[Bug]: PyTorch model, anyway to force to run in FP32?

[yi]

Describe the bug

hi, just tried fast-stable-diffusion, the UI is great and intuitive 👍

The problem I met is: I cant find a setting to force the program run in FP32. And It seems to run in FP16 by default, which is extremely slow on the gpu I'm using (1:32).

Could there be an command line argument, like the --no-half and --no-half-vae in A111111 ?

---

I've truned on USE FP32 in settings > API. But it still run in FP16

Reproduction

just run ./voltaml-manager

Expected behavior

some sort of command line argument, like the --no-half and --no-half-vae in A111111 ?

Installation Method

Local

Branch

Main

System Info

Linux

Logs

20:46:29 | core.config.config | INFO » Loading configuration from data/settings.json
20:46:29 | core.config.config | INFO » data/settings.json not found, creating a new one
20:46:29 | core.config.config | INFO » Saving configuration to data/settings.json
20:46:29 | core.config.config | INFO » Configuration saved to data/settings.json
20:46:29 | __main__ | INFO » Device: cuda:0
>>>>> 20:46:29 | __main__ | INFO » Precision: FP16
20:46:29 | api.app | INFO » No FastAPI Analytics key provided, skipping
INFO:     Started server process [3118742]

---

[After truned on `USE FP32` in settings > API. But it still run in FP16]

INFO:     127.0.0.1:58916 - "GET /api/models/loaded HTTP/1.1" 200 OK
21:13:04 | core.inference.pytorch | INFO » Loading v1-5-pruned-emaonly.safetensors with f16
21:13:04 | core.inference.functions | INFO » Loading v1-5-pruned-emaonly.safetensors with f16
21:13:04 | core.inference.functions | INFO » Loading model as safetensors

Additional context

No response

Validations

• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• I am writing the issue in English.

[Stax124]

It should be patched in the latest release. Can you please try again on the newest patch?