--- name: parity-testing-bc3fcd684572 description: Structured framework for exact HF↔MCore weight verification, forward-pass logit correlation, and optional strict numerical diagnostics. Use when debugging weight mismatches, verifying HF↔MCore checkpoint round-trips, choosing verification tools, or investigating conversion commits that caused parity failures. References existing tools and the add-model-support skill. --- # Parity Testing for Megatron Bridge This skill provides the decision framework for choosing the right verification tool and interpreting results. For the full model onboarding workflow (which includes parity testing as milestones 1 and 2), see the `add-model-support` skill. ## Quick Decision: Which Tool to Run | What you want to verify | Tool | GPU? | When to use | |---|---|---|---| | All weights round-trip exactly (single GPU) | `hf_megatron_roundtrip.py` | No | First check after writing a bridge | | Weights round-trip with TP/PP/EP | `hf_megatron_roundtrip_multi_gpu.py` | Yes | After single-GPU passes | | Forward-pass logit correlation | `compare_hf_and_megatron/compare.py` | Yes | After round-trip passes | | Text generation sanity | `hf_to_megatron_generate_text.py` | Yes | Separate inference evidence | | Programmatic weight check | `weights_verification_table()` | Yes | Inside Python scripts | | VLM generation sanity | `hf_to_megatron_generate_vlm.py` | Yes | VLM models | All tools live under `examples/conversion/`. ## 3-Level Test Strategy ### Level 1: State Dict Round-Trip (exact match) Check tensor coverage and mapping behavior before interpreting forward results. Pure layout transformations must round-trip weights exactly. Investigate arithmetic parameter transformations separately from missing or misplaced weights. ```bash # Single-GPU round-trip uv run python examples/conversion/hf_megatron_roundtrip.py \ --hf-model-id / # Multi-GPU with TP=2 uv run python -m torch.distributed.run --nproc_per_node=2 \ examples/conversion/hf_megatron_roundtrip_multi_gpu.py \ --hf-model-id / --tp 2 # Multi-GPU with PP=2 uv run python -m torch.distributed.run --nproc_per_node=2 \ examples/conversion/hf_megatron_roundtrip_multi_gpu.py \ --hf-model-id / --pp 2 ``` **Expected:** Every weight mapped by pure layout transformations shows "Matches Original: checkmark". Investigate every "X" before assigning a cause. **Tolerance:** Exact match (`max_diff == 0.0`) for pure reshaping, permutation, and sharding. If a mapping performs arithmetic or changes dtype, report that transformation and its measured effect explicitly. Use the controlled module-swap procedure below before classifying differences as numerical; do not silently relax the exact-match check. For programmatic verification inside scripts, use the built-in verifier: ```python from megatron.bridge.models.conversion.utils import weights_verification_table weights_verification_table(bridge, hf_pretrained, megatron_model) ``` ### Level 2: Forward-Pass Correlation (GPU / bfloat16) After round-trip passes, verify that the converted model produces strongly correlated logits and the same next-token prediction. This is a functional correlation gate, not a claim of bitwise-identical floating-point arithmetic. ```bash # Compare logits (loads both HF and Megatron models) uv run python -m torch.distributed.run --nproc_per_node=2 \ examples/conversion/compare_hf_and_megatron/compare.py \ --hf_model_path / --tp 2 \ --prompt "The capital of France is" ``` **Expected:** Matching next-token predictions and cosine similarity at least 0.99. Equivalently, cosine distance (`1 - cosine_similarity`) must be at most 1%. Record numeric maximum and mean absolute logit differences for diagnosis, but do not use either absolute difference as a pass/fail guard. A fixed absolute threshold is not scale- or ULP-aware across BF16 logits and implementations. For large models that OOM `compare.py` (which loads both models), use text generation as a separate inference sanity check: ```bash uv run python -m torch.distributed.run --nproc_per_node=2 \ examples/conversion/hf_to_megatron_generate_text.py \ --hf_model_path / --tp 2 \ --prompt "The capital of France is" --max_new_tokens 50 ``` Generation does not measure logit correlation and therefore cannot by itself verify the manual forward-pass item. ### Level 3: Training Parity (optional) Verify that a few training steps produce decreasing loss. This catches gradient computation issues that forward-pass tests miss. Use a toy model with 2 layers and small dimensions. See the functional test pattern in the `add-model-support` skill (Milestone 3, Phase 6). ## Tolerance Table | Test Level | Dtype | Device | Required Gate | Report Only | |---|---|---|---|---| | Round-trip, pure layout mappings | float32 | CPU | `max_diff == 0.0` | cosine similarity | | Forward pass | bfloat16 | GPU | next token matches and cosine `>= 0.99` | max/mean absolute logit difference | | Forward pass | float16 | GPU | next token matches and cosine `>= 0.99` | max/mean absolute logit difference | Keep stricter cosine or absolute-difference thresholds as optional diagnostic targets when investigating arithmetic differences. Do not use them to downgrade a model-support card after the correlation gate passes. ### Transformer Engine normalization Using Transformer Engine LayerNorm/RMSNorm can change floating-point operation ordering relative to the HF or PyTorch reference. Treat