Azure-HazyResearch/benchmarks/bert/implementations/ND96amsr_A100_v4/modeling.py
+
+import logging
+
+# Inspired by https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py
+import os
+import re
+from collections import OrderedDict
+from functools import partial
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from einops import rearrange, repeat
+from safetensors.torch import load_file as safe_load_file
+from transformers import GPT2Config, PreTrainedModel
+from transformers.models.bert.modeling_bert import (
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    MaskedLMOutput,
+    SequenceClassifierOutput,
+)
+from transformers.utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, WEIGHTS_NAME
+from transformers.utils.hub import cached_file, get_checkpoint_shard_files
+
+from .configuration_hf_nomic_bert import NomicBertConfig
+
+logger = logging.getLogger(__name__)
+
+
+# adapted from flash attention, added safe serialization option for hf models
+def state_dict_from_pretrained(model_name, safe_serialization=False, device=None, dtype=None):
+    # If not fp32, then we don't want to load directly to the GPU
+    mapped_device = "cpu" if dtype not in [torch.float32, None] else device
+    is_sharded = False
+    load_safe = False
+    resolved_archive_file = None
+
+    weights_path = os.path.join(model_name, WEIGHTS_NAME)
+    weights_index_path = os.path.join(model_name, WEIGHTS_INDEX_NAME)
+    safe_weights_path = os.path.join(model_name, SAFE_WEIGHTS_NAME)
+    safe_weights_index_path = os.path.join(model_name, SAFE_WEIGHTS_INDEX_NAME)
+
+    if os.path.isfile(weights_path):
+        resolved_archive_file = cached_file(model_name, WEIGHTS_NAME, _raise_exceptions_for_missing_entries=False)
+    elif os.path.isfile(weights_index_path):
+        resolved_archive_file = cached_file(model_name, WEIGHTS_INDEX_NAME, _raise_exceptions_for_missing_entries=False)
+        is_sharded = True
+    elif os.path.isfile(safe_weights_path):
+        resolved_archive_file = cached_file(model_name, SAFE_WEIGHTS_NAME, _raise_exceptions_for_missing_entries=False)
+        load_safe = True
+    elif os.path.isfile(safe_weights_index_path):
+        resolved_archive_file = cached_file(
+            model_name, SAFE_WEIGHTS_INDEX_NAME, _raise_exceptions_for_missing_entries=False
+        )
+        is_sharded = True
+        load_safe = True
+    else:  # Try loading from HF hub instead of from local files
+        weight_name = WEIGHTS_NAME if not safe_serialization else SAFE_WEIGHTS_NAME
+        resolved_archive_file = cached_file(model_name, weight_name, _raise_exceptions_for_missing_entries=False)
+        if resolved_archive_file is None:
+            weight_index = WEIGHTS_INDEX_NAME if not safe_serialization else SAFE_WEIGHTS_INDEX_NAME
+            resolved_archive_file = cached_file(model_name, weight_index, _raise_exceptions_for_missing_entries=False)
+            if resolved_archive_file is not None:
+                is_sharded = True
+
+        load_safe = safe_serialization
+
+    if resolved_archive_file is None:
+        raise EnvironmentError(f"Model name {model_name} was not found.")
+
+    if load_safe:
+        loader = partial(safe_load_file, device=mapped_device)
+    else:
+        loader = partial(torch.load, map_location=mapped_device)
+
+    if is_sharded:
+        # resolved_archive_file becomes a list of files that point to the different
+        # checkpoint shards in this case.
+        resolved_archive_file, sharded_metadata = get_checkpoint_shard_files(model_name, resolved_archive_file)
+        state_dict = {}
+        for sharded_file in resolved_archive_file:
+            state_dict.update(loader(sharded_file))
+    else:
+        state_dict = loader(resolved_archive_file)
+    # Convert dtype before moving to GPU to save memory
+    if dtype is not None:
+        state_dict = {k: v.to(dtype=dtype) for k, v in state_dict.items()}
+    state_dict = {k: v.to(device=device) for k, v in state_dict.items()}
+    return state_dict
+
+
+def filter_shapes(state_dict, model):
+    """
+    Filters the state dict to match the current model shape.
