Codeseys/composer-replication-framework

1# HF Repo Layout — composer-replication-framework2 3Per the [HF multi-artifact research project pattern](https://huggingface.co/docs/hub/repositories), this project will eventually span multiple HF repos. This document records the layout.4 5## Current state (2026-05-25)6 7Only the **methodology repo** exists. No trained variants, no datasets yet.8 9| Repo | Type | Status | Purpose |10|---|---|---|---|11| `Codeseys/composer-replication-framework` | model | ✅ exists (this repo) | Methodology, ADRs, framework spec, research deep-dives |12 13## Planned splits (post-spike)14 15When the v0.0 spike produces a result, the following repos will be created:16 17| Repo | Type | Created when | Contents |18|---|---|---|---|19| `Codeseys/composer-replication-traces-v0` | dataset | v0.0 spike data is collected | 100 frozen agentic-coding traces (JSON), used for trace-replay-distillation experiments |20| `Codeseys/composer-replication-qwen3-7b-v0` | model | v0.0 spike produces a checkpoint | LoRA adapter or full fine-tune of Qwen3-7B trained with Dr.GRPO + the framework's own additive trace-replay-DPO research channel (channel 3 is our addition, **not** part of Cursor's Composer recipe — see [ADR-014](adrs/ADR-014-policy-optimization-objective-menu.md)) |21| `Codeseys/composer-replication-qwen3-7b-v0-baseline` | model | v0.0 spike produces a baseline checkpoint | Same training, plain GRPO only (A/B comparison) |22 23After v0.1:24 25| Repo | Type | Contents |26|---|---|---|27| `Codeseys/composer-replication-traces-v1` | dataset | Larger trace corpus + Feature-Deletion environment seed repos |28| `Codeseys/composer-replication-feature-deletion-env-v1` | dataset | Repos with passing tests, with deletion masks for the env to apply |29| `Codeseys/composer-replication-qwen3-32b-v1` | model | Full Composer-channel (Dr.GRPO + SDPO) v1 trained variant, combined with the framework's additive trace-replay-DPO research channel (trace-replay is our addition, not Composer's recipe) |30 31All trained-variant repos will:32- Link back to **this repo** (`Codeseys/composer-replication-framework`) in their `README.md` as the methodology source.33- Live in an **HF Collection** (`composer-replication-*`) created when the second member repo is added.34 35## Why this split36 37Per the `huggingface-hub` skill's `references/multi-artifact-research-layout.md`:38 391. **Type semantics matter** — HF dataset repos have native handling for jsonl/parquet (streaming load, dataset viewer). The model repo type used for *this* repo treats markdown research as first-class.402. **Cite-ability** — each trained variant gets its own DOI / citation.413. **Variant training is unbounded** — we don't know how many variants will ship; per-variant repos keep eval results, model cards, and weights cleanly separated.424. **Discoverability via Collection** — single URL surfaces the whole study.43 44## Conventions45 46- **Repo prefix**: `composer-replication-` for every repo in this study.47- **Variant suffix**: `<base-model>-<size>-<scale-tag>` (e.g. `qwen3-7b-v0`, `qwen3-32b-v1`).48- **Dataset suffix**: `-traces-v<N>`, `-feature-deletion-env-v<N>`, `-bench-v<N>`.49- **Branch**: `master` locally → push to HF as `main` (refspec `master:main`).50- **License**: MIT for methodology and code; per-trained-variant license depends on base model's license.51 52## Sync pattern53 54When adding a new variant repo, use the `huggingface-hub` skill's `references/sync-to-hf-template.py` shape — `create_repo` + `upload_folder` + `add_collection_item(exists_ok=True)` in a single script, so shipping a new variant is one command.55