that as a forward-correlation question, not a weight-conversion failure: unchanged parameter representations must still round-trip exactly; investigate arithmetic parameter transformations as described above. Forward parity uses the cosine and next-token gates above. Do not replace an available TE affine norm solely to reduce a report-only absolute logit difference. Conversely, do not add affine parameters to weightless reference norms just to use TE; that changes the architecture and invalidates exact mapping coverage. ## Comparison Utilities These functions are useful when writing custom verification scripts or debugging failures. They are not part of the Bridge library — copy them into your script as needed. ```python import torch def compare_tensors(a, b, name=""): """Compare two tensors and report similarity metrics.""" max_diff = (a - b).abs().max().item() mean_diff = (a - b).abs().mean().item() cos_sim = torch.nn.functional.cosine_similarity( a.flatten().float(), b.flatten().float(), dim=0, ).item() print(f"{name}: max_diff={max_diff:.6e}, mean_diff={mean_diff:.6e}, cosine_sim={cos_sim:.8f}") return max_diff, mean_diff, cos_sim def compare_state_dicts(sd_a, sd_b, prefix=""): """Compare two state dicts key-by-key, reporting per-parameter differences.""" keys_a, keys_b = set(sd_a.keys()), set(sd_b.keys()) missing, extra = keys_a - keys_b, keys_b - keys_a if missing: print(f"{prefix}Missing keys: {sorted(missing)}") if extra: print(f"{prefix}Extra keys: {sorted(extra)}") max_diffs = {} for key in sorted(keys_a & keys_b): diff = (sd_a[key].float() - sd_b[key].float()).abs().max().item() if diff > 0: max_diffs[key] = diff print(f"{prefix}{key}: max_diff={diff:.6e}") if not max_diffs and not missing and not extra: print(f"{prefix}All {len(keys_a & keys_b)} parameters match exactly.") return missing, extra, max_diffs ``` ## Debugging Workflow When a parity test fails, follow this sequence: Before classifying any discrepancy as numerical, follow the controlled module-swap procedure in `skills/adding-model-support/SKILL.md`. Magnitude, BF16 precision, or a suspected normalization kernel is not sufficient evidence. Fix confirmed bugs; preserve reusable production blocks for confirmed numerical differences and retain the diagnostic evidence. 1. **Run single-GPU round-trip** — inspect the `mapping_registry()`, layouts, dtypes, and arithmetic parameter transformations. Missing tensors or wrong shapes indicate conversion defects; investigate value differences before assigning a cause. 2. **If single-GPU passes but multi-GPU fails** — inspect TP/PP/EP distribution, gathering, and checkpoint metadata. Compare the TP=1 result against each TP shard. See the `nccl-contiguous-tensors` skill for NCCL-specific issues. 3. **If round-trip passes but forward correlation fails** — the tested roundtrip recovered the source tensors. Verify runtime parameter placement and effective values; inverse mapping errors can cancel in a roundtrip. Record cosine similarity, absolute differences, token decisions, and continuations; use task-specific acceptance criteria. A high cosine alone does not explain a changed token or establish numerical equivalence. Check the bridge's config mapping in `hf_config_to_model_config_kwargs()`, or in `provider_bridge()` for families that have not migrated (normalization, activation, RoPE, etc.). 4. **Use the debugging script template** from the `add-model-support` skill to inspect runtime vs safetensors key naming and bridge config mapping. For the full catalog of pitfalls (QKV interleaving, MoE fused exports, tied embeddings, FP8 dequantization, TE LayerNorm aliases, etc.), see the Pitfalls section of the `add-model-support` skill. ## Code Anchors | Component | Path | |---|---| | Single-GPU round-trip | `examples/conversion/hf_megatron_roundtrip.py` | | Multi-GPU round-trip | `examples/conversion/hf_megatron_roundtrip_multi_gpu.py` | | Forward-pass comparison | `examples/conversion/compare_hf_and_megatron/compare.py` | | Text generation | `examples/conversion/hf_to_megatron_generate_text.py` | | VLM generation | `examples/conversion/hf_to_megatron_generate_vlm.py` | | Checkpoint CLI | `scripts/conversion/convert.sh` | | Toy model creator | `examples/conversion/create_hf_toy_model.py` | | Verification utility | `src/megatron/bridge/models/conversion/utils.py` | | Adapter verification | `examples/conversion/adapter/verify_adapter.py` | --- Source attribution: NVIDIA-NeMo Original project: https://github.com/NVIDIA-NeMo/Megatron-Bridge/blob/58382ddf9201475cf63c4aad535620b958ba55aa/skills/parity-testing/SKILL.md License: Apache-2.0 License information: https://www.apache.org/licenses/LICENSE-2.0 Original upstream instructions; downloaded from Mega AI Hub. ## Upstream license notice 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. # Mega AI Hub resource metadata Creator / publisher: NVIDIA-NeMo Original project: https://github.com/NVIDIA-NeMo/Megatron-Bridge/blob/58382ddf9201475cf63c4aad535620b958ba55aa/skills/parity-testing/SKILL.md Attribution basis: Source repository record Fields of work: Development & Engineering Roles: ML / AI Engineer, Data Engineer, Backend Developer Specialties: Backend Tools: Torch Country applicability: Country not specified Classification basis: Source category, role and specialization labels Instruction SHA-256: 80aa6c067a8c51dc8716325388923b9cc30f003cbfcfe4e6bb19b328764b346f