+    """
+    filtered_state_dict = {}
+    for key, value in state_dict.items():
+        if key in model.state_dict():
+            if value.shape == model.state_dict()[key].shape:
+                filtered_state_dict[key] = value
+    return filtered_state_dict
+
+
+def remap_bert_state_dict(state_dict, config, remove_bert=False, remove_cls_weights=False, add_pooling_layer=False):
+    """
+    Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
+    """
+
+    def add_bert_prefix(key):
+        # prepend bert. to the key
+        if key.startswith("bert.") or key.startswith("cls."):
+            return key
+        return f"bert.{key}"
+
+    state_dict = OrderedDict((add_bert_prefix(k), v) for k, v in state_dict.items())
+
+    # LayerNorm
+    def key_mapping_ln_gamma_beta(key):
+        key = re.sub(r"LayerNorm.gamma$", "LayerNorm.weight", key)
+        key = re.sub(r"LayerNorm.beta$", "LayerNorm.bias", key)
+        return key
+
+    state_dict = OrderedDict((key_mapping_ln_gamma_beta(k), v) for k, v in state_dict.items())
+
+    # Layers
+    def key_mapping_layers(key):
+        return re.sub(r"^bert.encoder.layer\.", "bert.encoder.layers.", key)
+
+    state_dict = OrderedDict((key_mapping_layers(k), v) for k, v in state_dict.items())
+
+    # LayerNorm
+    def key_mapping_ln(key):
+        key = re.sub(r"^bert.embeddings.LayerNorm.", "bert.emb_ln.", key)
+        key = re.sub(
+            r"^bert.encoder.layers.(\d+).attention.output.LayerNorm.(weight|bias)",
+            r"bert.encoder.layers.\1.norm1.\2",
+            key,
+        )
+        key = re.sub(
+            r"^bert.encoder.layers.(\d+).output.LayerNorm.(weight|bias)",
+            r"bert.encoder.layers.\1.norm2.\2",
+            key,
+        )
+        key = re.sub(
+            r"^cls.predictions.transform.LayerNorm.(weight|bias)",
+            r"cls.predictions.transform.layer_norm.\1",
+            key,
+        )
+        return key
+
+    state_dict = OrderedDict((key_mapping_ln(k), v) for k, v in state_dict.items())
+
+    # MLP
+    def key_mapping_mlp(key):
+        key = re.sub(
+            r"^bert.encoder.layers.(\d+).intermediate.dense.(weight|bias)",
+            r"bert.encoder.layers.\1.mlp.fc1.\2",
+            key,
+        )
+        key = re.sub(
+            r"^bert.encoder.layers.(\d+).output.dense.(weight|bias)",
+            r"bert.encoder.layers.\1.mlp.fc2.\2",
+            key,
+        )
+        return key
+
+    state_dict = OrderedDict((key_mapping_mlp(k), v) for k, v in state_dict.items())
+
+    # Attention
+    last_layer_subset = getattr(config, "last_layer_subset", False)
+    for d in range(config.num_hidden_layers):
+        if f"bert.encoder.layers.{d}.attention.self.query.weight" not in state_dict:
+            continue
+        Wq = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.query.weight")
+        Wk = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.key.weight")
+        Wv = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.value.weight")
+        bq = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.query.bias")
+        bk = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.key.bias")
+        bv = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.value.bias")
+        if not (last_layer_subset and d == config.num_hidden_layers - 1):
+            state_dict[f"bert.encoder.layers.{d}.attn.Wqkv.weight"] = torch.cat([Wq, Wk, Wv], dim=0)
+            state_dict[f"bert.encoder.layers.{d}.attn.Wqkv.bias"] = torch.cat([bq, bk, bv], dim=0)
+        else:
+            state_dict[f"bert.encoder.layers.{d}.attn.Wq.weight"] = Wq
+            state_dict[f"bert.encoder.layers.{d}.attn.Wkv.weight"] = torch.cat([Wk, Wv], dim=0)
+            state_dict[f"bert.encoder.layers.{d}.attn.Wq.bias"] = bq
+            state_dict[f"bert.encoder.layers.{d}.attn.Wkv.bias"] = torch.cat([bk, bv], dim=0)
+
+    def key_mapping_attn(key):
+        return re.sub(
+            r"^bert.encoder.layers.(\d+).attention.output.dense.(weight|bias)",
+            r"bert.encoder.layers.\1.attn.out_proj.\2",
+            key,
+        )
+
+    state_dict = OrderedDict((key_mapping_attn(k), v) for k, v in state_dict.items())
+
+    def key_mapping_decoder_bias(key):
+        return re.sub(r"^cls.predictions.bias", "cls.predictions.decoder.bias", key)
+
+    # remove nsp weights, we don't use
+    state_dict.pop("cls.seq_relationship.weight", None)
+    state_dict.pop("cls.seq_relationship.bias", None)
+    state_dict.pop("bert.embeddings.position_ids", None)
+
+    state_dict = OrderedDict((key_mapping_decoder_bias(k), v) for k, v in state_dict.items())
+
+    if remove_cls_weights:
+        cls_weights = [
+            "cls.predictions.decoder.bias",
+            "cls.predictions.transform.dense.weight",
+            "cls.predictions.transform.dense.bias",
+            "cls.predictions.transform.layer_norm.weight",
+            "cls.predictions.transform.layer_norm.bias",
+            "cls.predictions.decoder.weight",
+        ]
+        for weight in cls_weights:
+            state_dict.pop(weight, None)
+
+    # Word embedding
+    pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
+    if pad_vocab_size_multiple > 1:
+        word_embeddings = state_dict["bert.embeddings.word_embeddings.weight"]
+        state_dict["bert.embeddings.word_embeddings.weight"] = F.pad(
+            word_embeddings, (0, 0, 0, config.vocab_size - word_embeddings.shape[0])
+        )
+        if not remove_cls_weights:
+            decoder_weight = state_dict["cls.predictions.decoder.weight"]
+            state_dict["cls.predictions.decoder.weight"] = F.pad(
+                decoder_weight, (0, 0, 0, config.vocab_size - decoder_weight.shape[0])
+            )
+            # If the vocab was padded, we want to set the decoder bias for those padded indices to be
+            # strongly negative (i.e. the decoder shouldn't predict those indices).
+            # TD [2022-05-09]: I don't think it affects the MLPerf training.
+            if "cls.predictions.decoder.bias" in state_dict:
+                decoder_bias = state_dict["cls.predictions.decoder.bias"]
+                state_dict["cls.predictions.decoder.bias"] = F.pad(
+                    decoder_bias, (0, config.vocab_size - decoder_bias.shape[0]), value=-100.0
+                )
+
+    if add_pooling_layer is False:
+        pooler_weights = [
+            "bert.pooler.dense.weight",
+            "bert.pooler.dense.bias",
+        ]
+        for key in pooler_weights:
+            state_dict.pop(key, None)
+
+    if remove_bert:
+
+        def remove_bert_prefix(key):
+            key = re.sub(r"^bert.", "", key)
+            return key
+
+        state_dict = OrderedDict((remove_bert_prefix(k), v) for k, v in state_dict.items())
+
+    return state_dict
+
+
+class NomicBertPreTrainedModel(PreTrainedModel):
+    """An abstract class to handle weights initialization and
+    a simple interface for dowloading and loading pretrained models.
