setup failed: expected Layer Assignment

Start with the cake_core:🍰:sharding setup and handshake messages, then compare the topology.yml path with the working no-topology path. Reproduce the provided cake run command and inspect the Hello and LayerAssignment exchange. Done means the Android worker receives its layer assignment and the topology-based run proceeds without the setup error.

Summay

I build cake AMD --feature vulkan and cake for Android from the current master, and it is working ok in my PC, I can do cake run model "hello", but was trying to join it with a Android worker and I have the next issue in the phone:

9-18 21:33:55.123  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] zero-config mode: waiting for master discovery...
09-18 21:33:55.123  7404  7475 I cake-mobile: cake_core::cake::sharding: detected 1 GPU(s):
09-18 21:33:55.124  7404  7475 I cake-mobile: cake_core::cake::sharding:   CPU (aarch64) — 7.3 GiB (~2.0 TFLOPS)
09-18 21:33:55.124  7404  7475 I cake-mobile: cake_core::cake::sharding: listening on 0.0.0.0:10128 (setup mode)
09-18 21:33:55.124  7404  7475 I cake-mobile: cake_core::cake::sharding::discovery: listening for discovery queries on UDP port 10127
09-18 21:33:55.124  7404  7475 I cake-mobile: cake_core::cake::sharding: waiting for master to connect and assign layers...
09-18 21:33:55.124  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] status: {"stage":"discovery","message":"Waiting for master...","progress":0.0000}
09-18 21:34:36.936  7404  7475 I cake-mobile: cake_core::cake::sharding: [192.168.178.145:47088] master connected
09-18 21:34:36.936  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] status: {"stage":"connected","message":"Master connected (192.168.178.145:47088)","progress":0.0000}
09-18 21:34:36.956  7404  7475 I cake-mobile: cake_core::cake::sharding: [192.168.178.145:47088] authenticated
09-18 21:34:36.956  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] status: {"stage":"authenticated","message":"Authenticated with master","progress":0.0000}
09-18 21:34:36.958  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] status: {"stage":"error","message":"Setup failed: expected LayerAssignment, got Hello","progress":0.0000}
09-18 21:34:36.958  7404  7475 I cake-mobile: cake_mobile: [cake-mobile] shutting down tokio runtime...

Desktop

In the desktop I wrote my topology like this:

cpgreen:
  host: '192.168.178.183:10128'
  description: 'cpgreen'
  layers:
    - 'model.layers.0-5'

And launch cake with:

(main) avp:cakeȹ ./target/release/cake run Qwen/Qwen3-1.7B "Please give an adb line for return the size of Android apk?" --cluster-key "qwerty123" 
--topology topology.yml                                                                                                                            
[2026-09-18T20:13:32Z INFO ] [Master] dtype=F16 device=Cpu mem=10.3 MiB                                                                            
[2026-09-18T20:13:32Z INFO ] model 'Qwen/Qwen3-1.7B' found in cache at /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc
86ccca08cf5af4e1e306ecf908b1ad5e                                                                                                                   
[2026-09-18T20:13:32Z INFO ] loading topology from topology.yml                                                                                    
[2026-09-18T20:13:32Z INFO ] text model architecture: Qwen3                                                                                        
[2026-09-18T20:13:32Z INFO ] loading Qwen3 configuration from /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08c
f5af4e1e306ecf908b1ad5e/config.json                                                                                                                
[2026-09-18T20:13:32Z INFO ] loading 2 of 2 shard file(s) for local layers                                                                         
[2026-09-18T20:13:32Z INFO ] caching shard 1/2 (model-00001-of-00002.safetensors) ...                                                              
[2026-09-18T20:13:34Z INFO ] caching shard 2/2 (model-00002-of-00002.safetensors) ...                                                              
[2026-09-18T20:13:34Z INFO ] pre-cached 3.8 GiB in 1.8s                                                                                            
[2026-09-18T20:13:34Z INFO ] Vulkan backend (ash): AMD Radeon Graphics (RADV RENOIR)                                                               
[2026-09-18T20:13:34Z INFO ] UMA detected — using zero-copy DEVICE_LOCAL|HOST_VISIBLE buffers
[2026-09-18T20:13:34Z INFO ] using Vulkan backend for GPU-accelerated ops 
[2026-09-18T20:13:34Z INFO ] loading embeddings (prefix=model) ...
[2026-09-18T20:13:35Z INFO ] loading lm_head ...
[2026-09-18T20:13:35Z INFO ]   using tied word embeddings (lm_head = embed_tokens)
[2026-09-18T20:13:42Z INFO ] loading model.norm ...
[2026-09-18T20:13:42Z INFO ] loading 28 blocks ...
[2026-09-18T20:13:42Z INFO ] loading model.layers.6 ...
2026-09-18T20:13:43Z INFO ] loading model.layers.7 ...                                  
[2026-09-18T20:13:58Z INFO ] loading model.layers.27 ...
...                
[2026-09-18T20:13:58Z INFO ] connecting model.layers.0 to 192.168.178.183:10128 ...                                                                                           

