Why the handoff needs the exact semantic adapter insertion point
Constraint: The next session should not spend time rediscovering where the real MERT/MuQ adapter belongs in the song-centric pipeline Rejected: Leave the adapter step as a generic future task | does not identify the concrete file and function to change Confidence: high Scope-risk: narrow Directive: Keep future semantic-adapter handoffs anchored on enrich_songcentric_manifest_with_local_features.py unless the host pipeline entrypoint changes Tested: markdown link check on /workspace/docs after adding the semantic adapter handoff note Not-tested: No runtime install or adapter implementation yet; this commit records the verified insertion point only
Showing
1 changed file
with
48 additions
and
0 deletions
| ... | @@ -158,3 +158,51 @@ flowchart TD | ... | @@ -158,3 +158,51 @@ flowchart TD |
| 158 | ## 一句话 handoff | 158 | ## 一句话 handoff |
| 159 | 159 | ||
| 160 | > 下次不要再从总方案争论开始,直接跑 song-centric runner;如果 exact 正常、semantic 仍 fallback,就继续补真实 semantic adapter 和依赖。 | 160 | > 下次不要再从总方案争论开始,直接跑 song-centric runner;如果 exact 正常、semantic 仍 fallback,就继续补真实 semantic adapter 和依赖。 |
| 161 | |||
| 162 | --- | ||
| 163 | |||
| 164 | ## 10. 真实 semantic adapter 下一步应该接到哪里 | ||
| 165 | |||
| 166 | 当前最直接的接入点已经明确: | ||
| 167 | |||
| 168 | - 入口脚本:`acr-engine/scripts/enrich_songcentric_manifest_with_local_features.py` | ||
| 169 | - 关键函数:`build_semantic_feature(...)` | ||
| 170 | |||
| 171 | ### 当前真实状态 | ||
| 172 | |||
| 173 | - exact lane 已优先复用 `ChromaprintMatcher` | ||
| 174 | - semantic lane 还没有真实接入 `MERT / MuQ` | ||
| 175 | - runtime 就绪时,当前仍只会产出: | ||
| 176 | - `model_name = semantic_runtime_ready_placeholder` | ||
| 177 | - runtime 不就绪时,会走: | ||
| 178 | - `model_name = local_wavehash_embed` | ||
| 179 | |||
| 180 | ### fresh 依赖检查事实 | ||
| 181 | |||
| 182 | 当前 host 仍缺: | ||
| 183 | - `torch` | ||
| 184 | - `torchaudio` | ||
| 185 | - `transformers` | ||
| 186 | |||
| 187 | ### 下次 session 最直接的实现顺序 | ||
| 188 | |||
| 189 | 1. 安装 `torch / torchaudio / transformers` | ||
| 190 | 2. 在 `build_semantic_feature(...)` 内接真实 `MERT` 或 `MuQ` adapter | ||
| 191 | 3. 保留当前 `local_wavehash_embed` fallback 不删 | ||
| 192 | 4. 重跑: | ||
| 193 | |||
| 194 | ```bash | ||
| 195 | cd /workspace | ||
| 196 | /usr/local/miniconda3/bin/python acr-engine/scripts/run_songcentric_directory_pipeline_live.py \ | ||
| 197 | --dsn 'postgres://d2:d2pass@127.0.0.1:5432/d2' \ | ||
| 198 | --schema acr_songcentric_test \ | ||
| 199 | --input-root acr-engine/data/songcentric_builder_smoke \ | ||
| 200 | --output-dir acr-engine/data/pgvector_eval/music20 | ||
| 201 | ``` | ||
| 202 | |||
| 203 | ### 期望看到的 fresh 指标变化 | ||
| 204 | |||
| 205 | - `semantic_runtime_available = true` | ||
| 206 | - `semantic_runtime_ready_count > 0` | ||
| 207 | - `semantic_fallback_count` 明显下降或归零 | ||
| 208 | ... | ... |
-
Please register or sign in to post a comment