+    """
+
+    config_class = NomicBertConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Block"]
+    _skip_keys_device_placement = "past_key_values"
+
+    def __init__(self, config, *inputs, **kwargs):
+        super().__init__(config)
+        if not isinstance(config, GPT2Config):
+            raise ValueError(
+                "Parameter config in `{}(config)` should be an instance of class `GPT2Config`. "
+                "To create a model from a Google pretrained model use "
+                "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format(
+                    self.__class__.__name__, self.__class__.__name__
+                )
+            )
+        self.config = config
+
+    @classmethod
+    def from_pretrained(cls, model_name, config=None, *inputs, **kwargs):
+        """
+        Instantiate a NomicBertPreTrainedModel from a pre-trained model file or a pytorch state dict.
+        Download and cache the pre-trained model file if needed.
+        Params:
+            pretrained_model_name_or_path: either:
+                - a path or url to a pretrained model archive containing:
+                    . `bert_config.json` a configuration file for the model
+                    . `pytorch_model.bin` a PyTorch dump of a NomicBertForPretraining instance
+                - a path or url to a pretrained model archive containing:
+                    . `bert_config.json` a configuration file for the model
+                    . `model.chkpt` a TensorFlow checkpoint
+            *inputs, **kwargs: additional input for the specific NomicBert class
+                (ex: num_labels for NomicBertForSequenceClassification)
+        """
+        # Instantiate model.
+        if config is None:
+            config = cls.config_class.from_pretrained(model_name)
+        remove_cls = cls != NomicBertForPreTraining
+        remove_bert_prefix = cls != NomicBertForPreTraining
+        ignore_mismatched_shapes = kwargs.pop("ignore_mismatched_sizes", False)
+        num_labels = kwargs.pop("num_labels", None)
+        rotary_scaling_factor = kwargs.pop("rotary_scaling_factor", None)
+        if rotary_scaling_factor:
+            config.rotary_scaling_factor = rotary_scaling_factor
+
+        if config.n_positions <= 0 and config.rotary_emb_fraction > 0:
+            config.n_positions = 2048
+        if num_labels:
+            config.num_labels = num_labels
+
+        if "add_pooling_layer" in kwargs:
+            model = cls(config, *inputs, add_pooling_layer=kwargs.pop("add_pooling_layer"))
+        else:
+            if cls == NomicBertModel:
+                model = cls(config, *inputs, add_pooling_layer=False)
+            else:
+                model = cls(config, *inputs)
+        # TODO: fix this
+        # Assuming we know what we're doing when loading from disk
+        # Prob a bad assumption but i'm tired and want to train this asap
+        if os.path.exists(model_name):
+            model_path = f"{model_name}/pytorch_model.bin"
+            if os.path.exists(model_path):
+                state_dict = torch.load(f"{model_name}/pytorch_model.bin")
+            else:
+                model_path = f"{model_name}/model.safetensors"
+                if not os.path.exists(model_path):
+                    raise ValueError(f"Model path {model_path} not found")
+                state_dict = safe_load_file(model_path)
+
+            if ignore_mismatched_shapes:
+                state_dict = filter_shapes(state_dict, model)
+            load_return = model.load_state_dict(state_dict, strict=False)
+        else:
+            # TODO: can probably check config class and see if we need to remap from a bert model
+            state_dict = state_dict_from_pretrained(model_name, safe_serialization=kwargs.get("safe_serialization", False))
+            state_dict = remap_bert_state_dict(
+                state_dict,
+                config,
+                remove_bert=remove_bert_prefix,
+                remove_cls_weights=remove_cls,
+                add_pooling_layer=getattr(config, "add_pooling_layer", False),
+            )
+            if ignore_mismatched_shapes:
+                state_dict = filter_shapes(state_dict, model)
+
+            load_return = model.load_state_dict(state_dict, strict=True)
+        logger.warning(load_return)
+        return model
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, NomicBertEncoder):
+            module.gradient_checkpointing = value
+
+
+# https://github.com/huggingface/transformers/blob/7032e0203262ebb2ebf55da8d2e01f873973e835/src/transformers/models/bert/modeling_bert.py#L748
+def _init_weights(module, initializer_range=0.02):
+    if isinstance(module, nn.Linear):
+        nn.init.normal_(module.weight, std=initializer_range)
+        if module.bias is not None:
+            nn.init.zeros_(module.bias)
+    elif isinstance(module, nn.Embedding):
+        nn.init.normal_(module.weight, std=initializer_range)
+        if module.padding_idx is not None:
+            nn.init.zeros_(module.weight[module.padding_idx])
+
+
+class NomicBertEmbeddings(nn.Module):
+    def __init__(self, config):
+        """
+        If max_position_embeddings <= 0, there's no position embeddings
+        If type_vocab_size <= 0, there's no token type embeddings
+        """
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.max_position_embeddings = config.max_position_embeddings if config.rotary_emb_fraction <= 0 else 0
+        self.type_vocab_size = config.type_vocab_size
+        if self.max_position_embeddings > 0 and config.rotary_emb_fraction <= 0:
+            self.position_embeddings = nn.Embedding(
