Trying to split model with --split-max-size, but gguf-split ignores it

[RichardErkhov]

Latest version, ubuntu 2204, conda python=3.10.
Trying to split model with gguf-split, but something is going wrong

(base) richard@richard-ProLiant-DL580-Gen9:~/Desktop/ramdisk/banana/llama.cpp$ ./gguf-split --split --split-max-size 4000M --dry-run /media/richard/5fbd0bfa-8253-4803-85eb-80a13218a927/grok-1-fp16-gguf/grok-1-Q5_K.gguf Q5_K/grok-1 
n_split: 1
split 00001: n_tensors = 2115, total_size = 214437M
gguf_split: 1 gguf split written with a total of 2115 tensors.
(base) richard@richard-ProLiant-DL580-Gen9:~/Desktop/ramdisk/banana/llama.cpp$ ./gguf-split --split --split-max-size 4G --dry-run /media/richard/5fbd0bfa-8253-4803-85eb-80a13218a927/grok-1-fp16-gguf/grok-1-Q5_K.gguf Q5_K/grok-1 
n_split: 17
split 00001: n_tensors = 128, total_size = 14609M
split 00002: n_tensors = 128, total_size = 13184M
split 00003: n_tensors = 128, total_size = 12648M
split 00004: n_tensors = 128, total_size = 12597M
split 00005: n_tensors = 128, total_size = 12648M
split 00006: n_tensors = 128, total_size = 12750M
split 00007: n_tensors = 128, total_size = 12836M
split 00008: n_tensors = 128, total_size = 13088M
split 00009: n_tensors = 128, total_size = 13197M
split 00010: n_tensors = 128, total_size = 12597M
split 00011: n_tensors = 128, total_size = 12597M
split 00012: n_tensors = 128, total_size = 12699M
split 00013: n_tensors = 128, total_size = 12699M
split 00014: n_tensors = 128, total_size = 12597M
split 00015: n_tensors = 128, total_size = 13137M
split 00016: n_tensors = 128, total_size = 13675M
split 00017: n_tensors = 67, total_size = 6868M
gguf_split: 17 gguf split written with a total of 2115 tensors.

[phymbert]

See:

• How to use the `gguf-split` / Model sharding demo #6404 (reply in thread)
• split: allow --split-max-size option #6259

[RichardErkhov]

See:

• How to use the `gguf-split` / Model sharding demo #6404 (reply in thread)
• split: allow --split-max-size option #6259

So basically it doesnt work and I need to use split max tensors for now ?