Docker SDK model placement status

[!NOTE] For the latest implementation status, please refer to Functional Implementation Status (Remaining Functionality).

Update date: 2026-04-09

Confirmation results: - Starting SDK API container: evospikenet-api - Public port: http://127.0.0.1:8000 - There is only one item in the data_artifacts table - artifact content is model_metadata.json, not weights

artifact content:

{"source_model": "src/models/model_evospikenet.pt"}

Fact check: - src/models/model_evospikenet.pt does not exist in the host workspace - /home/appuser/app/src/models/model_evospikenet.pt does not exist in the evospikenet-api container either. - saved_models Docker volume is also empty

Additional information: - evospikenet-llm-trainer-gpu is restarting repeatedly, and the log shows import failure of bitsandbytes.BitsAndBytesConfig. - Therefore, it is highly likely that the model has not been generated or saved yet.

Conclusion: - The current Docker SDK has "production model reference metadata" but does not have "body weight". - To proceed with Phase 3 conversion/validation on a production model, one of the following is required:

1. Correct the dependency inconsistency in the learning container and generate the actual .pt / .pth
2. Re-register the existing production weights as SDK artifacts

Auxiliary tools: - Added scripts/device/fetch_sdk_model_artifact.py - You can try to fetch both metadata and actual files and output the results to sdk_model_fetch_result.json