YanTianlong commited on
Commit
f7bbebe
·
1 Parent(s): b961f46

Record model storage verification

Browse files
Files changed (2) hide show
  1. docs/todo.md +4 -0
  2. docs/work-log.md +29 -0
docs/todo.md CHANGED
@@ -14,6 +14,8 @@ from scaffold to a working Gradio app.
14
  - [x] Copy the validated API workflow into `workflows/voicegate_api.json`.
15
  - [x] Confirm SSH access to the running Space container and document the
16
  runbook.
 
 
17
 
18
  ## Phase 1: Repository Hygiene
19
 
@@ -117,6 +119,8 @@ from scaffold to a working Gradio app.
117
  - [ ] Is `flash_attention_2` available and reliable in the ZeroGPU environment?
118
  - [ ] Does VoxCPM2 fit comfortably in ZeroGPU memory with ASR and
119
  MelBandRoFormer in the same run?
 
 
120
  - [ ] Should the first public demo disable background separation to reduce
121
  runtime and memory pressure?
122
  - [ ] What maximum uploaded audio/video duration should the first version allow?
 
14
  - [x] Copy the validated API workflow into `workflows/voicegate_api.json`.
15
  - [x] Confirm SSH access to the running Space container and document the
16
  runbook.
17
+ - [x] Confirm `DEEPSEEK_API_KEY` is visible in the Space without printing it.
18
+ - [x] Download VoxCPM2 and MelBand RoFormer to persistent Space storage.
19
 
20
  ## Phase 1: Repository Hygiene
21
 
 
119
  - [ ] Is `flash_attention_2` available and reliable in the ZeroGPU environment?
120
  - [ ] Does VoxCPM2 fit comfortably in ZeroGPU memory with ASR and
121
  MelBandRoFormer in the same run?
122
+ - [x] Where should large model files live? `/data/voicegate_models`, with
123
+ symlinks into ComfyUI's expected model directories.
124
  - [ ] Should the first public demo disable background separation to reduce
125
  runtime and memory pressure?
126
  - [ ] What maximum uploaded audio/video duration should the first version allow?
docs/work-log.md CHANGED
@@ -341,6 +341,35 @@ Override the root with:
341
  VOICEGATE_MODEL_ROOT
342
  ```
343
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
344
  ## Current Known Good Commits
345
 
346
  - `683b147` Add ComfyUI runtime bootstrap scripts
 
341
  VOICEGATE_MODEL_ROOT
342
  ```
343
 
344
+ On 2026-06-05, the first two explicit ComfyUI-path models were downloaded to
345
+ persistent storage:
346
+
347
+ ```text
348
+ /data/voicegate_models/voxcpm/VoxCPM2/model.safetensors
349
+ /data/voicegate_models/voxcpm/VoxCPM2/audiovae.pth
350
+ /data/voicegate_models/diffusion_models/MelBandRoFormer_comfy/MelBandRoformer_fp32.safetensors
351
+ ```
352
+
353
+ Verified symlinks:
354
+
355
+ ```text
356
+ /home/user/app/ComfyUI/models/voxcpm/VoxCPM2
357
+ -> /data/voicegate_models/voxcpm/VoxCPM2
358
+
359
+ /home/user/app/ComfyUI/models/diffusion_models/MelBandRoFormer_comfy
360
+ -> /data/voicegate_models/diffusion_models/MelBandRoFormer_comfy
361
+ ```
362
+
363
+ `DEEPSEEK_API_KEY` was also verified as present in the Space environment without
364
+ printing its value.
365
+
366
+ Model download pitfall:
367
+
368
+ - `huggingface-cli download` is deprecated and failed in the Space.
369
+ - `hf download` also failed because of a CLI dependency compatibility issue.
370
+ - `scripts/bootstrap_comfy.py` now uses the `huggingface_hub` Python API
371
+ directly for model downloads.
372
+
373
  ## Current Known Good Commits
374
 
375
  - `683b147` Add ComfyUI runtime bootstrap scripts