Skip to content

[NVFp4][MoE Calibration] Fix MoE Calibration Context #1864

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
dsikka merged 3 commits into from
Sep 25, 2025
Merged

[NVFp4][MoE Calibration] Fix MoE Calibration Context #1864

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cd05b9f
only use 20 samples for nvfp4, fix boolean
dsikka Sep 24, 2025
b6d0448
Merge branch 'main' into fix_qwen_calibration
dsikka Sep 24, 2025
daf7823
Merge branch 'main' into fix_qwen_calibration
dsikka Sep 25, 2025
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
3 changes: 1 addition & 2 deletions examples/quantization_w4a4_fp4/qwen_30b_a3b.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@
DATASET_SPLIT = "train_sft"

# Select number of samples
NUM_CALIBRATION_SAMPLES = 200
NUM_CALIBRATION_SAMPLES = 20
MAX_SEQUENCE_LENGTH = 2048

# Load dataset and preprocess.
Expand Down Expand Up @@ -84,7 +84,6 @@ def tokenize(sample):
print(tokenizer.decode(output[0]))
print("==========================================\n\n")


# Save to disk in compressed-tensors format.
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-NVFP4"
model.save_pretrained(SAVE_DIR, save_compressed=True)
Expand Down
2 changes: 1 addition & 1 deletion src/llmcompressor/modeling/prepare.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,7 @@ def update_qwen3_moe(model, stack, calibrate_all_experts):
def moe_calibration_context(
model: PreTrainedModel,
stack,
calibrate_all_experts: bool = False,
calibrate_all_experts: bool = True,
):
# Temporarily updates the MoE modules within the context
# Once the context exists, parameter updates persist
Expand Down