Dmitry
Dmitry
Ok, so I've made some measurements that I'm now confident of. Please note that I redid the tests using manual measurement since I discovered that for some reason `tracing` introduced...
> From that I am seeing here I think the most likely scenario is the obvious one: Faster execution of the compiled code cannot recuperate the big up front costs...
> Correct me if I am wrong: Your measurement always measure compilation + execution. This means even if the execution is much faster it could still be offset by some...
> But why would it be the host function calls? You measured them as roughly equal in time. I guess the A/B test I mentioned would be the next step:...
> 1600 host function calls for a single deploy of an ERC20 contract? That seems too much. Or is it a batch of calls? Unfortunately, not. It's for a single...
> To confirm that assumption @0x7CFE can disable gas metering and do one more measurement. I no longer work on this.
> However I can't seem to find which modes are missing. Those both seem to be in the Mode enum? Also I'd like to know why they're mentioned as `ScaleType`...
> I haven't lost interest in this library, just need some help from you guys because this is a challenging project :) I'm really glad to hear that. Thank you...
Have similar issues with `ggml-vicuna-13b-4bit.bin` model. I downloaded it quite a while ago, around a time when Vicuna was first announced. ``` Loaded hyperparameters thread 'main' panicked at 'Failed to...
> Anyone that owns the Playstation model can talk to me to try to add support to [new-lg4ff](https://github.com/berarma/new-lg4ff). I own the PS model.