Skip to content

fix(ci): match Nemotron remote reloads with legacy PEFT exports #3952

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
yuhezhang-ai merged 6 commits into from
Sep 22, 2026
Merged

fix(ci): match Nemotron remote reloads with legacy PEFT exports #3952

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5ad3e74
fix(ci): bound GPT-OSS cross-framework parity to 512 tokens
yuhezhang-ai Sep 20, 2026
8b0bd42
fix(ci): pin Mamba cumsum tile during checkpoint parity
yuhezhang-ai Sep 21, 2026
24416ff
fix(ci): reload Nemotron PEFT adapters with built-in HF models
yuhezhang-ai Sep 20, 2026
7156f57
fix(ci): keep Nano HF reference on one CUDA device
yuhezhang-ai Sep 21, 2026
446557d
fix(ci): match Nemotron remote reloads with legacy PEFT exports
yuhezhang-ai Sep 21, 2026
dd2b963
Merge branch 'main' into yuhez/fix/amint-356-hf-peft-reload
yuhezhang-ai Sep 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions examples/llm_finetune/nemotron/customizer_nemotron_nano_peft.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,4 +94,6 @@ ci:
checkpoint_robustness:
tokenizer_name: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
experts_implementation: grouped_mm
# Pair the remote-code HF receiver with the legacy adapter export.
trust_remote_code: true
checkpoint.v4_compatible: true
2 changes: 2 additions & 0 deletions examples/llm_finetune/nemotron/customizer_nemotron_nano_peft_packing.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,4 +97,6 @@ ci:
checkpoint_robustness:
tokenizer_name: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
experts_implementation: grouped_mm
# Pair the remote-code HF receiver with the legacy adapter export.
trust_remote_code: true
checkpoint.v4_compatible: true
2 changes: 2 additions & 0 deletions examples/llm_finetune/nemotron/nemotron_nano_4b_squad_peft.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,6 +125,8 @@ ci:
# tp_size stays at the base value (1): the dense Nemotron-H custom path is FSDP2-only.
# Use the same cached family tokenizer for the long parity document.
tokenizer_name: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
# Pair the remote-code HF receiver with the legacy adapter export.
trust_remote_code: true
checkpoint.v4_compatible: true
dataset.limit_dataset_samples: 500
validation_dataset.limit_dataset_samples: 500
2 changes: 2 additions & 0 deletions examples/llm_finetune/nemotron/nemotron_nano_v3_hellaswag_peft.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,7 +119,9 @@ ci:
checkpoint_robustness:
hf_device_map_auto: true
experts_implementation: grouped_mm
# Pair the remote-code HF receiver with the legacy adapter export.
trust_remote_code: true
checkpoint.v4_compatible: true
tokenizer_name: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
distributed.ep_size: 8
dataset.num_samples_limit: 500
Expand Down
Loading