Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Server Startup Issue #110

Open
ItsCRC opened this issue May 4, 2024 · 5 comments
Open

Server Startup Issue #110

ItsCRC opened this issue May 4, 2024 · 5 comments

Comments

@ItsCRC
Copy link

ItsCRC commented May 4, 2024

Hi All,

I am trying to run cargo run --release and I get the following:
Finished release profile [optimized] target(s) in 0.21s
Running target/release/ai00_server
2024-05-04T04:35:17.007Z INFO [ai00_server] reading config assets/configs/Config.toml...
2024-05-04T04:35:17.008Z INFO [ai00_server::middleware] ModelInfo {
version: V5,
num_layer: 32,
num_emb: 2560,
num_hidden: 8960,
num_vocab: 65536,
num_head: 40,
time_mix_adapter_size: 0,
time_decay_adapter_size: 0,
}
2024-05-04T04:35:17.008Z INFO [ai00_server::middleware] type: SafeTensors
2024-05-04T04:35:17.013Z WARN [wgpu_hal::gles::egl] EGL_MESA_platform_surfaceless not available. Using default platform
2024-05-04T04:35:17.027Z WARN [wgpu_hal::gles::egl] No config found!
2024-05-04T04:35:17.027Z WARN [wgpu_hal::gles::egl] No config found!
2024-05-04T04:35:17.048Z INFO [ai00_server] server started at 0.0.0.0:65530 with tls
2024-05-04T04:35:17.071Z ERROR [ai00_server::middleware] reload model failed: failed to request device

Currently, it is Cpu ins Config.toml. Also, I have Nvidia 22C (24GB) GPU and it gives the same error when I set it to Gpu.
Result of lsb_release -a:

No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 18.04.6 LTS
Release:        18.04
Codename:       bionic

Config.toml:

[model]
embed_device = "Cpu"                                       # Device to put the embed tensor ("Cpu" or "Gpu").
max_batch = 8                                              # The maximum batches that are cached on GPU.
model_name = "RWKV-5-World-3B-v2-20231113-ctx4096.st" # Name of the model.
model_path = "assets/models"                               # Path to the folder containing all models.
quant = 32                                                  # Layers to be quantized.
quant_type = "Int8"                                        # Quantization type ("Int8" or "NF4").
stop = ["\n\n"]                                            # Additional stop words in generation.
token_chunk_size = 128                                     # Size of token chunk that is inferred at once. For high end GPUs, this could be 64 or 128 (faster).

Can anyone help? Thanks

@cryscan
Copy link
Collaborator

cryscan commented May 5, 2024

Ai00 requires Vulkan. Does your GPU support Vulkan and Do you have Vulkan driver installed? You may check vulkaninfo.

@ItsCRC
Copy link
Author

ItsCRC commented May 5, 2024

I just checked. Vulkan drivers are not installed. Let me install and then come back.

@ItsCRC
Copy link
Author

ItsCRC commented May 13, 2024

@cryscan I have A10 G Nvidia card and driver version is 515.65.01 with CUDA 11.7. I want to know how can I install vulkan drivers? I don't seem to find any good article to install them.

@lingfengchencn
Copy link

==========
VULKANINFO
==========

Vulkan Instance Version: 1.3.283

the some issues

2024-05-16T11:06:22.752Z INFO  [ai00_server::middleware] type: SafeTensors
error: XDG_RUNTIME_DIR not set in the environment.
error: XDG_RUNTIME_DIR not set in the environment.
2024-05-16T11:06:22.802Z ERROR [ai00_server::middleware] reload model failed: failed to request device
2024-05-16T11:06:22.807Z INFO  [ai00_server] server started at 0.0.0.0:65530 with tls

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants