Skip to content

Apertus1p5 support - #5930

Closed
theely wants to merge 85 commits into
NVIDIA-NeMo:mainfrom
theely:apertus1p5-support
Closed

Apertus1p5 support#5930
theely wants to merge 85 commits into
NVIDIA-NeMo:mainfrom
theely:apertus1p5-support

Conversation

@theely

@theely theely commented Sep 1, 2026

Copy link
Copy Markdown

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Changelog

  • Add specific line by line info of high level changes in this PR.

GitHub Actions CI

See the CI section in the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

Before your PR is "Ready for review"

Pre checks:

  • Make sure you read and followed Contributor guidelines
  • Did you write any new necessary tests?
  • Did you add or update any necessary documentation?
  • Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • Reviewer: Does the PR have correct import guards for all optional libraries?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

  • Related to # (issue)

github-actions Bot and others added 30 commits July 28, 2026 06:25
@github-actions
[🤖]: Update docs-versions after code-freeze for r0.6.0
e7540cb
@github-actions
Bump docs version to r0.6.0
2b56a25
@svcnvidia-nemo-ci @kevalmorabia97
cp: `Add SequentialMLP mappings for MTP heads in nemotron_h_bridge.py…
6c0ad53 
… (5140)` into `r0.6.0` (NVIDIA-NeMo#5152)

Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: [test] fix: repair Nemotron-H CI regressions (5143) into `r0.6.…
0720b49 
…0` (NVIDIA-NeMo#5146)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @cuichenx
cp: Nemotron 3.5 Nano verification card and final-checkpoint refresh (N…
08acf56 
…VIDIA-NeMo#5124, NVIDIA-NeMo#5199) into r0.6.0 (NVIDIA-NeMo#5178)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @balasaajay
cp: `fix(docker): update dependencies in Dockerfile and fw_pyproject.…
ff37951 
…toml for compatibility and security (5202)` into `r0.6.0` (NVIDIA-NeMo#5208)

Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ajay <abalasa@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: fix(model): map GLM-5 routed expert count (5220) into r0.6.0 (N…
9544c93 
…VIDIA-NeMo#5248)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @weijiac0619 @claude
cp: feat(sft): DSv4 Flash offline-packed SFT with THD + CP (5126) i…
c2b85a3 
…nto `r0.6.0` (NVIDIA-NeMo#5247)

Signed-off-by: weijia chen <weijiac@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: weijiac0619 <weijiac@nvidia.com>
Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
@svcnvidia-nemo-ci @zhongbozhu
cp: [Doc] Document latest Mcore toggle moe_hybridep_pad_uneven_disp…
359609f 
…atch_inputs` for HybridEP THD training (4213)` into `r0.6.0` (NVIDIA-NeMo#5159)

Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Zhongbo Zhu <42691305+zhongbozhu@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(perf): port r0.5.0 VR200 performance configs (5196) into `r…
aad7177 
…0.6.0` (NVIDIA-NeMo#5197)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @minitu
cp: feat(misc): add memory snapshot analysis skill (5150) into `r0.…
9472ca2 
…6.0` (NVIDIA-NeMo#5198)

Signed-off-by: Jaemin Choi <jaeminc@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Jaemin Choi <jaeminc@nvidia.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(model): add Gemma 3 VL verification coverage (5084) into `r…
a1e812e 
…0.6.0` (NVIDIA-NeMo#5204)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(training): preserve alltoall flex fallback (5207) into `r0.…
77e25d2 
…6.0` (NVIDIA-NeMo#5214)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: `fix(model): align MiniMax evidence and inference task routing (5…
4c10b8e 
…194)` into `r0.6.0` (NVIDIA-NeMo#5235)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(peft): sync grouped expert LoRA over expert DP (5231) into …
44f5668 
…`r0.6.0` (NVIDIA-NeMo#5269)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @balasaajay
cp: chore(deps): update mlflow version to 3.15.0 (5267) into `r0.6.…
af2d5b4 
…0` (NVIDIA-NeMo#5275)

Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ajay <abalasa@nvidia.com>
@weijiac0619
chore: bump mcore dev commit for DSv4 support (NVIDIA-NeMo#5284)
a2fec25 
Signed-off-by: Weijia Chen <weijiac@nvidia.com>
@cuichenx
cp: fix(recipes): derive pretraining vocab from tokenizer (4996) in…
d4bf3f1 
…to `r0.6.0` (NVIDIA-NeMo#5280)

Signed-off-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(data): preserve VLM pixel defaults (5218) into r0.6.0 (NV…
a91e2ab 
…IDIA-NeMo#5271)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(data): reject truncation without supervision (5217) into `r…
3626e28 
…0.6.0` (NVIDIA-NeMo#5272)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(data): normalize GLM VLM assistant content (5216) into `r0.…
fdffbe0 
…6.0` (NVIDIA-NeMo#5273)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(data): pad Qwen VLM token type IDs (5215) into r0.6.0 (NV…
c5a6aae 
…IDIA-NeMo#5274)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: perf(recipe): tune Nemotron 3.5 Nano H100 benchmark (5266) into…
b757704 
… `r0.6.0` (NVIDIA-NeMo#5279)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: `fix(training): restore weak scaling for benchmark recipes (5222)…
df9cfb1 
…` into `r0.6.0` (NVIDIA-NeMo#5282)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@balasaajay
chore(deps): update transformer-engine dependency to a new commit hash
6770772 
Updated the transformer-engine dependency in pyproject.toml and uv.lock to use the latest commit (cfda3834) for improved stability and features.

