Fix MLA cache shape init without layerid

[Socratesa]

Motivation

cache manager 临时构造 backend 只为了算 cache shape，没有 layer_id，self.layerid 报错

Modifications

Usage or Command

Accuracy Tests

Checklist

• Add at least a tag in the PR title.
  • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
  • You can add new tags based on the PR content, but the semantics must be clear.
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[Copilot]

Pull request overview

该 PR 修复了在 Cache Manager 为了计算 MLA KV cache shape 而“临时构造 backend”时可能缺失 layer_id，从而导致 get_kv_cache_shape() 访问 self.layer_id 报错的问题，使该函数在缺少 layer_id 的场景下能够继续工作。

Changes:

• 在 get_kv_cache_shape() 中将 self.layer_id 改为通过 getattr(self, "layer_id", None) 获取，避免属性不存在时抛异常。
• 仅在 layer_id 存在时才使用 window_attn_skip_freq[layer_id] 的分支判断。

PR 标题/描述检查（需要补充）

• 标题缺少规定的 tag，建议改为：[BugFix] Fix MLA cache shape init without layerid
• 描述中的 Modifications / Usage or Command / Accuracy Tests 基本为空；若无测试/命令/精度结果，建议在对应小节说明原因与验证方式。
• 该改动属于 bugfix，通常不需要额外文档；但若该行为会影响使用方（例如 cache manager 的构造方式），建议确认是否需要在相关说明中补一句。

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[PaddlePaddle-bot]

🤖 Paddle-CI-Agent | pr_review | 2026-06-23 21:44:21

📋 Review 摘要

PR 概述：修复 MLA cache shape 计算在无 layer_id 临时 backend 上的异常。
变更范围：fastdeploy/model_executor/layers/attention/mla_attention_backend.py
影响面 Tag：[OP] [KVCache]

问题

级别	文件	概述
🔴 Bug	fastdeploy/model_executor/layers/attention/mla_attention_backend.py:626	无 layer_id 时默认普通 MLA shape，会导致 window_attn_skip_freq 层的 KV cache 维度被低估

📝 PR 规范检查

标题缺少官方 Tag，且 Modifications、Usage or Command、Accuracy Tests 仍为空/占位内容。可直接替换为：

标题建议（可直接复制）：

• [BugFix] Fix MLA cache shape init without layer_id

PR 描述建议（点击展开，可直接复制）

## Motivation
cache manager 临时构造 MLA attention backend 只用于计算 cache shape 时没有 `layer_id`，原逻辑直接访问 `self.layer_id` 会报错。

## Modifications
- 在 `MLAAttentionBackend.get_kv_cache_shape` 中通过 `getattr(self, "layer_id", None)` 读取 layer id。
- 仅在存在 `layer_id` 且当前层命中 `window_attn_skip_freq` 时使用 SWA/FP8 MLA cache shape。

## Usage or Command
N/A

## Accuracy Tests
N/A

## Checklist

- [x] Add at least a tag in the PR title.
  - Tag list: [`[FDConfig]`,`[APIServer]`,`[Engine]`, `[Scheduler]`, `[PD Disaggregation]`, `[Executor]`, `[Graph Optimization]`, `[Speculative Decoding]`, `[RL]`, `[Models]`, `[Quantization]`, `[Loader]`, `[OP]`, `[KVCache]`, `[DataProcessor]`, `[BugFix]`, `[Docs]`, `[CI]`, `[Optimization]`, `[Feature]`, `[Benchmark]`, `[Others]`, `[XPU]`, `[HPU]`, `[GCU]`, `[DCU]`, `[Iluvatar]`, `[Metax]`]
  - You can add new tags based on the PR content, but the semantics must be clear.
- [ ] Format your code, run `pre-commit` before commit.
- [ ] Add unit tests. Please write the reason in this PR if no unit tests.
- [ ] Provide accuracy results.
- [ ] If the current PR is submitting to the `release` branch, make sure the PR has been submitted to the `develop` branch, then cherry-pick it to the `release` branch with the `[Cherry-Pick]` PR tag.

总体评价

当前改动能避免 AttributeError，但会把启用 window_attn_skip_freq 的 MLA cache 初始化路径变成静默的 shape 低估，建议修复后再合入。

[PaddlePaddle-bot]

🔴 Bug layer_id is None 时直接跳过 window_attn_skip_freq 会把含 SWA/窗口层的 MLA cache shape 降成普通 MLA shape。

PrefixCacheManager._get_kv_cache_shape 会临时构造 backend 后立即调用 get_kv_cache_shape，这类实例没有 layer_id；V1 CacheController.initialize_kv_cache 也会在旧路径逐层设置 attn_backend.layer_id 之前调用 create_kv_cache。当前分支会返回 kv_lora_rank + qk_rope_head_dim 的普通 shape，但 runner 对 window_attn_skip_freq[i] == 1 的层使用 kv_lora_rank + 4 * (kv_lora_rank // 128) + 2 * qk_rope_head_dim 并按 uint8 cache attach/创建。这样 PR 虽然避免了 AttributeError，却会给窗口层分配或传递偏小的 IPC/cache shape，后续 attach 或 attention kernel 可能出现 shape mismatch 或越界读写。

建议修复方式：
让无 layer_id 的调用不要默认走普通 MLA shape。可以让 cache manager/cache controller 在调用前按层设置/传入 layer_id 并逐层取 shape；如果该接口必须返回单一 shape，则在 layer_id is None and any(window_attn_skip_freq) 时返回覆盖所有层的保守最大 shape，并同步 dtype/IPC 分配策略。