Error: error receiving response for Hello: early eof         

Without topology

Running only cake everything works fine

(main) avp:cakeȹ ./target/release/cake run Qwen/Qwen3-1.7B "Please give an adb line for return the size of Android apk?" --cluster-key "qwerty123"
[2026-09-18T20:18:00Z INFO ] model 'Qwen/Qwen3-1.7B' found in cache at /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e
[2026-09-18T20:18:00Z INFO ] model has 28 transformer layers (prefix: model.layers)
[2026-09-18T20:18:00Z INFO ] discovering workers (timeout: 10s)...
[2026-09-18T20:18:10Z INFO ] discovery complete: 0 worker(s) found
[2026-09-18T20:18:10Z WARN ] no workers discovered — all layers will be loaded locally
[2026-09-18T20:18:10Z INFO ] [Master] dtype=F16 device=Cpu mem=14.7 MiB
[2026-09-18T20:18:10Z INFO ] model 'Qwen/Qwen3-1.7B' found in cache at /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e
[2026-09-18T20:18:10Z INFO ] text model architecture: Qwen3
[2026-09-18T20:18:10Z INFO ] loading Qwen3 configuration from /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json
[2026-09-18T20:18:10Z INFO ] loading tensors from /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/model.safetensors.index.json ...
[2026-09-18T20:18:10Z INFO ] caching shard 1/2 (model-00001-of-00002.safetensors) ...
[2026-09-18T20:18:12Z INFO ] caching shard 2/2 (model-00002-of-00002.safetensors) ...
[2026-09-18T20:18:12Z INFO ] pre-cached 3.8 GiB in 1.9s
[2026-09-18T20:18:12Z INFO ] Vulkan backend (ash): AMD Radeon Graphics (RADV RENOIR)
[2026-09-18T20:18:12Z INFO ] UMA detected — using zero-copy DEVICE_LOCAL|HOST_VISIBLE buffers
[2026-09-18T20:18:12Z INFO ] using Vulkan backend for GPU-accelerated ops
[2026-09-18T20:18:12Z INFO ] loading embeddings (prefix=model) ...
[2026-09-18T20:18:13Z INFO ] loading lm_head ...
[2026-09-18T20:18:13Z INFO ]   using tied word embeddings (lm_head = embed_tokens)
[2026-09-18T20:18:21Z INFO ] loading model.norm ...
[2026-09-18T20:18:21Z INFO ] loading 28 blocks ...
[2026-09-18T20:18:21Z INFO ] loading model.layers.0 ...
[2026-09-18T20:18:21Z INFO ] loading model.layers.1 ..
...
[2026-09-18T20:18:42Z INFO ]   model.layers.27 (local)
[2026-09-18T20:18:42Z INFO ] loading tokenizer from /home/avp/.cache/huggingface/hub/models--Qwen--Qwen3-1.7B/snapshots/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/tokenizer.json
[2026-09-18T20:18:42Z INFO ] model loaded - mem=6.8 GiB
[2026-09-18T20:18:42Z INFO ] starting the inference loop (mem=6.8 GiB)
    
    
<think>
Okay, the user is asking for an adb command to get the size of an Android APK. Let me think about how to approach this.

First, I remember that adb (Android Debug Bridge) has commands to interact with devices. The basic structure would be something like 'adb shell' followed by a command.

The size of an APK can be found using the 'dumpsys package' command, but I'm not sure if that gives the file size directly. Wait, maybe there's a better way. Oh right, there's a command called 'pm path' which shows the package path, and then maybe some other commands.

Thanks for any help :)