Nate Nethercott

Question about weight quantization methodology memory savings

1

Thanks for your quick implementation! I was reading through `bitnet/bitbnet_b158.py` and just had a short question. In your implementation of `quantize_weights` you use the same procedure as outlined in the...

RuntimeError: Expected all tensors to be on the same device using Deepspeed with QLoRA and DPOTrainer

2

Couldn't find any similar other issues in `accelerate`, `peft`, or `trl` so I'm opening one here. When using the DPOTrainer on a single GPU with QLoRA I have no issues,...

This fix allows for loading lora checkpoints based on Qwen without this we default to `LlavaLlamaForCausalLM`

Implement Hamming distance for binary strings

9

# Pull Request Had some time this weekend and wanted to try to incorporate hamming into arroy ! Check comments in the PR to explain some of the logic ##...

# Pull Request Timings for facet post-processing give the illusion of a missing trace since `Progress` retains only a single measurement per `PostProcessingFacets` type. `Progress.update_progress` [truncates its internal list of...

[ann-benchmarks](https://github.com/erikbern/ann-benchmarks) is a benchmarking environment for ann implementations written in Python by the same guy who did annoy. It would be cool to submit arroy there to see how it...