Allow independent QKV precision in PyTorch quantization passes

[Ti-Tai Wang (titaiwangms)]

Summary

Add an opt-in independent_qkv: true setting to the native PyTorch RTN, KQuant,
and GPTQ passes. It preserves separate quantization settings for split attention
Q/K/V projections (for example, INT4 Q/K and INT8 V). The default stays false,
so existing packed-QKV workflows retain their current normalization behavior.

The option guards both QKV-normalization points: fresh quantization and
quantization of a partially quantized checkpoint. Existing quantized weights
remain locked, and the option is invocation-local; it is not persisted in the
checkpoint schema. SelectiveMixedPrecision, AutoClip, ONNX exporters, and
quantization kernels are unchanged.

{
  "type": "Rtn",
  "bits": 4,
  "group_size": 128,
  "independent_qkv": true,
  "overrides": {
    "re:.*\\.self_attn\\.v_proj": {"bits": 8}
  }
}

Only use this option with downstream exporters that retain separate Q/K/V
projections; it does not enable mixed-width packed-QKV MatMul.

Validation

• test_independent_qkv.py exercises real RTN, KQuant, and GPTQ passes, the
  original default, explicit/regex overrides, disk save/reload and packed
  buffers, locked checkpoint merges, and exclusions.
• A tiny offline Qwen3-MoE RTN checkpoint retains Q/K/V at 4/4/8 after reload
  while fused experts remain quantized.
• Full test_quant_utils.py plus the new suite: 94 passed.
• Existing targeted RTN/KQuant/GPTQ checkpoint and MoE tests: 7 passed.
• Ruff format and targeted Ruff check passed excluding CPY001, which also
  reports on unchanged files in the base checkout; git diff --check passed.

Full 30B ONNX export, ORT CUDA execution, and the separate mixed-width QMoE
full-logit parity gate were not part of this change and are not claimed to pass.

This upstream-head PR supersedes #2697. The follow-up commit removes Pylint W0212 protected-member access in the new tests.

[Copilot]

Copilot review overview

🟢 Approval recommended

The implementation is backward-compatible, scoped, documented, and thoroughly tested.

Review effort: Balanced
Findings: None

What changed in this PR

Adds opt-in independent Q/K/V precision to native PyTorch quantization while preserving existing behavior by default.

Changes:

• Exposes independent_qkv for RTN, KQuant, and GPTQ.
• Skips QKV normalization when enabled, including checkpoint merges.
• Adds documentation and comprehensive offline tests.

File	Description
olive/​passes/​pytorch/​quant_utils.py	Defines and applies the new option.
olive/​passes/​pytorch/​rtn.py	Enables the option for RTN.
olive/​passes/​pytorch/​kquant.py	Enables the option for KQuant.
olive/​passes/​pytorch/​gptq.py	Enables the option for GPTQ.
test/​passes/​pytorch/​test_independent_qkv.py	Tests defaults, overrides, reloads, exclusions, and MoE.
docs/​source/​features/​quantization.md	Documents usage and exporter limitations.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.