+                config.max_position_embeddings,
+                config.hidden_size,
+            )
+        if self.type_vocab_size > 0:
+            self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
+
+    def forward(self, input_ids, position_ids=None, token_type_ids=None):
+        """
+        input_ids: (batch, seqlen)
+        position_ids: (batch, seqlen)
+        token_type_ids: (batch, seqlen)
+        """
+        batch_size, seqlen = input_ids.shape
+        embeddings = self.word_embeddings(input_ids)
+
+        if self.type_vocab_size > 0:
+            if token_type_ids is None:
+                token_type_ids = torch.zeros(seqlen, dtype=torch.long, device=input_ids.device)
+            token_type_embeddings = self.token_type_embeddings(token_type_ids)
+            embeddings = embeddings + token_type_embeddings
+
+        if self.max_position_embeddings > 0:
+            if position_ids is None:
+                position_ids = torch.arange(seqlen, dtype=torch.long, device=input_ids.device)
+            position_embeddings = self.position_embeddings(position_ids)
+            embeddings = embeddings + position_embeddings
+        return embeddings
+
+
+class NomicBertMLP(nn.Module):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        activation=F.gelu,
+        bias1=True,
+        bias2=True,
+        return_residual=False,
+        fused_bias_fc=False,
+    ):
+        super().__init__()
+        out_features = out_features if out_features is not None else in_features
+        hidden_features = hidden_features if hidden_features is not None else in_features * 4
+        self.return_residual = return_residual
+        self.fc1 = nn.Linear(in_features, hidden_features, bias=bias1)
+        approximate = "tanh" if activation in ["gelu_new", "gelu_fast", "gelu_pytorch_tanh"] else "none"
+        self.activation = nn.GELU(approximate=approximate) if activation == "gelu" else activation
+        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias2)
+
+    def forward(self, x):
+        y = self.fc1(x)
+        y = self.activation(y)
+        y = self.fc2(y)
+        return y if not self.return_residual else (y, x)
+
+
+class NomciBertGatedMLP(nn.Module):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        activation=F.sigmoid,
+        bias1=True,
+        bias2=True,
+        multiple_of=256,
+        return_residual=False,
+        fused_bias_fc=True,
+        device=None,
+        dtype=None,
+    ):
+        super().__init__()
+        out_features = out_features if out_features is not None else in_features
+        hidden_features = hidden_features if hidden_features is not None else int(8 * in_features / 3)
+        hidden_features = (hidden_features + multiple_of - 1) // multiple_of * multiple_of
+        self.return_residual = return_residual
+
+        self.fc11 = nn.Linear(in_features, hidden_features, bias=bias1)
+        self.fc12 = nn.Linear(in_features, hidden_features, bias=bias1)
+        self.activation = activation
+        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias2)
+
+    def forward(self, x):
+        y = self.fc11(x)
+        gate = self.fc12(x)
+        if self.activation == F.sigmoid:  # Special case for GLU
+            y = F.glu(torch.cat([y, gate], dim=-1), dim=-1)
+        else:
+            y = y * self.activation(gate)
+        y = self.fc2(y)
+        return y if not self.return_residual else (y, x)
+
+
+def rotate_half(x, interleaved=False):
+    if not interleaved:
+        x1, x2 = x.chunk(2, dim=-1)
+        return torch.cat((-x2, x1), dim=-1)
+    else:
+        x1, x2 = x[..., ::2], x[..., 1::2]
+        return rearrange(torch.stack((-x2, x1), dim=-1), "... d two -> ... (d two)", two=2)
+
+
+def apply_rotary_emb(x, cos, sin, offset=0, interleaved=False):
+    """
+    x: (batch_size, seqlen, nheads, headdim)
+    cos, sin: (seqlen, rotary_dim / 2) or (batch_size, seqlen, rotary_dim / 2)
+    """
+    ro_dim = cos.shape[-1] * 2
+    assert ro_dim <= x.shape[-1]
+    cos, sin = (
+        cos[offset : offset + x.shape[1]],
+        sin[offset : offset + x.shape[1]],
+    )
+    cos = repeat(cos, "... d -> ... 1 (2 d)" if not interleaved else "... d -> ... 1 (d 2)")
+    sin = repeat(sin, "... d -> ... 1 (2 d)" if not interleaved else "... d -> ... 1 (d 2)")
+    return torch.cat(
+        [x[..., :ro_dim] * cos + rotate_half(x[..., :ro_dim], interleaved) * sin, x[..., ro_dim:]],
+        dim=-1,
+    )
+
+
+class NomicBertRotaryEmbedding(nn.Module):
+    def __init__(
+        self,
+        dim: int,
+        base=10000.0,
+        interleaved=False,
+        scale_base=None,
+        pos_idx_in_fp32=True,
+        device=None,
+    ):
+        """
+        interleaved: if True, rotate pairs of even and odd dimensions (GPT-J style) instead
+            of 1st half and 2nd half (GPT-NeoX style).
+        pos_idx_in_fp32: if True, the position indices [0.0, ..., seqlen - 1] are in fp32,
+            otherwise they might be in lower precision.
+            This option was added because previously (before 2023-07-02), when we construct
+            the position indices, we use the dtype of self.inv_freq. In most cases this would
+            be fp32, but if the model is trained in pure bf16 (not mixed precision), then
+            self.inv_freq would be bf16, and the position indices are also in bf16.
+            Because of the limited precision of bf16 (e.g. 1995.0 is rounded to 2000.0), the
+            embeddings for some positions will coincide.
+            To maintain compatibility with models previously trained in pure bf16,
+            we add this option.
