Reinforcement Learning
Transformers
English
post-training
distillation
agentic-coding
composer-2.5
cursor
kimi-k2
grpo
dapo
diloco
openenv
trl
verl
research
methodology
Instructions to use Codeseys/composer-replication-framework with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Codeseys/composer-replication-framework with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Codeseys/composer-replication-framework", dtype="auto") - Notebooks
- Google Colab
- Kaggle
Baladithya Balamurugan Claude Opus 4.8 (1M context) commited on
Commit ·
ace4fac
1
Parent(s): 7d9dbbc
docs: file R13 (pre-existing serverless ruff B904 debt) as tracked backlog item
Browse files
docs/BACKLOG_RESOLUTION_2026-06-09.md
CHANGED
|
@@ -77,6 +77,8 @@ Goal-driven systematic resolution of every pending item. This doc is the live au
|
|
| 77 |
|
| 78 |
**Wave 3 — DONE (Phase-7 reconciliation):** R1 ✅ (HeldOutGuard wired into ComposerReplicationTrainer — optional, OFF by default, soft/hard stop; + integration test), R2 ✅ (HeldoutSplit disjointness enforcer `safety/holdout.py` + 10 tests), R3 ✅ (EKS entrypoint contract bug fixed — `replica_entrypoint.__main__` now resolves from env OR argv; proven end-to-end with a pure-env invocation), R4 ✅ (calibrate_kl_threshold rejects factor<=0/negative-baseline + positive floor), R7 ✅ (API_REFERENCE §15-17: EKS/SageMaker/DockerSandbox/safety), R8 ✅ (ADR-015 authored + indexed), R10 ✅ (path-(c) divergence-rate test). R12 ✅ (B4/B7 complete). R5 ✅ (EKS+SageMaker cancel now re-raise unexpected errors, swallow only 404/409/already-terminal + propagation test), R6 ✅ (EKS collect() result dicts include `result`=rendezvous URI), R11 ✅ (spike-006 test seeded torch.manual_seed(0) → no longer contention-flaky). ALL Wave-3 items (R1-R11) CLOSED.
|
| 79 |
|
|
|
|
|
|
|
| 80 |
Sandbox refactor verdict: **clean** (no regression to LocalSubprocessSandbox/FeatureDeletionEnv).
|
| 81 |
|
| 82 |
## Wave plan
|
|
|
|
| 77 |
|
| 78 |
**Wave 3 — DONE (Phase-7 reconciliation):** R1 ✅ (HeldOutGuard wired into ComposerReplicationTrainer — optional, OFF by default, soft/hard stop; + integration test), R2 ✅ (HeldoutSplit disjointness enforcer `safety/holdout.py` + 10 tests), R3 ✅ (EKS entrypoint contract bug fixed — `replica_entrypoint.__main__` now resolves from env OR argv; proven end-to-end with a pure-env invocation), R4 ✅ (calibrate_kl_threshold rejects factor<=0/negative-baseline + positive floor), R7 ✅ (API_REFERENCE §15-17: EKS/SageMaker/DockerSandbox/safety), R8 ✅ (ADR-015 authored + indexed), R10 ✅ (path-(c) divergence-rate test). R12 ✅ (B4/B7 complete). R5 ✅ (EKS+SageMaker cancel now re-raise unexpected errors, swallow only 404/409/already-terminal + propagation test), R6 ✅ (EKS collect() result dicts include `result`=rendezvous URI), R11 ✅ (spike-006 test seeded torch.manual_seed(0) → no longer contention-flaky). ALL Wave-3 items (R1-R11) CLOSED.
|
| 79 |
|
| 80 |
+
**Pre-existing tech-debt (discovered, tracked, OUT of this effort's scope — do not silently reformat the existing codebase):** R13 — ~14 `ruff B904` (raise-without-from) + import-order nits in PRE-EXISTING serverless files (executor.py, hf_jobs.py, modal.py, modal_spawn.py, + 4 pre-existing tests) and spike-006 smoke files. These predate Wave 1-3; a `ruff --fix` would touch unauthored code. Filed for a dedicated lint-debt pass. My Wave 1-3 files are all ruff-clean.
|
| 81 |
+
|
| 82 |
Sandbox refactor verdict: **clean** (no regression to LocalSubprocessSandbox/FeatureDeletionEnv).
|
| 83 |
|
| 84 |
## Wave plan
|