Signed-off-by: Ajay Balasa <abalasa@nvidia.com>
@svcnvidia-nemo-ci @ghadiaravi13 @malay-nagda
cp: `feat (recipe): add Nemotron 3 Ultra MXFP8 GB200 pretraining perf…
1f2d8e6 
…-recipe (5287)` into `r0.6.0` (NVIDIA-NeMo#5309)

Signed-off-by: Ravi Ghadia <ghadiaravi13@users.noreply.github.com>
Signed-off-by: Malay Nagda <malayn@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Ravi Ghadia <40660742+ghadiaravi13@users.noreply.github.com>
Co-authored-by: Ravi Ghadia <ghadiaravi13@users.noreply.github.com>
Co-authored-by: Malay Nagda <malayn@nvidia.com>
@svcnvidia-nemo-ci @yaoyu-33
@cuichenx @meatybobby
cp: `fix(training): count expert parameter norms with TP and EP (4998…
9d4eef4 
…)` into `r0.6.0` (NVIDIA-NeMo#5177)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
Co-authored-by: meatybobby <meatybobby@gmail.com>
@svcnvidia-nemo-ci @yaoyu-33 @cuichenx
cp: fix(model): restore DeepSeek V3 sequence auxiliary loss (5165) …
9182eed 
…into `r0.6.0` (NVIDIA-NeMo#5172)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: Nemotron 3.5 Lightning checkpoint and naming updates (NVIDIA-NeMo…
e1f8367 
…#5293, NVIDIA-NeMo#5303) into r0.6.0 (NVIDIA-NeMo#5300)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@cuichenx @malay-nagda
fix(training): backport HybridEP uneven-dispatch compatibility to r0.…
3bc4e4d 
…6.0 (NVIDIA-NeMo#5288)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: Malay Nagda <malayn@nvidia.com>
Co-authored-by: Malay Nagda <malayn@nvidia.com>
svcnvidia-nemo-ci and others added 27 commits August 8, 2026 12:35
@svcnvidia-nemo-ci @yaoyu-33
cp: docs(recipe): retire stale Gemma 4 VL SFT launchers (5433) into…
3264eda 
… `r0.6.0` (NVIDIA-NeMo#5435)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(peft): synchronize shared expert LoRA across EP (5396) into…
75e6be5 
… `r0.6.0` (NVIDIA-NeMo#5428)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(model): preserve Step3.7 text config fields (5417) into `r0…
8fe66b5 
….6.0` (NVIDIA-NeMo#5429)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: [ckpt] fix: Deinterleave low-memory GLU checkpoints (5366) into…
07d182e 
… `r0.6.0` (NVIDIA-NeMo#5370)

Signed-off-by: Yu Yao <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@balasaajay
Update wandb dependency version to 0.27.2 (NVIDIA-NeMo#5461)
9905b29 
Signed-off-by: Ajay <abalasa@nvidia.com>
@cuichenx
fix(model): backport Gemma4 PLE sequence parallel sharding to r0.6.0 (N…
13ed820 
…VIDIA-NeMo#5450)

Signed-off-by: Chen Cui <chcui@nvidia.com>
@yaoyu-33
cp: fix(data): collate ragged packed sequence metadata (5400) into …
e78f969 
…`r0.6.0` (NVIDIA-NeMo#5426)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
@svcnemo-autobot
build: AUT-1430 upgrade GitPython to 3.1.58 (NVIDIA-NeMo#5480)
2bc595f 
Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: `fix(models): use HybridEP for MiniMax-M3 expert parallelism (544…
802b7f6 
…8)` into `r0.6.0` (NVIDIA-NeMo#5473)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @cuichenx
cp: `fix(diffusion): calculate WAN FLOPs from runtime geometry (5249)…
e6b2cbb 
…` into `r0.6.0` (NVIDIA-NeMo#5477)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @yaoyu-33
@cuichenx @claude @maanug-nv
cp: `feat(data): infer assistant loss mask for gpt-oss Harmony chat t…
0ed053d 
…emplate (5389)` into `r0.6.0` (NVIDIA-NeMo#5447)

