[RWKV] Align module names with Transformers conventions

[Cai-z-us]

What does this PR do?

Follows up on #10 by making the RWKV-7 model layout native to the current Transformers and FLA ecosystem. Canonical BlinkDL names now exist only at the .pth conversion boundary; the Transformers model and serialized state dict use standard decoder/module names without duplicate aliases.

• rename the model layout to embed_tokens, embedding_norm, layers, norm, and lm_head
• rename the decoder components to RwkvDecoderLayer, linear_attn, and mlp
• rename dense projections to r_proj, k_proj, v_proj, o_proj, up_proj, and down_proj
• rename the linear-attention GroupNorm to g_norm
• rename public stateful-training fields to attention_shift, recurrent_state, and mlp_shift
• translate BlinkDL checkpoint keys once in temp/rwkv_pth2st.py, including blocks.0.ln0.* -> model.embedding_norm.*
• remove the obsolete RWKV-4 converter, which emitted a third incompatible naming scheme
• update LoRA targets, documentation, structure tests, conversion round trips, and inference-layout tests

The RWKV math, raw formula parameters, canonical head_size=64, FlashRWKV2 provider calls, Git-pinned tokenizer dependency, and support boundaries are unchanged. Qwen2Rwkv remains out of scope. External consumers such as vLLM will be updated separately to consume the new state-dict contract.

Coordination

No repository issue owns this follow-up, and the repository owner requested the rename directly. Duplicate-work checks found no overlapping open PR in this repository. Upstream already has huggingface#47780 and issue huggingface#47787, so this PR targets only the rwkv-rs integration repository and does not create competing upstream work.

Validation

• local RWKV modeling/GPU suite: 50 passed, 1 skipped
• local RWKV tokenizer suite: 11 passed
• focused shared linear-attention cache test: 1 passed
• rwkv-sha-pro6000x8, exact commit 38fd0b212a1e, official pinned vocabulary: 63 passed, 9 subtests passed
• make typing
• RWKV modeling structure check with all TRF rules enabled and cache disabled
• make check-repository-consistency
• wheel metadata: unconditional exact Git tokenizer commit; rwkv extra adds only FlashRWKV2==0.1.0a8
• make style
• git diff --check

Before submitting

• Read the contributor and PR checks
• Updated the public naming and conversion documentation
• Added/updated regression tests for every renamed public path

[github-actions]

Thank you for your contribution 🤗!

CI Security Gate — automatic approval blocked

This PR was not automatically approved for CI because the security gate failed.

Possible reasons:

• The PR touches 50 or more files — only PRs with fewer than 50 changed files are automatically approved
• A changed file is outside the allowed directories (src/, tests/, docs/, utils/), has a disallowed extension (only .py, .txt, .md permitted outside tests/ and docs/), or is not .md/.yml inside docs/ — this covers files the PR deletes or renames, not only the ones it edits
• A new high-severity security issue was detected in the changed Python files (Bandit check)
• The PR touches a path this repository protects from untrusted PRs, such as the file that decides who reviews it — a maintainer must make that change in a separate PR

See the workflow run for the exact violations.

A maintainer can review and manually approve CI if a finding is a false positive.

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: rwkv

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 33057174854
Result: failure | Grafana metrics are not available yet.

Code quality check failed: test jobs were skipped. Fix the code quality issues and push again to run tests.