Can get wan 2.1 to start generating

[KintCark]

I get all the way past the model loading and when I get ti the generation process I keep getting this error:

ggml_new_object: not enough space in the context's memory pool (needed 16810528, available 11370496)
/root/stable-diffusion.cpp/ggml/src/ggml.c:1663: GGML_ASSERT(obj_new) failed
Aborted

I can't get past this error

[Amin456789]

not a tech but i think it is because of the ram. i mean iit needs more ram, maybe y should use gguf of wan
go hugging face and search wan 1.3 gguf q3 k m and use umt5 q3 k s too
im still cant test wan due to cpu problem i got so it is all a wild guess

[KintCark]

not a tech but i think it is because of the ram. i mean iit needs more ram, maybe y should use gguf of wan
go hugging face and search wan 1.3 gguf q3 k m and use umt5 q3 k s too
im still cant test wan due to cpu problem i got so it is all a wild guess

That's the problem I'm using gguf versions I even quantize it to 2_K and the vae still not work

[Amin456789]

even wan 1.3b q2 eats that much of a ram? then my laptop wont load it at all hah
maybe choose 320x320 resouliton and test see

[KintCark]

even wan 1.3b q2 eats that much of a ram? then my laptop wont load it at all hah
maybe choose 320x320 resouliton and test see

I tried every small resolution possible even 32x32 still same error

[wbruna]

I get all the way past the model loading and when I get ti the generation process I keep getting this error:

ggml_new_object: not enough space in the context's memory pool (needed 16810528, available 11370496) /root/stable-diffusion.cpp/ggml/src/ggml.c:1663: GGML_ASSERT(obj_new) failed Aborted

As Amin456789 said, it really looks like a not-enough-VRAM issue. I don't remember the details, but people commented on the original PR about the memory requirements: #778

If you have plenty of RAM, you could try the flag --offload-to-cpu .

[leejet]

Could you paste the full command here to take a look?

[phil2sat]

try:
--offload-to-cpu --vae-tiling --vae-on-cpu --clip-on-cpu
hope you have enough system ram also.
if it works first remove vae, then tiling.
on a modern cpu clip runs not that bad, maybe compare times, for me on a 4790k from 2013 its damn slow

[repsac-by]

It was the same
It turned out that I had missed the -M vid_gen parameter