+        """
+        super().__init__()
+        self.dim = dim
+        self.base = float(base)
+        self.pos_idx_in_fp32 = pos_idx_in_fp32
+        # Generate and save the inverse frequency buffer (non trainable)
+        inv_freq = self._compute_inv_freq(device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.interleaved = interleaved
+        self.scale_base = scale_base
+        scale = (
+            (torch.arange(0, dim, 2, device=device, dtype=torch.float32) + 0.4 * dim) / (1.4 * dim)
+            if scale_base is not None
+            else None
+        )
+        self.register_buffer("scale", scale, persistent=False)
+
+        self._seq_len_cached = 0
+        self._cos_cached = None
+        self._sin_cached = None
+        self._cos_k_cached = None
+        self._sin_k_cached = None
+
+    def _compute_inv_freq(self, device=None):
+        return 1.0 / (self.base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))
+
+    def _update_cos_sin_cache(self, seqlen, device=None, dtype=None):
+        # Reset the tables if the sequence length has changed,
+        # if we're on a new device (possibly due to tracing for instance),
+        # or if we're switching from inference mode to training
+        if (
+            seqlen > self._seq_len_cached
+            or self._cos_cached is None
+            or self._cos_cached.device != device
+            or self._cos_cached.dtype != dtype
+            or (self.training and self._cos_cached.is_inference())
+        ):
+            self._seq_len_cached = seqlen
+            # We want fp32 here, not self.inv_freq.dtype, since the model could be loaded in bf16
+            # And the output of arange can be quite large, so bf16 would lose a lot of precision.
+            # However, for compatibility reason, we add an option to use the dtype of self.inv_freq.
+            if self.pos_idx_in_fp32:
+                t = torch.arange(seqlen, device=device, dtype=torch.float32)
+                # We want fp32 here as well since inv_freq will be multiplied with t, and the output
+                # will be large. Having it in bf16 will lose a lot of precision and cause the
+                # cos & sin output to change significantly.
+                # We want to recompute self.inv_freq if it was not loaded in fp32
+                if self.inv_freq.dtype != torch.float32:
+                    inv_freq = self._compute_inv_freq(device=device)
+                else:
+                    inv_freq = self.inv_freq
+            else:
+                t = torch.arange(seqlen, device=device, dtype=self.inv_freq.dtype)
+                inv_freq = self.inv_freq
+            # Don't do einsum, it converts fp32 to fp16 under AMP
+            # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
+            freqs = torch.outer(t, inv_freq)
+            self._cos_cached = torch.cos(freqs).to(dtype)
+            self._sin_cached = torch.sin(freqs).to(dtype)
+
+    def forward(
+        self,
+        qkv: torch.Tensor,
+        kv: Optional[torch.Tensor] = None,
+        seqlen_offset: Union[int, torch.Tensor] = 0,
+        max_seqlen: Optional[int] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        qkv: (batch, seqlen, 3, nheads, headdim) if kv is none,
+             else it's just q of shape (batch, seqlen, nheads, headdim)
+        kv: (batch, seqlen, 2, nheads, headdim)
+        seqlen_offset: (batch_size,) or int. Each sequence in x is shifted by this amount.
+            Most commonly used in inference when we have KV cache.
+            If it's a tensor of shape (batch_size,), then to update the cos / sin cache, one
+            should pass in max_seqlen, which will update the cos / sin cache up to that length.
+        Apply rotary embedding *inplace* to qkv and / or kv.
+        """
+        seqlen = qkv.shape[1]
+        if seqlen > self._seq_len_cached:
+            self._update_cos_sin_cache(seqlen, device=qkv.device, dtype=qkv.dtype)
+        elif max_seqlen is not None:
+            self._update_cos_sin_cache(max_seqlen, device=qkv.device, dtype=qkv.dtype)
+        elif isinstance(seqlen_offset, int):
+            self._update_cos_sin_cache(seqlen + seqlen_offset, device=qkv.device, dtype=qkv.dtype)
+
+        q_rot = apply_rotary_emb(qkv[:, :, 0], self._cos_cached, self._sin_cached, seqlen_offset, self.interleaved)
+        k_rot = apply_rotary_emb(qkv[:, :, 1], self._cos_cached, self._sin_cached, seqlen_offset, self.interleaved)
+        return torch.stack((q_rot, k_rot, qkv[:, :, 2]), dim=2)
+
+
+class NomicBertDynamicNTKRotaryEmbedding(NomicBertRotaryEmbedding):
+    def __init__(self, rotary_scaling_factor, max_position_embeddings, **kwargs):
+        super().__init__(**kwargs)
+        self.rotary_scaling_factor = rotary_scaling_factor
+        self.max_position_embeddings = max_position_embeddings
+
+    def _compute_inv_freq(self, base=None, device=None):
+        if base is None:
+            base = self.base
+        return 1.0 / (base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))
+
+    def _update_cos_sin_cache(self, seqlen, device=None, dtype=None):
+        # Reset the tables if the sequence length has changed,
+        # if we're on a new device (possibly due to tracing for instance),
+        # or if we're switching from inference mode to training
+        if seqlen > self.max_position_embeddings:
+            base = self.base * (
+                (self.rotary_scaling_factor * seqlen / self.max_position_embeddings) - (self.rotary_scaling_factor - 1)
+            ) ** (self.dim / (self.dim - 2))
+            inv_freq = self._compute_inv_freq(base=base, device=device)
+            self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        if (
+            seqlen > self._seq_len_cached
+            or self._cos_cached is None
+            or self._cos_cached.device != device
+            or self._cos_cached.dtype != dtype
+            or (self.training and self._cos_cached.is_inference())
+        ):
+            self._seq_len_cached = seqlen
+            # We want fp32 here, not self.inv_freq.dtype, since the model could be loaded in bf16
+            # And the output of arange can be quite large, so bf16 would lose a lot of precision.
+            # However, for compatibility reason, we add an option to use the dtype of self.inv_freq.
+            if self.pos_idx_in_fp32:
+                t = torch.arange(seqlen, device=device, dtype=torch.float32)
+                # We want fp32 here as well since inv_freq will be multiplied with t, and the output
+                # will be large. Having it in bf16 will lose a lot of precision and cause the
+                # cos & sin output to change significantly.