Signed-off-by: Yu Yao <yaoyu.094@gmail.com>
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Maanu Grover <maanug@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: `fix(scripts): preserve FHT and fetch missing MCore commits (5478…
0494f09 
…)` into `r0.6.0` (NVIDIA-NeMo#5481)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@cuichenx @yaoyu-33
cp: fix(models): use HybridEP for Kimi-K2.5 inference (5421) into `…
dd61c0c 
…r0.6.0` (NVIDIA-NeMo#5485)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: Chen Cui <chcui@nvidia.com>
Co-authored-by: yaoyu-33 <yaoyu.094@gmail.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: fix(model): handle disabled GLM-5 MTP export (5445) into `r0.6.…
e790259 
…0` (NVIDIA-NeMo#5488)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: yaoyu-33 <yaoyu.094@gmail.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: docs(ckpt): document roundtrip scaling limits (5491) into `r0.6…
dd05e1a 
….0` (NVIDIA-NeMo#5492)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
@svcnvidia-nemo-ci @yaoyu-33
cp: docs(model-card): add Nemotron 3 Super verification card (5062)…
3ee5173 
… into `r0.6.0` (NVIDIA-NeMo#5494)

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
Signed-off-by: yaoyu-33 <yuya@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Yu Yao <54727607+yaoyu-33@users.noreply.github.com>
Co-authored-by: yaoyu-33 <yaoyu.094@gmail.com>
@svcnvidia-nemo-ci @malay-nagda
cp: docs: update performance summary for 26.08 (5496) into r0.6.0 (…
5796c06 
…NVIDIA-NeMo#5507)

Signed-off-by: Malay Nagda <malayn@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: malay-nagda <malayn@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: fix(deepseek-v4): use full-prefix inference (5509) into r0.6.0 (
c2e93d8 
NVIDIA-NeMo#5511)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: fix(recipe): prevent Qwen3.5-VL FSDP SFT OOM (5508) into `r0.6.…
b888318 
…0` (NVIDIA-NeMo#5515)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@yaoyu-33
fix(recipe): recompute Gemma 4 VL SFT layers (NVIDIA-NeMo#5506)
5b70a4c 
Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>
@svcnvidia-nemo-ci @cuichenx
cp: `docs(dsv4): preserve container packages during MCore sync (5514)…
7b5057e 
…` into `r0.6.0` (NVIDIA-NeMo#5516)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@balasaajay
Update nemo-evaluator path in fw_pyproject.toml (NVIDIA-NeMo#5571)
c932511 
Signed-off-by: Ajay <abalasa@nvidia.com>
@balasaajay
docs: Update 0.6.0 version entries for docs (NVIDIA-NeMo#5567)
8e43ce6 
Signed-off-by: Ajay <abalasa@nvidia.com>
@svcnvidia-nemo-ci @cuichenx
cp: fix(deepseek): default DeepSeek V4 to HybridEP (5543) into `r0.…
cea3a73 
…6.0` (NVIDIA-NeMo#5582)

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
Co-authored-by: Chen Cui <chcui@nvidia.com>
@ko3n1g
docs: fix preferred version and add 0.5.1 to the version picker (NVID…
0f113ab 
…IA-NeMo#5623)

Signed-off-by: oliver könig <okoenig@nvidia.com>
@github-actions
beep boop 🤖: Bumping Megatron-Bridge to v0.6.1 [skip ci]
51885cf 
Signed-off-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
@theely
models: add Apertus / Apertus 1.5 bridges
53bea08 
Upstream 0.6.x has no Apertus support. This adds:

- models/apertus/: ApertusBridge for ApertusForCausalLM, plus MCoreXIELU
  (a Megatron-side wrapper of transformers' XIELUActivation that requires the
  CUDA xielu kernel and TP-sums its alpha grads) and get_apertus_decoder_block_spec
  (RMSNorm q/k-layernorm + MCoreXIELU MLP activation).
- models/apertus1p5/: Apertus1p5Bridge for the multimodal
  Apertus1p5ForConditionalGeneration -- hyperparameters read from
  config.text_config, HF weight keys under model.language_model.*, pruned-LM-head
  handling (output_vocab_size < vocab_size), vision/audio tokenizer towers left
  unmapped (target is a text-only GPTModel).
- models/__init__.py: import both (the @register_bridge decorator only fires on
  import) and add to __all__.
- hf_pretrained/safe_config_loader.py: replace the filelock around
  AutoConfig.from_pretrained with contextlib.nullcontext -- fcntl.flock raises
  ENOLCK/ESTALE in-container on CSCS Lustre and the config is written once by
  rank 0 and read-only afterward.
@copy-pr-bot

copy-pr-bot Bot commented Sep 1, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@theely theely closed this Sep 1, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@malay-nagda malay-nagda Awaiting requested review from malay-nagda malay-nagda will be requested when the pull request is marked ready for review malay-nagda is a code owner
@dingqingy-nv dingqingy-nv Awaiting requested review from dingqingy-nv dingqingy-nv will be requested when the pull request is marked ready for review dingqingy-nv is a code owner

Assignees

No one assigned

Labels

community-request

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

9 participants

@theely @svcnvidia-nemo-ci @weijiac0619 @cuichenx @balasaajay @malay-nagda @svcnemo-autobot @yaoyu-33 @ko3n1g