Add support for GPTBigCode(starcoder) #63
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Should add support for GPTBigCode models which is used for bigcode/starcoder
BigDisclaimer: I only tested it with my textgen PR that is actually using the #43 branch also I don't have enough VRAM to load this model fully on gpu, so I used splitting. Note: The memory limits are not very accurately respected and I had to set GPU memory to 6GiB to actually fit it into 10 GB, this is separate issue though.
Eitherway I got the model loaded and it generated reasonable code, so I assume it works. Model used: mayank31398/starcoder-GPTQ-4bit-128g
I did not try quantizing the model myself with this.
Some testing needed by someone more competent and with better hardware.