fpgaminer
fpgaminer
Are you still having this issue? Just checked from here (Firefox) and it's working okay.
That's fair, though a 10% performance hit is rather painful. To that end, here's my attempt to optimize `apply_rotary_pos_emb`: ``` def ref_apply_rotary_pos_emb(q, k, cos, sin, position_ids): gather_indices = position_ids[:, None,...
I can, yes. Which would be best: a fix for those two lines, or re-writing to calculate loss outside the model? The latter better matches how the guide explains things...
Hello @gante, Thanks for the response. I have no intention of using beam sampling myself. I'm bubbling up a bug report by @diegomontoya from my GPTQ-triton repo, that turned out...
I can rebase and fix this up when I get a chance, but I haven't seen any discussion on the API from the maintainers.
> Do you know where this constant 45.335163 comes from? I didn't write that Python code, but based on context I can give you my best guess. https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner/blob/fd76bc932ae04be2b27be769871105c4678af9dc/projects/VHDL_Xilinx_Port/miner.py#L389-L392 That code...
Ran into this issue as well. Seems like adding a check before (https://github.com/huggingface/safetensors/blob/079781fd0dc455ba0fe851e2b4507c33d0c0d407/bindings/python/py_src/safetensors/torch.py#L389) would workaround the issue. Check if the buffer is length 0, and if so, create an empty...
Thank you for reporting the bug. Looks like a problem on Aersia's side. I could work around the problem, but it seems that there is an official HTML5 player these...