diff --git a/persistent-memory.md b/persistent-memory.md index 85ff647..aab8de1 100644 --- a/persistent-memory.md +++ b/persistent-memory.md @@ -176,7 +176,7 @@ Every vLLM seat on a shared card (GPU 0, GPU 1) has byte-pinned KV (`--kv-cache- ### Scriberr (fv-ml1 GPU 3) -- **Diarization FIXED 2026-10-06 ~1320 (Prime: "whisper doesn't work at all, need diarized transcripts").** Three faults: no `HF_TOKEN` (pyannote is gated; now passed via compose, vault `fv-ml1/hf-token`); torchcodec 0.16 in the pyannote env against torch 2.8 (`AudioDecoder` undefined; pinned to 0.7 by hand, durable as `patches/0003`, applied at the next rebuild); and the WhisperX adapter's CPU/small/float32 defaults. Both diarization paths verified on GPU 3. → `stacks/scriberr/README.md` § Diarization +- **Diarization FIXED 2026-10-06 ~1320 (Prime: "whisper doesn't work at all, need diarized transcripts").** Three faults: no `HF_TOKEN` (pyannote is gated). FINAL: the gated models were downloaded once, caches moved to /tank (`HF_HOME`/`TORCH_HOME`/`NLTK_DATA` under `/app/whisperx-env`), and `HF_TOKEN` is a placeholder (Prime 1330: "move the downloads and drop the token"). Cached gated models load with an invalid token; a NEW gated model needs one download with vault `fv-ml1/hf-token`. torchcodec 0.16 in the pyannote env against torch 2.8 (`AudioDecoder` undefined; pinned to 0.7 by hand, durable as `patches/0003`, applied at the next rebuild); and the WhisperX adapter's CPU/small/float32 defaults. Both diarization paths verified on GPU 3. → `stacks/scriberr/README.md` § Diarization - **LIVE `scriberr:local-blackwell-a353078-dropout2`:** upstream a353078 plus patch 0001 (overlap slicer) and patch 0002 (gap retry, `PARAKEET_MODEL_PATH`), carried LOCALLY ONLY (Prime 2026-10-01: no upstream). v3 stays. `scripts/scriberr-rebuild` re-applies both. → `persistent-memory.d/2026-09-30-scriberr-slicer-gap-retry-gpu3.md` - **URL is https://scriberr.nh3.phasefinal.com since 2026-10-01** (fronted by the fleet TLS caddy on nh3-dev; Caddyfile backup `.bak-20261001-scriberr`). Prime's "can't initialize recorder" was the SecureContext rule: `getUserMedia` is refused on http://. The HTTPS name is in `ALLOWED_ORIGINS` (`playbooks/scriberr-https-origin.yaml`); http://10.251.50.54:8080 still works except for recording. ⚠ `/opt/docker/compose/scriberr` is lkraven-owned, so `deploy-stack.sh` as infra-ops FAILS there (Permission denied); upload single files with `elway --upload --sudo --owner lkraven:lkraven`. diff --git a/stacks/scriberr/README.md b/stacks/scriberr/README.md index b136e87..b0c3ed5 100644 --- a/stacks/scriberr/README.md +++ b/stacks/scriberr/README.md @@ -138,11 +138,11 @@ rebuild — a local patch to carry, which is why it was not done. Prime: "whisper doesn't work at all, and I need diarized transcripts." There were three separate faults, and all three looked like "transcription is broken": -1. **No Hugging Face token.** pyannote's diarization models are gated. Every diarized job died at the diarize step: WhisperX with `GatedRepoError 401` on `pyannote/speaker-diarization-3.1`, but only after a full CPU transcription (12m45s for a 1 h file), and Parakeet+pyannote with "HuggingFace token is required". **Fix:** `HF_TOKEN` is passed through compose. Its value is in the host `.env` only (vault `fv-ml1/hf-token`, account lkraven, read role; access to `speaker-diarization-3.1`, `segmentation-3.0` and `speaker-diarization-community-1` was checked). +1. **No Hugging Face token.** pyannote's diarization models are gated. Every diarized job died at the diarize step: WhisperX with `GatedRepoError 401` on `pyannote/speaker-diarization-3.1`, but only after a full CPU transcription (12m45s for a 1 h file), and Parakeet+pyannote with "HuggingFace token is required". **Fix, final form (same day, Prime: "move the downloads and drop the token"):** the gated models were downloaded ONCE with a real token. All model caches now live on /tank (`HF_HOME`, `TORCH_HOME`, `NLTK_DATA` under `/app/whisperx-env`; before this they sat in the container layer and every recreate threw away ~6 GB). `HF_TOKEN` is a **placeholder** in compose, so no real credential lives in Scriberr. Measured: with an invalid token, cached gated models still load (huggingface_hub falls back to the cache on the 401), and public models still download (HTTP 200). The placeholder exists only because Scriberr refuses an EMPTY token. **A new gated model** needs one first download with the real token (vault `fv-ml1/hf-token`). 2. **The separate pyannote env had a torchcodec that cannot load.** uv resolved torchcodec 0.16.0 beside torch 2.8.0+cu128. 0.16 needs torch ≥2.11 and CUDA 13 (`libnvrtc.so.13`), so pyannote.audio 4 has no decoder and fails with `name 'AudioDecoder' is not defined`. **Fix (live, by hand):** `torchcodec==0.7.