+                # We want to recompute self.inv_freq if it was not loaded in fp32
+                if self.inv_freq.dtype != torch.float32:
+                    if seqlen > self.max_position_embeddings:
+                        base = self.base * (
+                            (self.scaling_factor * seqlen / self.max_position_embeddings) - (self.scaling_factor - 1)
+                        ) ** (self.dim / (self.dim - 2))
+                    else:
+                        base = self.base
+                    inv_freq = self._compute_inv_freq(device=device, base=base)
+                else:
+                    inv_freq = self.inv_freq
+            else:
+                t = torch.arange(seqlen, device=device, dtype=self.inv_freq.dtype)
+                inv_freq = self.inv_freq
+            # Don't do einsum, it converts fp32 to fp16 under AMP
+            # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
+            freqs = torch.outer(t, inv_freq)
+            if self.scale is None:
+                self._cos_cached = torch.cos(freqs).to(dtype)
+                self._sin_cached = torch.sin(freqs).to(dtype)
+            else:
+                power = (
+                    torch.arange(seqlen, dtype=self.scale.dtype, device=self.scale.device) - seqlen // 2
+                ) / self.scale_base
+                scale = self.scale.to(device=power.device) ** rearrange(power, "s -> s 1")
+                # We want the multiplication by scale to happen in fp32
+                self._cos_cached = (torch.cos(freqs) * scale).to(dtype)
+                self._sin_cached = (torch.sin(freqs) * scale).to(dtype)
+                self._cos_k_cached = (torch.cos(freqs) / scale).to(dtype)
+                self._sin_k_cached = (torch.sin(freqs) / scale).to(dtype)
+
+
+class NomicBertAttention(nn.Module):
+    """Multi-head self-attention and cross-attention"""
+
+    def __init__(
+        self,
+        config,
+    ) -> None:
+        """
+        num_heads_kv: can be used to toggle MQA / GQA. If None, use num_heads.
+        return_residual: whether to return the input x along with the output. This is for
+            performance reason: for post-norm architecture, returning the input allows us
+            to fuse the backward of nn.Linear with the residual connection.
+        """
+        super().__init__()
+        self.embed_dim = config.n_embd
+        self.use_flash_attn = config.use_flash_attn
+        self.fused_bias_fc = config.fused_bias_fc
+
+        self.num_heads = config.n_head
+        self.num_heads_kv = config.num_heads_kv if getattr(config, "num_heads_kv", None) is not None else self.num_heads
+        assert self.embed_dim % self.num_heads == 0, "embed_dim must be divisible by num_heads"
+        self.head_dim = self.embed_dim // self.num_heads
+        # we don't really support mqa / gqa for now
+        qkv_dim = self.head_dim * (self.num_heads + 2 * self.num_heads_kv)
+
+        self.register_buffer(
+            "norm_factor",
+            torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype()),
+            persistent=False,
+        )
+
+        self.rotary_emb_dim = self.head_dim * config.rotary_emb_fraction
+        if self.rotary_emb_dim > 0:
+            if config.rotary_scaling_factor:
+                self.rotary_emb = NomicBertDynamicNTKRotaryEmbedding(
+                    dim=self.rotary_emb_dim,
+                    base=config.rotary_emb_base,
+                    scale_base=config.rotary_emb_scale_base,
+                    interleaved=config.rotary_emb_interleaved,
+                    rotary_scaling_factor=config.rotary_scaling_factor,
+                    max_position_embeddings=config.max_trained_positions,
+                )
+            else:
+                self.rotary_emb = NomicBertRotaryEmbedding(
+                    dim=self.rotary_emb_dim,
+                    base=config.rotary_emb_base,
+                    scale_base=config.rotary_emb_scale_base,
+                    interleaved=config.rotary_emb_interleaved,
+                )
+            # bug in xformers: https://github.com/facebookresearch/xformers/issues/841
+            # uses the head dimension instead of the sequence dimension
+            self.rotary_head_dim = getattr(config, "rotary_head_dim", False)
+
+        self.Wqkv = nn.Linear(self.embed_dim, qkv_dim, bias=config.qkv_proj_bias)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.qkv_proj_bias)
+        self.causal = config.causal
+        self.drop = nn.Dropout(config.attn_pdrop)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        is_padded_inputs: Optional[bool] = True,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        max_seq_len: Optional[int] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+
+        has_layer_past = past_key_value is not None
+
+        if has_layer_past:
+            past_key_value = past_key_value[0]
+            past_len = past_key_value[1]
+        else:
+            past_len = 0
+
+        qkv = self.Wqkv(hidden_states)
+        qkv = rearrange(qkv, "... (three h d) -> ... three h d", three=3, d=self.head_dim)
+
+        past_key_value = (past_key_value, past_len + qkv.size(1)) if use_cache else None
+
+        if self.rotary_emb_dim > 0:
+            if self.rotary_head_dim:
+                qkv = rearrange(qkv, "b s three h d -> b h three s d")
