hazratisulton

Press "Run Filter" on Finetune tab. **Expected result:** Filtering is working **Actual result:**  Filtering not working. In logs: `REJECTED FILTER refact/code_contrast/format_2023q2/el_file.py Boolean value of Tensor with more than one...

Allow renaming of LoRA's Completed Runs

2

By allowing users to rename their completed runs, they can quickly identify and distinguish between different runs. - Give names to loras when training, "all files", "only python" - what...

Latest lora checkpoints for deepseek-coder/5.7b/mqa-base only generate 1 token to some requests

1

Use latest lora checkpoint for deepseek-coder/5.7b/mqa-base (iter0010-testloss0.683 in this example)  Try first completion. It's working:  Accept completion and press Enter:  **Expected result:** Completion is working and generating...

https://github.com/smallcloudai/refact/assets/140423660/47acabdd-a726-459b-bcb9-f6666abd1a6d No completion after pressing Enter and going to new line. No completion in this case even if use force completion.  Force completion works if cursor is at the...

Handle OOM better on smaller/older GPUs, or bigger models on regular GPUs

4

```-- 4522 -- 20230930 23:34:53 FTUNE /model load 20853.9ms downloads-refact-1 | -- 4522 -- FILTER explanation: initial loss too big calculated on a single file, threshold is 3.500. Likely downloads-refact-1...

OOM issues cause auto-completion and chat response to stop

1

Preconditions: Run with a large context or little amount of memory Steps: Start to chat or trigger an auto-completion request in IDE Wait for the system to respond Expected result:...