[OMNIML-2850] [3/n] Adds sparse attention calibration

[kaix-nv]

What does this PR do?

Type of change: ?
new feature

Overview: ?

• This PR adds the sparse attention calibration algorithm.
• This PR adds sparse attention integration to the llm_eval evaluation examples including lm_eval and MMLU.

Usage

import modelopt.torch.sparsity.attention_sparsity as mtsa

# Apply sparse attention with calibration
model = mtsa.sparsify(model, config=SKIP_SOFTMAX_CALIB)

# Print summary - now shows actual thresholds
mtsa.print_sparse_attention_summary(model)
# Output:
# Method: flash_skip_softmax, Threshold: Dynamic (λ=437.395926)

# Or llm_eval integration
# HuggingFace sparse attention example
python examples/llm_sparsity/attention_sparsity/hf_sa.py \
    --pyt_ckpt_path Qwen/Qwen3-4B \
    --sparse_attn skip_softmax_calib \
    --verify_output

# LM Eval with sparse attention
python examples/llm_eval/lm_eval_hf.py \
    --model hf \
    --model_args pretrained=Qwen/Qwen3-4B \
    --tasks hellaswag \
    --sparse_cfg SKIP_SOFTMAX_CALIB

# MMLU with sparse attention (NEW)
python examples/llm_eval/mmlu.py \
    --model_name causal \
    --model_path Qwen/Qwen3-4B \
    --sparse_cfg SKIP_SOFTMAX_CALIB

Testing

Before your PR is "Ready for review"

• Make sure you read and follow Contributor guidelines and your commits are signed.
• Is this change backward compatible?: Yes/No
• Did you write any new necessary tests?: Yes/No
• Did you add or update any necessary documentation?: Yes/No
• Did you update Changelog?: Yes/No

Additional Information

[codecov]

Codecov Report

❌ Patch coverage is 81.33484% with 165 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.69%. Comparing base (8188a01) to head (c9d7008).

Files with missing lines	Patch %	Lines
...rsity/attention_sparsity/calibration/calibrator.py	32.45%	77 Missing ⚠️
...arsity/attention_sparsity/calibration/calibrate.py	51.72%	28 Missing ⚠️
...ch/sparsity/attention_sparsity/sparse_attention.py	75.36%	17 Missing ⚠️
...pt/torch/sparsity/attention_sparsity/conversion.py	91.85%	11 Missing ⚠️
...y/attention_sparsity/methods/flash_skip_softmax.py	91.00%	9 Missing ⚠️
...orch/sparsity/attention_sparsity/model_sparsify.py	76.00%	6 Missing ⚠️
...sparsity/attention_sparsity/calibration/dataset.py	97.43%	5 Missing ⚠️
...ch/sparsity/attention_sparsity/methods/registry.py	80.76%	5 Missing ⚠️
...delopt/torch/sparsity/attention_sparsity/config.py	95.40%	4 Missing ⚠️
modelopt/torch/sparsity/attention_sparsity/mode.py	90.32%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #538      +/-   ##
==========================================
+ Coverage   74.37%   74.69%   +0.32%     
==========================================
  Files         182      196      +14     
  Lines       18219    19103     +884     
==========================================
+ Hits        13550    14269     +719     
- Misses       4669     4834     +165     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.