+            qkv = self.rotary_emb(qkv, seqlen_offset=past_len)
+
+            if self.rotary_head_dim:
+                qkv = rearrange(qkv, "b h three s d -> b s three h d")
+
+        query, key, value = qkv[:, :, 0], qkv[:, :, 1], qkv[:, :, 2]
+
+        query = query.permute(0, 2, 1, 3)
+        key = key.permute(0, 2, 1, 3)
+        value = value.permute(0, 2, 1, 3)
+
+        attention_scores = torch.matmul(query, key.transpose(-1, -2)) / self.norm_factor
+        if attention_mask is not None:
+            attention_scores = attention_scores + attention_mask
+
+        attentions_probs = F.softmax(attention_scores, dim=-1)
+        attentions_probs = self.drop(attentions_probs)
+
+        attn_output = torch.matmul(attentions_probs, value)
+        attn_output = rearrange(attn_output.permute(0, 2, 1, 3), "... h d -> ... (h d)")
+
+        attn_output = self.out_proj(attn_output)
+
+        return attn_output
+
+
+class NomicBertBlock(nn.Module):
+    def __init__(
+        self,
+        config,
+    ):
+        super().__init__()
+        self.prenorm = config.prenorm
+        self.fused_dropout_add_ln = config.fused_dropout_add_ln
+
+        self.attn = NomicBertAttention(config)
+        activation = (
+            F.sigmoid
+            if config.activation_function == "glu"
+            else (F.silu if config.activation_function == "swiglu" else F.gelu)
+        )
+        if config.activation_function in ["glu", "swiglu", "geglu"]:
+            self.mlp = NomciBertGatedMLP(
+                config.n_embd,
+                hidden_features=config.n_inner,
+                bias1=config.mlp_fc1_bias,
+                bias2=config.mlp_fc2_bias,
+                activation=activation,
+                fused_bias_fc=config.fused_bias_fc,
+            )
+        else:
+            self.mlp = NomicBertMLP(
+                config.n_embd,
+                hidden_features=config.n_inner,
+                bias1=config.mlp_fc1_bias,
+                bias2=config.mlp_fc2_bias,
+                activation=activation,
+                fused_bias_fc=config.fused_bias_fc,
+            )
+
+        self.dropout1 = nn.Dropout(config.resid_pdrop)
+        self.norm1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.norm2 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.dropout2 = nn.Dropout(config.resid_pdrop)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        hidden_states2: torch.Tensor,
+        residual: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        is_padded_inputs: Optional[bool] = True,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cu_seqlens: Optional[torch.Tensor] = None,
+        max_seq_len: Optional[int] = None,
+    ):
+        r"""Pass the input through the encoder layer.
+        Args:
+            hidden_states: the sequence to the encoder layer (required).
+            residual: if postnorm, residual=None, If prenorm, hidden_states = Attn/MLP(LN(residual))
+            mixer_subset: for cross-attention only. If not None, will take a subset of x
+                before applying the query projection. Useful for e.g., ViT where we only care
+                about the CLS token in the last layer.
+        """
+        if self.prenorm:
+            dropped = self.dropout1(hidden_states)
+            residual = (dropped + residual) if residual is not None else dropped
+            hidden_states = self.norm1(residual.to(dtype=self.norm1.weight.dtype))
+            hidden_states = self.attn(
+                hidden_states,
+                attention_mask=attention_mask,
+                is_padded_inputs=is_padded_inputs,
+                cu_seqlens=cu_seqlens,
+                max_seq_len=max_seq_len,
+            )
+
+            dropped = self.dropout2(hidden_states)
+            residual = (dropped + residual) if residual is not None else dropped
+            hidden_states = self.norm2(residual.to(dtype=self.norm2.weight.dtype))
+            hidden_states = self.mlp(hidden_states)
+
+            return hidden_states, None, residual
+        else:
+            assert residual is None
+            attn_outputs = self.attn(
+                hidden_states,
+                attention_mask=attention_mask,
+                is_padded_inputs=is_padded_inputs,
+                cu_seqlens=cu_seqlens,
+                max_seq_len=max_seq_len,
+            )
+            hidden_states = self.norm1((self.dropout1(attn_outputs) + hidden_states).to(dtype=self.norm1.weight.dtype))
+            mlp_out = self.mlp(hidden_states)
+
+            hidden_states = self.norm2((self.dropout2(mlp_out) + hidden_states).to(dtype=self.norm2.weight.dtype))
+            return hidden_states, None, None
+
+
+class NomicBertEncoder(nn.Module):
+    def __init__(self, config: GPT2Config):
+        super().__init__()
+        self.layers = nn.ModuleList([NomicBertBlock(config) for _ in range(config.n_layer)])
+        self.gradient_checkpointing = False
+        self.config = config
+
+    def forward(
+        self,
+        hidden_states: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        is_padded_inputs: Optional[bool] = True,
+    ):
+        """If subset_mask is not None, we only want output for the subset of the sequence.
+        This means that we only compute the last layer output for these tokens.