*` (its table pairs 0.7 with torch 2.8), pinned in `/tank/scriberr/whisperx-env/pyannote/pyproject.toml`, then `uv lock && uv sync` as uid 10001; the old files are kept as `*.bak-20261006`. Scriberr only rewrites that pyproject if its import test fails at startup. **Durable fix:** `patches/0003-pyannote-env-pin-torchcodec.patch` pins torch/torchaudio 2.8 and torchcodec 0.7 in the embedded pyproject, so it lands at the next rebuild. 3. **WhisperX defaults to CPU.** The adapter's defaults are `device cpu`, `model small`, `compute_type float32` (`whisperx_adapter.go`). A quick transcribe with no profile ran exactly that. **Use** device `cuda`, model `large-v3`, compute `float16`. -**Verified on a 4-min, multi-speaker clip of a real upload, GPU 3:** WhisperX large-v3/cuda/float16 + `pyannote/speaker-diarization-3.1` exited 0 in 65 s with 2 speakers. The pyannote env script with `speaker-diarization-community-1` (the adapter's only option, used by the Parakeet profile) exited 0 in 11 s with 3 speakers. The two pipelines disagree on the speaker count, so set min/max speakers when you know it. WhisperX's own env (pyannote.audio 4.0.4) loads 3.1 fine and feeds the pipeline in-memory audio, so it never needed torchcodec. +**Re-verified with the placeholder token and the /tank cache (60 s clip): community-1 exited 0 in 8 s, WhisperX+3.1 exited 0 in 28 s, both 2 speakers, with no downloads and no 401s.** First verification, with the real token, on a 4-min multi-speaker clip of a real upload, GPU 3: WhisperX large-v3/cuda/float16 + `pyannote/speaker-diarization-3.1` exited 0 in 65 s with 2 speakers. The pyannote env script with `speaker-diarization-community-1` (the adapter's only option, used by the Parakeet profile) exited 0 in 11 s with 3 speakers. The two pipelines disagree on the speaker count, so set min/max speakers when you know it. WhisperX's own env (pyannote.audio 4.0.4) loads 3.1 fine and feeds the pipeline in-memory audio, so it never needed torchcodec. ## Optional: summarisation via the LiteLLM gateway diff --git a/stacks/scriberr/compose.yaml b/stacks/scriberr/compose.yaml index 6178430..52a0d03 100644 --- a/stacks/scriberr/compose.yaml +++ b/stacks/scriberr/compose.yaml @@ -72,15 +72,26 @@ services: # and takes out the Parakeet + Sortformer backends (WhisperX survives). # `copy` trades a little disk and time for it actually working. - UV_LINK_MODE=${SCRIBERR_UV_LINK_MODE:-copy} - # Hugging Face token for the GATED pyannote models (speaker-diarization-3.1, - # segmentation-3.0). Without it every diarized job dies at the diarize step: - # WhisperX with a 401 GatedRepoError AFTER the whole transcription has run, and - # Parakeet+pyannote with "HuggingFace token is required". So it looks as if - # nothing works. Scriberr's config and both adapters read HF_TOKEN - # (internal/config/config.go), and it reaches the uv subprocesses. Value: vault - # `fv-ml1/hf-token` (account lkraven, read role; access to the gated repos was - # checked 2026-10-06). It lives only in the host .env, never in git. - - HF_TOKEN=${HF_TOKEN:-} + # ── Model caches on /tank, not in the container layer (Prime, 2026-10-06) ── + # Before this they lived under /home/appuser, which is the container's writable + # layer, so EVERY recreate threw away ~6 GB (Whisper large-v3, pyannote, the + # alignment model, NLTK) and the next job had to download it all again. + # /app/whisperx-env is the persistent /tank/scriberr/whisperx-env bind mount. + - HF_HOME=/app/whisperx-env/hf-cache + - TORCH_HOME=/app/whisperx-env/torch-cache + - NLTK_DATA=/app/whisperx-env/nltk_data + # ⚠ HF_TOKEN is a PLACEHOLDER on purpose. No real credential lives here. + # pyannote's diarization models are gated: downloading them needs a token whose + # account accepted their terms. LOADING an already-downloaded copy does not. + # Measured 2026-10-06: with an invalid token, huggingface_hub's check gets a 401 + # and falls back to the cached copy, both online and with HF_HUB_OFFLINE=1. + # Public repos (any faster-whisper size) still download with an invalid token + # (HTTP 200). The placeholder exists only because Scriberr refuses to start + # pyannote diarization when HF_TOKEN is EMPTY (pyannote_adapter.go checks that + # it is non-empty, never that it is valid). + # To add a NEW gated model: run its first download once with a real token + # (vault `fv-ml1/hf-token`) set for that one run, then put the placeholder back. + - HF_TOKEN=${HF_TOKEN:-hf_placeholder_models_cached_on_tank} # ── Memory settings, measured while Scriberr shared GPU 1 (2026-09-30; it moved to # GPU 3 at 1322 the same day, so the 5.5 GB budget no longer binds, but the values stand). Prime: Scriberr shared GPU 1 with # intern-decision, which holds ~9.7 GB resting / 10.3 GB peak). ──────────