Deprecate unused DeepSpeed features

Summary

Tracking issue for retiring unused / unmaintained DeepSpeed features so the runtime, docs, tests, and packaging stay consistent.

Goal: remove code paths that are no longer supported, then sweep the leftover docs, extras, CODEOWNERS, and tests so they cannot be re-enabled by accident.

PRs

PR	State	Scope
#8490	Merged 2026-09-12	First sweep: compression library, 1-bit Adam/Lamb, 0-1 Adam, MiCS, Nebula (71d316d60, +169 / −8,298)
#8099	Merged 2026-07-01	ZeRO-3 elastic_checkpoint (use Universal Checkpointing). Stage 1/2 unchanged
#8493	Merged 2026-09-14	Sparse Attention
#8504	Merged 2026-09-16	sparse_gradients
#8534	Open	LoCo-Zero++ (zeropp_loco_param)
#8535	Open	Compression leftovers: 1-bit comm backends, MoQ/eigenvalue, fake-quant kernels, docs/CODEOWNERS/issue template

There is no PR for the full follow-up sweep described below (MoQ + 1-bit comm + PLD + Apex AMP as one bundle). #8535 covers the 1-bit comm and MoQ/eigenvalue leftovers. Progressive Layer Dropping and Apex AMP still have no PR.

Coverage statistics

Line counts are from git diff --numstat. Code = runtime/library sources; tests = tests/; docs = docs/ plus CODEOWNERS / issue templates / CI where noted.

Removed in #8490

Feature	Code −	Tests −	Docs −	Net
Compression library	2,430	301	455	−3,166
1-bit / 0-1 Adam optimizers	1,132	1,255	745	−3,131
MiCS ZeRO-3	650	225	34	−909
Nebula checkpoint engine	229	0	0	−229
Runtime wiring (config.py / engine.py / launchers / CODEOWNERS)	264	0	3	−208
Shared docs (config-json, training pages, nav)	0	0	513	−506
Shared tests (config dict, AutoEP MiCS gate, ZeRO)	0	62	0	−19
Helper relocation (recursive_getattr → module_utils)	0	0	0	+39
PR total				+169 / −8,298

Follow-up sweep (no dedicated PR)

Local working-tree estimate vs master when this was written: +104 / −2,814. Combined vs pre-#8490: +253 / −11,092.

Feature	Code −	Tests −	Docs −	Status
1-bit comm backends	717	676	0	Open in #8535
MoQ / eigenvalue	333	18	212	Open in #8535
Progressive Layer Dropping	40	116	155	No PR
Apex AMP / runtime wiring	383	0	0	No PR (engine amp path)
Docs / CI / packaging	24	0	113	Partially in #8535
ZeRO-3 elastic checkpoint	16	—	config-json	Done in #8099
Rejection tests	0	+52	0	Split across follow-up PRs

Leftover configs should fail at parse time (DeepSpeedConfigError or ZeRO-3 ValidationError) instead of silently training without the feature.

Features to deprecate

Training / checkpointing

• Nebula checkpoint engine (deepspeed/nebula, NebulaCheckpointEngine) — #8490
• MiCS ZeRO-3 sharding (MiCS_Init, mics_shard_size, mics_hierarchical_params_gather) — #8490
• ZeRO-3 elastic checkpoint (zero_optimization.elastic_checkpoint with stage 3; use Universal Checkpointing). ZeRO-1/2 elastic checkpoints remain. — #8099

Optimizers

• 1-bit Adam (OneBitAdam) — #8490
• 1-bit Lamb (OneBitLamb) — #8490
• 0/1 Adam (ZeroOneAdam) — #8490

Compression / quantization

• Model compression library (deepspeed.compression, compression scheduler, related tests) — #8490
• Mixture-of-Quantization (MoQ) / quantize_training / eigenvalue scheduling — #8535

Mixed precision

• Apex AMP (amp config). Use fp16, bf16, or torch_autocast. No PR.

Training schedule

• Progressive Layer Dropping (progressive_layer_drop). No PR.

