Is this the updated version or the bugged version?

#2
by PrimeD - opened
PrimeD changed discussion title from Is this the updated version of the bugged version? to Is this the updated version or the bugged version?

He hasn't even fixed it in the original model yet as the last commit is dated 6th September 2024 14:16:52 GMT while his tweat is from 7th september 2024 14:23 GMT. So how can you expect the GGUF quants to contain the yet non-existant fixed version?

He hasn't even fixed it in the original model yet as the last commit is dated 6th September 2024 14:16:52 GMT while his tweat is from 7th september 2024 14:23 GMT. So how can you expect the GGUF quants to contain the yet non-existant fixed version?

is this the same model- https://huggingface.co/mattshumer/Reflection-Llama-3.1-70B ? I am bit confused..

"Reflection Llama-3.1 70B
| IMPORTANT UPDATE – There was an issue with the model when we first uploaded it. If you tried it and didn't have good results, please, try again, we think we've fixed the issue.

Reflection Llama-3.1 70B is (currently) the world's top open-source LLM, trained with a new technique called Reflection-Tuning that teaches a LLM to detect mistakes in its reasoning and correct course.

The model was trained on synthetic data generated by Glaive. If you're training a model, Glaive is incredible β€” use them."

This quant is based on https://huggingface.co/leafspark/Reflection-Llama-3.1-70B-bf16 and not https://huggingface.co/mattshumer/Reflection-Llama-3.1-70B as stated in the model card but seems to be the exact same model uploaded by different persons. Not sure if leafspark's version is affected by the issue that affected mattshumer's upload. @mradermacher I recommend you quant mattshumer's fixed version to ensure we have the fixed version.

I see one difference. On leafspark's version there is the following comment added to the model card:

Notes
This model has been updated with the latest embedding fixes.

It seems like leafspark's forked and fixed mattshumer's version before he fixed it himself in which case this would be the fixed version. Maybe still worth to quant the officially fixed one just to clear any confusion.

The original version was also quanted: https://huggingface.co/mradermacher/Reflection-70B-i1-GGUF https://huggingface.co/mradermacher/Reflection-70B-GGUF

If somebody tells me when the original repo has an updated version I'll happily requant it.

The original model from mattshumer was fixed at 6th September 2024 09:50:38 GMT so you likely already have the fixed version of the original under https://huggingface.co/mradermacher/Reflection-70B-i1-GGUF and https://huggingface.co/mradermacher/Reflection-70B-GGUF but please check inside the logs to make sure.

It was queued 9 hours before that, but the second download was ten hours after that. If I remember correctly, it wouldn't convert originally anyway, so if it was already fixed that long ago, it should be the new version.

mradermacher changed discussion status to closed

Sign up or log in to comment