+        subset_mask: (batch, seqlen), dtype=torch.bool
+        """
+        hidden_states2 = None
+        residual = None
+
+        for _, layer in enumerate(self.layers):
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs)
+
+                    return custom_forward
+
+                hidden_states, hidden_states2, residual = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer),
+                    hidden_states,
+                    hidden_states2,
+                    residual,
+                    attention_mask,
+                    None,
+                    None,
+                    is_padded_inputs,
+                    # if you freeze ANY layers, you need `use_reentrant=False`
+                    # https://github.com/huggingface/transformers/issues/21381
+                    # https://discuss.pytorch.org/t/checkpoint-with-no-grad-requiring-inputs-problem/19117/7
+                    use_reentrant=False,
+                )
+
+            else:
+                hidden_states, hidden_states2, residual = layer(
+                    hidden_states,
+                    hidden_states2,
+                    residual,
+                    attention_mask,
+                    position_ids,
+                    None,
+                    is_padded_inputs,
+                    output_attentions,
+                    use_cache,
+                )
+        return hidden_states
+
+
+class NomicBertPooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.n_embd, config.n_embd)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states, pool=True):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0] if pool else hidden_states
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class NomicBertPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.n_embd, config.n_embd, bias=config.mlp_fc1_bias)
+        approximate = "tanh" if config.activation_function in ["gelu_new", "gelu_fast", "gelu_pytorch_tanh"] else "none"
+        if config.activation_function == "swiglu":
+            self.transform_act_fn = F.silu
+        else:
+            self.transform_act_fn = nn.GELU(approximate=approximate)
+
+        self.layer_norm = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.layer_norm(hidden_states)
+
+        return hidden_states
+
+
+class NomicBertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        self.transform = NomicBertPredictionHeadTransform(config)
+
+        self.decoder = nn.Linear(config.n_embd, config.vocab_size, bias=config.mlp_fc1_bias)
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class NomicBertPreTrainingHeads(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = NomicBertLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class NomicBertModel(NomicBertPreTrainedModel):
+    def __init__(self, config: GPT2Config, add_pooling_layer=True):
+        super().__init__(config)
+        self.pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
+        if config.vocab_size % self.pad_vocab_size_multiple != 0:
+            config.vocab_size += self.pad_vocab_size_multiple - (config.vocab_size % self.pad_vocab_size_multiple)
+
+        assert config.activation_function in [
+            "gelu",
+            "gelu_new",
+            "gelu_fast",
+            "gelu_pytorch_tanh",
+            "swiglu",
+            "geglu",
+            "glu",
+        ]
+
+        self.embeddings = NomicBertEmbeddings(config)
+        self.emb_drop = nn.Dropout(config.resid_pdrop)
+        self.emb_ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.encoder = NomicBertEncoder(config)
+        self.pooler = NomicBertPooler(config) if add_pooling_layer else None
+
+        self.apply(partial(_init_weights, initializer_range=config.initializer_range))
+
+    def forward(
+        self,
+        input_ids,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        return_dict=None,
+    ):
+        if token_type_ids is None:
+            token_type_ids = torch.zeros_like(input_ids)
+        hidden_states = self.embeddings(input_ids, position_ids=position_ids, token_type_ids=token_type_ids)
+        hidden_states = self.emb_ln(hidden_states)
+        hidden_states = self.emb_drop(hidden_states)
+
+        # FailureLens: transformers 5 removed get_extended_attention_mask.
+        if attention_mask is not None:
+            if attention_mask.dim() == 2:
+                attention_mask = attention_mask[:, None, None, :].to(dtype=torch.float32)
+                attention_mask = (1.0 - attention_mask) * torch.finfo(torch.float32).min
+            elif attention_mask.dim() != 4:
+                raise ValueError("Expected a 2-D or 4-D attention mask")
+
+        sequence_output = self.encoder(hidden_states, attention_mask=attention_mask, return_dict=return_dict)
+
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+        )
+
+
+class NomicBertForPreTraining(NomicBertPreTrainedModel):
+    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]
+
+    def __init__(self, config: GPT2Config):
+        super().__init__(config)
+
+        self.bert = NomicBertModel(config, add_pooling_layer=getattr(config, "add_pooling_layer", False))
+        self.cls = NomicBertPreTrainingHeads(config)
+        self.mlm_loss = nn.CrossEntropyLoss()
+
+        # Initialize weights and apply final processing
+        self.apply(partial(_init_weights, initializer_range=config.initializer_range))
+        self.tie_weights()
+
+    def tie_weights(self):
+        self.cls.predictions.decoder.weight = self.bert.embeddings.word_embeddings.weight
+
+    def forward(
+        self,
+        input_ids,
+        position_ids=None,
+        token_type_ids=None,
+        attention_mask=None,
+        labels=None,
+    ):
+        """
+        If labels are provided, they must be -100 for masked out tokens (as specified in the attention
+        mask).
+        Outputs:
+            if `labels` and `next_sentence_label` are not `None`:
+                Outputs the total_loss which is the sum of the masked language modeling loss and the next
+                sentence classification loss.
+            if `labels` or `next_sentence_label` is `None`:
+                Outputs a tuple comprising
+                - the masked language modeling logits of shape [batch_size, sequence_length, vocab_size], and
+                - the next sentence classification logits of shape [batch_size, 2].
+        """
+        outputs = self.bert(
+            input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            attention_mask=attention_mask.bool() if attention_mask is not None else None,
+        )
+        sequence_output, _ = outputs.last_hidden_state, outputs.pooler_output
+
+        prediction_scores = self.cls(sequence_output)
+
+        total_loss = None
+        if labels is not None:
+            masked_lm_loss = self.mlm_loss(
+                rearrange(prediction_scores, "... v -> (...) v"),
+                rearrange(labels, "... -> (...)"),
+            )
+            total_loss = masked_lm_loss.float()
+
+        return MaskedLMOutput(
+            loss=total_loss,
+            logits=prediction_scores,
+            hidden_states=outputs.hidden_states,
+            attentions=None,
+        )
+
+
+class NomicBertForSequenceClassification(NomicBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.bert = NomicBertModel(config)
+        classifier_dropout = getattr(config, "classifier_dropout", config.embd_pdrop)
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.n_embd, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        outputs = self.bert(
+            input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            attention_mask=attention_mask.bool() if attention_mask is not None else None,
+        )
+
+        pooled_output = outputs[1]
+
+        pooled_output = self.dropout(pooled_output)
+        logits = self.classifier(pooled_output)
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = nn.MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = nn.CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = nn.BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

modules.json

@@ -0,0 +1,14 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "sentence_transformers.models.Transformer"
+  },
+  {
+    "idx": 1,
+    "name": "1",
+    "path": "1_Pooling",
+    "type": "sentence_transformers.models.Pooling"
+  }
+]

sentence_bert_config.json

@@ -0,0 +1,8 @@
+{
+  "max_seq_length": 8192,
+  "do_lower_case": false,
+  "model_args": {
+    "add_pooling_layer": false,
+    "safe_serialization": true
+  }
+}

special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,63 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "mask_token": "[MASK]",
+  "max_length": 512,
+  "model_max_length": 8192,
+  "pad_to_multiple_of": null,
+  "pad_token": "[PAD]",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}