auto-round
Stream AutoScheme's per-layer sensitivity scoring block-by-block
#2063

Merged

Stream AutoScheme's per-layer sensitivity scoring block-by-block #2063

xin3he merged 28 commits into intel:main from aquilarubra:pr/autoscheme-streaming

aquilarubra force pushed from df4f536c to 2b84c1a4 51 days ago

aquilarubra force pushed from 445f953f to 5721fe1a 51 days ago

aquilarubra force pushed from bb4d4b2e to 499daf1f 51 days ago

chensuyue requested a review from

wenhuach21 50 days ago

wenhuach21 requested a review from

lvliang-intel 49 days ago

lvliang-intel commented on 2026-07-22

aquilarubra force pushed from 58e3ccf9 to 21a28b4f 48 days ago

hshen14 requested a review from

xin3he 47 days ago

wenhuach21 approved these changes on 2026-07-27

aquilarubra force pushed from 21a28b4f to 2c6a023b 43 days ago

aquilarubra force pushed from cb538119 to b71b8b95 43 days ago

xin3he requested a review from

copilot-pull-request-reviewer 42 days ago

copilot-pull-request-reviewer commented on 2026-07-28

xin3he approved these changes on 2026-07-29

Add AR_DISK_STREAM_MODEL and AR_RESUME_DIR env vars

dac780c3

Add disk-streaming primitive for per-block checkpoint materialization

083bb440

context/model: build meta-device skeleton under AR_DISK_STREAM_MODEL

eaaf80c3

offload: materialize/reload blocks from meta correctly

c90acd4d

offload: use a deterministic offload dir when AR_RESUME_DIR is set

a72847be

compressors/base: propagate model_dir to offloader; defer resume clear

011874a1

compressors/data_driven: honor disk streaming outside resumability

97696fe7

[pre-commit.ci] auto fixes from pre-commit.com hooks

a863e847

Strip leaked internal-repo comments; fix meta-materialize dtype bug

933cb6b3

Add unit tests for disk_stream_util.py primitives

09450c8e

Re-trigger CI (unrelated HF Hub 503 flake on test_gptoss)

de389450

Re-trigger CI (retest after flaky VRAM/RAM perf checks and llmc_forma…

b60e8646

Re-trigger CI (unrelated HF Hub 503 flake on test_gptoss)

22e0c2dc

Re-trigger CI (retest after flaky VRAM/RAM perf checks and llmc_forma…

06f1a0b9

Add AR_DISK_STREAM_MODEL and AR_RESUME_DIR env vars

949ff2ca

Re-trigger CI (unrelated HF Hub 503 flake on test_gptoss)

0050b4dd

compressors/data_driven: resumable standard tuning loop (AR_RESUME_DIR)

11d1d45d

Add ResumeState: crash/resume checkpointing for the tuning loop

d9cc0fce

shard_writer: discover pre-existing shards to resume across processes

f595ce1a

export: skip pack_layer() for a resumed, still-meta layer

3140a5f4

[pre-commit.ci] auto fixes from pre-commit.com hooks

e0ca142a

Strip leaked internal-repo comment markers

d7f0e36c

Add unit and integration tests for AR_RESUME_DIR

995afeaf

Re-trigger CI (retest after flaky VRAM/RAM perf checks and llmc_forma…

ead1edac

Fix envs.py syntax corrupted during rebase conflict resolution

2092f07c

auto_scheme: disk-streaming support prep (gen_auto_scheme low_cpu_mem…

7434f229

auto_scheme: stream per-layer sensitivity scoring block-by-block

b9eb98ff

aquilarubra force pushed from b71b8b95 to b9eb98ff 41 days ago

[pre-commit.ci] auto fixes from pre-commit.com hooks

8f141239

xin3he merged 73661edb into main 40 days ago

Reviewers

xin3he

wenhuach21

lvliang-intel

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

auto-round Stream AutoScheme's per-layer sensitivity scoring block-by-block #2063 Merged

Stream AutoScheme's per-layer sensitivity scoring block-by-block #2063

auto-round
Stream AutoScheme's per-layer sensitivity scoring block-by-block
#2063

Merged