1-bit communication leftovers (optimizers are gone; backends remain)

• Compressed allreduce backends (deepspeed/runtime/comm/{nccl,mpi,compressed}.py, deepspeed/runtime/compression/cupy.py) — #8535
• tests/onebit/ and remaining unit tests for those backends — #8535
• setup.py extras 1bit / 1bit_mpi and requirements/requirements-1bit-mpi.txt — #8535

Docs / repo hygiene

• Docs nav Model Compression (removed in #8490)
• Docs nav Mixture-of-Quantization and Progressive Layer Dropping — MoQ nav in #8535; PLD still open
• .github/ISSUE_TEMPLATE/compression_bug_report.md — #8535
• CODEOWNERS /deepspeed/runtime/fp16/onebit/ (removed in #8490)
• CODEOWNERS /deepspeed/runtime/compression/ — #8535
• Config / tutorial leftovers for MoQ, Apex AMP, PLD (bulk Nebula/one-bit optimizer/MiCS docs already removed in #8490)

GitHub usage consensus (2026-09-14)

Three independent classifications (Opus, Codex, Grok) over the same GitHub pull agreed on the next deprecation set below.

Method

• Issues/PRs: unauthenticated /search/issues on 2026-09-14. Primary column is -org:deepspeedai. Relative score vs "zero_optimization" -org:deepspeedai = 1,362.
• Code: authenticated /search/code on 2026-09-14 via gh. Relative score vs "zero_optimization" -org:deepspeedai = 49,280. Code search over-counts vendored DeepSpeed copies (e.g. DeepSpeed-0.9.5/ inside other repos) and generic identifiers. Treat as ordinal rank, not user counts.
• Not consensus / do not add from this scan: inference v1/v2, Windows, ZeRO offload, pipeline, DeepCompile, Hybrid Engine, autotuning, DataStates, ZenFlow.

Consensus candidates

Feature	Issue query	Issues −org	Rel. issues	Issues in DS repo	Code query	Code −org	Rel. code	Status
Sparse Attention	"sparse_attention"	358 (158 w/ "mode")	0.26 (0.12)	86	"sparse_attention"	27,264	0.55	Done #8493
sparse_gradients	"sparse_gradients"	20	0.015	8	"sparse_gradients"	4,616	0.094	Open #8504
Elastic training	"max_train_batch_size" "micro_batch_sizes"	0	0	64 (word “elasticity”)	same pair	1,900	0.039	No PR
graph_harvesting	"graph_harvesting"	83	0.061	10	"graph_harvesting"	1,052	0.021	No PR (0 tests; config-dump noise)
Legacy curriculum_learning	"curriculum_learning"	185	0.14	27	"curriculum_learning"	15,136	0.31	No PR (homonym; superseded by Data Efficiency)
LoCo-Zero++	"zeropp_loco_param"	37	0.027	3	"zeropp_loco_param"	152	0.003	Open #8534

2-of-3 only (not unanimous; do not treat as consensus)

Data Efficiency / Random-LTD, dump_state (debug flag; "dump_state" code search is 164k unrelated hits), disable_allgather, FusedLion.

Other unused / already-deprecated candidates

• Sparse Attention (deepspeed/ops/sparse_attention, extras_require['sparse_attn']) — #8493
• sparse_gradients (config-json already calls this essentially deprecated) — #8504
• Elastic training (elasticity / max_train_batch_size + micro_batch_sizes)
• graph_harvesting
• Legacy curriculum learning (superseded by Data Efficiency)
• LoCo-Zero++ (zero_optimization.zeropp_loco_param) — #8534

Notes

• Checked items landed in #8490, #8099, or #8493.
• The original follow-up sweep is not a single open PR. #8535 takes 1-bit comm + MoQ leftovers; PLD and Apex AMP still need a PR.
• MiCS removal also dropped the AutoEP ZeRO-3 “no MiCS” compatibility gate and its unit test.
• 1-bit optimizers are gone in #8490; 1-bit communication backends are in #8535 (coalesced_collectives.py stays for ZeRO-3).
• deepspeed/compression/helper.py remains as a FutureWarning shim for DeepSpeed-Chat after #8490.
• Apex AMP removal is still local / un-PRed; PyTorch torch_autocast is unchanged. Local import apex remains only to detect apex.optimizers.FusedAdam for FP16 optimizer wrapping.
• Stats methodology: #8490 from git diff --numstat 71d316d60^1 71d316d60; follow-up from git diff --numstat origin/master on the working tree at the time; combined vs b726f4edb.

Start by separating the unchecked features from work already covered by PRs #8490, #8099, #8493, #8534, and #8535. Read the listed runtime paths, configuration entries, tests, setup.py extras, docs, and CODEOWNERS files, beginning with the specific feature you select. Done means the unsupported configuration fails at parse time and its runtime code, tests, packaging, and documentation no longer advertise it.