Export DeepSeek V4 ratio-4 compressed sparse attention

[justinchuby]

Summary

• emit the frozen-v1 pkg.nxrt::CompressedSparseAttention path for DeepSeek-V4-Flash ratio-4 layers using live query-dependent indexer and compressor projections
• emit canonical planar block-FP8 BlockQuantizedMatMul and planar-FP4 sparse BlockQuantizedMoE nodes with strict native safetensors streaming
• thread explicit compressed/index graph state while preserving ratio-128 HCA and all non-native/non-CSA paths

Dependency and scope

• Depends on Add planar auxiliary-scale ABI for block-quantized ops justinchuby/onnx-genai#2321 for the planar auxiliary-scale ABI.
• Engine state-group/runtime session wiring remains separate work owned by Roy; this PR only defines and threads the producer graph I/O.
• No dense fallback or zero anchor is used to satisfy native CSA capability.

Validation

• 2135 passed, 1 skipped focused config/mapping/model/registry/GGUF/export suite
• Ruff format check and compatible project lint rules passed; targeted mypy passed for 8 producer modules
• real checkpoint header preflight: 46 shards, 69,187 tensors, 159,617,149,040 bytes
• real target graph: 41 CSA ops (21 ratio-4, 20 ratio-128), 666 block-quant matmuls, 43 block-quant MoEs; 1,822 streamed targets
• real MTP graph: 0 CSA ops, 17 block-quant matmuls, 1 block-quant MoE; 45 streamed targets

[github-actions]

Performance Comparison

Comparing dcf8bab → 0cd078d

Model	Metric	Baseline	Current	Delta
bert (feature-extraction)	model_size_bytes	359 KB	359 KB	+0.0%	⚪
bert (feature-extraction)	num_nodes	68	68	+0.0%	⚪
falcon	model_size_bytes	364 KB	364 KB	+0.0%	⚪
falcon	num_nodes	66	66	+0.0%	⚪
gemma2	model_size_bytes	428 KB	428 KB	+0.0%	⚪
gemma2	num_nodes	105	105	+0.0%	⚪
gpt2	model_size_bytes	324 KB	324 KB	+0.0%	⚪
gpt2	num_nodes	54	54	+0.0%	⚪
llama	model_size_bytes	425 KB	425 KB	+0.0%	⚪
llama	num_nodes	60	60	+0.0%	⚪
llama (static-cache)	model_size_bytes	425 KB	425 KB	+0.0%	⚪
llama (static-cache)	num_nodes	56	56	+0.0%	⚪
mamba (ssm-text-generation)	model_size_bytes	296 KB	296 KB	+0.0%	⚪
mamba (ssm-text-generation)	num_nodes	94	94	+0.0%	⚪
phi3	model_size_bytes	421 KB	421 KB	+0.0%	⚪
phi3	num_nodes	58	58	+0.0%	⚪
phi3 (static-cache)	model_size_bytes	421 KB	421 KB	+0.0%	⚪
phi3 (static-cache)	num_nodes	54	54	+0.0%	⚪
qwen2	model_size_bytes	425 KB	425 KB	+0.0%	⚪
qwen2	num_nodes	60	60	+0.0%	⚪
qwen2 (static-cache)	model_size_bytes	425 KB	425 KB	+0.0%	⚪
qwen2 (static-cache)	num_nodes	56	56	+0.0%	⚪
qwen3_5_moe (hybrid-text-generation)	model_size_bytes	506 KB	506 KB	+0.0%	⚪
qwen3_5_moe (hybrid-text-generation)	num_nodes	265	265	+0.0%	⚪
qwen3_5_text (hybrid-text-generation)	model_size_bytes	458 KB	458 KB	+0.0%	⚪
qwen3_5_text (hybrid-text-generation)	num_nodes	127	127	+0.0%	⚪
qwen3_5_vl (hybrid-qwen-vl)	model_size_bytes	977 KB	977 KB	+0.0%	⚪
qwen3_5_vl (hybrid-qwen-vl)	num_nodes	450	450	+0.0%	⚪
t5 (seq2seq)	model_size_bytes	836 KB	836 KB	+0.0%	⚪
t5 (seq2seq)	num_nodes	176	176	+0.0%	⚪
whisper (speech-to-text)	model_size_bytes	1008 KB	1008 KB	+0.0%	⚪
whisper (speech-to-text)	num_nodes	128	128	+0.0%	⚪

No performance regressions.

[github-actions]

🏗️ Architecture Diff

Comparing dcf8bab → 0cd078d

Model	Sub-model	Changes	Status
bert (feature-extraction)	model	0	⚪
falcon	model	0	⚪
gemma2	model	0	⚪
gemma4 (gemma4)	decoder	0	⚪
gemma4 (gemma4)	embedding	0	⚪
gemma4 (gemma4)	vision_encoder	0	⚪
gemma4_text	model	0	⚪
gpt2	model	0	⚪
llama	model	0	⚪
llama (static-cache)	model	0	⚪
mamba (ssm-text-generation)	model	0	⚪
phi3	model	0	⚪
phi3 (static-cache)	model	0	⚪
qwen	model	0	⚪
qwen (static-cache)	model	0	⚪
qwen2	model	0	⚪
qwen2 (static-cache)	model	0	⚪
qwen2_moe	model	0	⚪
qwen2_moe (static-cache)	model	0	⚪
qwen3	model	0	⚪
qwen3 (static-cache)	model	0	⚪
qwen3_5_moe (hybrid-text-generation)	model	0	⚪
qwen3_5_text (hybrid-text-generation)	model	0	⚪
qwen3_5_vl (hybrid-qwen-vl)	decoder	0	⚪
qwen3_5_vl (hybrid-qwen-vl)	embedding	0	⚪
qwen3_5_vl (hybrid-qwen-vl)	vision_encoder	0	⚪
qwen3_moe	model	0	⚪
qwen3_moe (static-cache)	model	0	⚪
qwen3_next (hybrid-text-generation)	model	0	⚪
t5 (seq2seq)	decoder	0	⚪
t5 (seq2seq)	encoder	0	⚪
whisper (speech-to-text)	decoder	0	⚪
whisper (speech-to-text)	encoder	0	⚪

No architecture changes detected. ✅

---

Legend: ⚪ No change · 🔵 Minor (attrs/inits) · 🟡 Moderate (nodes added/removed) · 🔴 Major (interface changed)