A bagel, with everything
Just a fiction oriented 4bpw exl2 quantization of https://huggingface.co/jondurbin/bagel-dpo-34b-v0.2
Quantized on 300K tokens of two Vicuna format chats, a sci fi story and a fiction story at a long context. This should yield better storywriting performance than the default exl2 quantization.
If anyone wants sizes other than 4bpw, for more/less context or smaller GPUs, just ask.
Running
Being a Yi model, try running a lower temperature with ~0.05 MinP, a little repitition penalty, maybe mirostat with a low tau, and no other samplers. Yi tends to run "hot" by default.
24GB GPUs can run Yi-34B-200K models at 45K-75K context with exllamav2, and performant UIs like exui. I go into more detail in this post
Commands
First pass:
python convert.py --in_dir /home/alpha/FastModels/jondurbin_bagel-dpo-34b-v0.2 -o /home/alpha/FastModels/scratch -om /home/alpha/FastModels/bagelmeas.json --cal_dataset /home/alpha/Documents/stories.parquet -ml 32768 -mr 7 -ss 4096 -b 4.0 -hb 6 -nr
Second pass:
python convert.py --in_dir /home/alpha/FastModels/jondurbin_bagel-dpo-34b-v0.2 -o /home/alpha/FastModels/scratch -m /home/alpha/FastModels/bagelmeas.json --cal_dataset /home/alpha/Documents/stories.parquet -l 12288 -r 25 -ml 32768 -mr 9 -ss 4096 -b 4.0 -hb 6 -cf /home/alpha/FastModels/jondurbin_bagel-dpo-34b-v0.2-exl2-4bpw-fiction -nr
- Downloads last month
- 14
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.