firebase-ios-sdk
firebase-ios-sdk copied to clipboard
firebase
Remote Config not producing throttling error accurately
Description
In trying to add non fatal logging specifically for throttling I have discovered what appears to be a bug. Whilst the error status of the connection is 429, as shown in the logs, and the catch let error as RemoteConfig hitting, its error code is never that of throttling (8002), its always internal errors code (8003). (minimumFetchInterval = 0 seconds btw)
Thanks!
Reproducing the issue
func forceThrottle() async {
let RemoteConfig = RemoteConfig.remoteConfg()
do {
for i in 0...1000 {
let foo = try await remoteConfig.fetchAndActivate()
}
} catch let error as RemoteConfigError {
//prints false
print("Error code is correct: \(error.code == .throttled)")
} catch {}
}
Firebase SDK Version
10.21.0
Xcode Version
15.2
Installation Method
Swift Package Manager
Firebase Product(s)
AB Testing, Analytics, App Distribution, Crashlytics, Performance, Remote Config
Targeted Platforms
iOS
Relevant Log Output
10.21.0 - [FirebaseRemoteConfig][I-RCN000026] RCN fetch failure. Response http error code: 429
If using Swift Package Manager, the project's Package.resolved
Expand Package.resolved snippet
Replace this line with the contents of your Package.resolved.
If using CocoaPods, the project's Podfile.lock
Expand Podfile.lock snippet
Replace this line with the contents of your Podfile.lock!
Ollama is very close to supporting AMD GPUs through ROCm using mainline llama.cpp.
Having Vulkan support would be huge. ROCm is not that widely supported thorough AMD GPUs, plus it might allow supporting GPU accel on devices like Intel Arc or older AMD GPUs, like Navi 14
ggerganov/llama.cpp#2059 got merged
That backend seems to be more performant and works on more hardware including Intel GPUs.
Out of curiosity: does Vulkan runner support NPUs included in very recent generations of CPUs (e.g. Intel Ultra)? I remember Vulkan is a graphics framework, right? So probably it will not? Then if not, what would be the right (universal/uniformed) framework/tool for that?
Vulkan support would be wonderful. I have an AMD RX 6800 XT, and using KoboldCPP with Vulkan support made using LLMs go from slow-but-kinda-usable to near-instantaneous speeds. My desktop is running Gentoo, which is currently stuck on ROCm 5, and Ollama seems to use ROCm 6.
Vulkan support would be wonderful. I have an AMD RX 6800 XT, and using KoboldCPP with Vulkan support made using LLMs go from slow-but-kinda-usable to near-instantaneous speeds. My desktop is running Gentoo, which is currently stuck on ROCm 5, and Ollama seems to use ROCm 6.
@ProjectMoon I have experimental vulkan support here: https://github.com/ollama/ollama/pull/2578
I still haven't figured out how to automatically get the memory size from vulkan (especially for iGPUs). Otherwise it works really well. I get massive speedups on my intel iGPU. The default value of VRAM is hardcoded, you should change it for your device (making that automatic is the most important bit).
@ddpasa couldn't that be set so that it has no limit by default and reads such limit from an environment variable, if later a way to find the memory automatically is found it will be nice, but that should not be a blocker IMHO.
@ProjectMoon I have experimental vulkan support here: #2578
I tried it out! Unfortunately, ollama crashes when it tries to run a model, with some kind of illegal instruction error.
I still haven't figured out how to automatically get the memory size from vulkan (especially for iGPUs). Otherwise it works really well. I get massive speedups on my intel iGPU. The default value of VRAM is hardcoded, you should change it for your device (making that automatic is the most important bit).
Based on what I've seen, there doesn't seem to be a real way to get VRAM from video cards via API, in any standardized manner. Maybe best way for integrated GPUs is to just make a very small default VRAM size that can be overridden by the user as an env var/command line option?
@ProjectMoon I have experimental vulkan support here: #2578
I tried it out! Unfortunately, ollama crashes when it tries to run a model, with some kind of illegal instruction error.
Can you share the error? It's likely related to llama.cpp and not to ollama, but maybe we can forward it there.