Add masked language modeling models

[shujaatTracebloc]

Summary

• Add masked_language_modeling/pytorch/ directory with three model definitions for biomedical graph MLM pretraining
• simple_mlm.py — ~30M param, 4-layer transformer encoder for smoke testing and pipeline validation
• netmedgpt_style_scratch.py — ~110M param, 12-layer BERT-scale encoder for training from scratch on domain corpora (e.g. PrimeKG random walks)
• netmedgpt_style_warmstart.py — BERT-base warm-started from HuggingFace pretrained weights, embedding layer resized for custom tokenizer vocab
• Update test_model_contract.py to include masked_language_modeling in known categories

Test plan

• pytest tests/test_model_contract.py passes with new category and all three model files
• Verify simple_mlm forward pass produces correct output shape (batch, seq_len, vocab_size)
• Verify netmedgpt_style_scratch forward pass produces correct output shape
• Verify netmedgpt_style_warmstart loads pretrained weights and resizes embeddings

🤖 Generated with Claude Code

---

Note

Medium Risk
Adds new PyTorch MLM model entrypoints, including one that imports HuggingFace transformers, which may break model-contract imports if that dependency isn’t present in relevant CI/runtime environments.

Overview
Introduces a new masked_language_modeling task in the model zoo with three PyTorch model definitions: a small transformer encoder (SimpleMaskedLM), a BERT-base-scale from-scratch encoder with tied LM head (NetMedGPTScratch), and a warm-started HuggingFace BERT loader that resizes token embeddings (NetMedGPTWarmStart).

Updates the model contract test to recognize masked_language_modeling as a valid category so these new model files are included in the import/metadata checks.

^{Reviewed by Cursor Bugbot for commit 084cb95. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

👋 Heads-up — Code review queue is at 18 / 8

Above the WIP limit. The team convention is to review existing PRs before opening new work.

Open PRs currently in Code review (oldest first):

• .github#41 — fix(profile): point install snippets to canonical i.sh / i.ps1 URLs · author: @saadqbal · reviewer: @LukasWodka
• client#100 — feat(client): scaffolding for stateless requests-proxy auth (HC-1) · author: @saadqbal · no reviewer assigned
• client-runtime#35 — feat(Develop #21): jobs-manager submit-ingestion-run HTTP endpoint · author: @saadqbal · no reviewer assigned
• client-runtime#36 — Bump requests from 2.32.3 to 2.33.0 in /Node-deploy · author: @dependabot · no reviewer assigned
• client-runtime#37 — Bump black from 25.1.0 to 26.3.1 in /Node-deploy · author: @dependabot · no reviewer assigned
• docs#3 — Block bots from crawling Mintlify static assets · author: @LukasWodka · reviewer: @saadqbal
• frontend-app#469 — chore(deps): bump postcss from 8.5.6 to 8.5.10 · author: @dependabot · no reviewer assigned
• frontend-app#470 — chore(deps): bump dompurify from 3.3.1 to 3.4.0 · author: @dependabot · no reviewer assigned
• frontend-app#471 — chore(deps): bump @babel/plugin-transform-modules-systemjs from 7.27.1 to 7.29.4 · author: @dependabot · no reviewer assigned
• model-zoo#67 — Add 10 trending model architectures across all task families · author: @divyasinghds · no reviewer assigned

Pull from review before opening new work. (This is a nudge from the kanban WIP check, not a block.)

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6576d4b. Configure here.}