[Jiajia Qin (qjia7)]

Activating a LoRA adapter after a generator has captured decode currently reuses the base path's gpu_graph_id. Graph replay bypasses input processing and kernel selection, so decode can silently ignore the newly activated adapter.

This change includes the active-adapter revision in capture identity, alongside input length and execution variant. It reuses the process-wide GraphAnnotationIds allocator to prevent collisions between generators sharing a session, retains older captures for cleanup, and uses the actual capture ID for recurrent-state capture fixup. Adapter activation remains additive; it does not recompute earlier KV-cache entries.

The new Python regression is included in the regular test runner. It captures the base path, activates A and then A+B on the same generator after rewind, compares full logits with fresh active generators, and verifies a fresh base generator sharing the session. It also checks distinct capture IDs and adapter unloading.

Validation on Windows / NVIDIA RTX 5080:

• Release native library and Python wheel built successfully. Formatting and whitespace checks pass.
• WebGPU regression: 2 passed (capture off/on); the two CUDA parameterizations were not run locally.
• Full-model reproduction before the fix: capture-off comparisons were exact, but captured decode differed from a fresh active generator by up to 2.82764 in logits. After the fix, all six forced-prefix logit rows match exactly with built-in WebGPU, and with the plugin under capture.
• Isolated native performance checks cover FP16 and INT8 adapters, on/off, at 128 and 1024 prompt tokens. Two initially slower cases were checked in a three-configuration comparison separating the ORT update from GenAI. Final 1024-token confirmation changes versus the previous build were +0.14% decode / -0.41% TTFT for FP16 adapter-off and -0.25% decode / +0.15% TTFT for INT8 adapter-on. No repeatable regression was observed on this GPU.

Related ORT operator and capture coverage: microsoft/onnxruntime#33370. This GenAI fix addresses capture identity independently of that operator implementation.

[Copilot]

🔵 Needs a closer look

CUDA parameterizations remain unvalidated, and adapter recapture with recurrent state needs hardware-backed confirmation.

0 open findings

What changed in this PR

Fixes graph-capture identity so activating LoRA adapters does not replay a previously captured base-model path.

Changes:

• Includes adapter revision in capture identity using process-wide IDs, retaining older captures for cleanup.
• Uses actual capture IDs for recurrent-state fixup.
• Adds regression coverage and documents additive activation and KV-cache behavior.

File	Description
test/​python/​test_onnxruntime_genai.py	Includes the new regression in the regular runner.
test/​python/​test_adapter_graph_capture.py	Tests adapter activation, capture isolation, logits, and unloading.
src/​ort_genai_c.h	Documents activation and KV-cache semantics.
src/​models/​multi_modal.cpp	Uses actual capture IDs for recurrent-state fixup.
src/​models/​model.h	Replaces local capture-ID storage with the shared allocator.
src/​models/​model.cpp	Allocates adapter-aware capture IDs and releases retained captures.
src/​models/​decoder_only.cpp	Uses actual capture IDs for recurrent-state fixup.

🧠 Review effort: Balanced

---

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

[Baiju Meswani (baijumeswani)]

The main fix looks sound. I did not find a confirmed correctness bug. These four comments are non-blocking suggestions, mainly about test coverage and making fallback behavior easier to understand.