float16 weights

#1
by jartine - opened

Thanks so much for converting this model to GGUF. Could you post the f16 weights (llava-v1.5-13b-f16.gguf) for the model itself please? I love using MP3s but I like owning the CD too, you know? Thanks!

yes of course sorry.

Got lost in the world of MOE madness, shall up as some as pipe is clear.

yeah. I made one but it's being weird. Investigating cos I used a new pipeline. I might just do it by hand again if I cant figure out why :\

llama_build_graph: non-view tensors processed: 844/844
llama_new_context_with_model: compute buffer total size = 197.07 MiB

encode_image_with_clip: image encoded in 2656.21 ms by CLIP ( 4.61 ms per image patch)

The image showcases a computer screen displaying a large thumbnail image. The thumbnail is surrounded by text, which seems to be a discussion or a commentary on the image. The words appear to be in a foreign language, indicating that the image and its context might be unfamiliar or unrelated to the viewer. The layout of the image and the presence of text may suggest a digital art piece or an online discussion involving the thumbnail.

llama_print_timings: load time = 13087.37 ms
llama_print_timings: sample time = 2.71 ms / 92 runs ( 0.03 ms per token, 33898.31 tokens per second)
llama_print_timings: prompt eval time = 101298.42 ms / 617 tokens ( 164.18 ms per token, 6.09 tokens per second)
llama_print_timings: eval time = 99984.21 ms / 92 runs ( 1086.78 ms per token, 0.92 tokens per second)
llama_print_timings: total time = 206940.64 ms

seems ok now. Take a while to upload

yeah. happening. Be.... A while
image.png

image.png
Did this twice so far... give me strength. Might have to switch to my python game upload cos the web client explodifies . After the best part of a day :\

api.upload_folder(folder_path="D:\\models\\PsiPi\\llava\\f16\\", repo_id="PsiPi/liuhaotian_llava-v1.5-13b-GGUF", re po_type="model", multi_commits=True, multi_commits_verbose=True)

C:\Users\new\122\Lib\site-packages\huggingface_hub-0.19.4-py3.11.egg\huggingface_hub\utils_experimental.py:57: UserWarning: 'plan_multi_commits' is experimental and might be subject to breaking changes in the future. You can disable this warning by setting HF_HUB_DISABLE_EXPERIMENTAL_WARNING=1 as environment variable.
warnings.warn(
C:\Users\new\122\Lib\site-packages\huggingface_hub-0.19.4-py3.11.egg\huggingface_hub\utils_experimental.py:57: UserWarning: 'HfApi.create_commits_on_pr' is experimental and might be subject to breaking changes in the future. You can disable this warning by setting HF_HUB_DISABLE_EXPERIMENTAL_WARNING=1 as environment variable.
warnings.warn(
Will create 0 deletion commit(s) and 1 addition commit(s), totalling 1 atomic operations. Multi-commits strategy with ID e1ac506dd3d2faa596e37aef0db415326348cb3d5fcb5f40e94a5b59da9fb926. New PR created: https://huggingface.co/PsiPi/liuhaotian_llava-v1.5-13b-GGUF/discussions/2

llava-v1.5-13b-f16.gguf: 1%|▋ | 377M/26.0G [11:39<13:22:35, 533kB/s]

Seeing if this is more reliable.
give me strength

To future self. next time just use python to start with, be an hour
image.png

it's done. Sorry for the delays @jartine

Thanks again!

jartine changed discussion status to closed

Sign up or log in to comment