redbrain

This results in `NotImplementedError: Failed to find assignment for logical_axis_index 1 of size 64 with remaining assignable mesh [4, 4, 8].` Any clue what went wrong?

It appears that since a v4-256 has half the chips of a v4-512, the appropriate mesh topology would be `-1,32,1`. But running it with that mesh and with batch sizes...

Still not working, even with the parameters you suggested for mesh_dim and batch_size. Full command ```sh export LIBTPU_INIT_ARGS='--xla_jf_spmd_threshold_for_windowed_einsum_mib=0 --xla_tpu_spmd_threshold_for_allgather_cse=10000 --xla_tpu_spmd_rewrite_einsum_with_reshape=true --xla_enable_async_all_gather=true --jax_enable_async_collective_offload=true --xla_tpu_enable_latency_hiding_scheduler=true TPU_MEGACORE=MEGACORE_DENSE' python -m EasyLM.models.llama.llama_train \ --mesh_dim='!-1,64,1' \...

Same here: https://wzrd.in/standalone/color-convert

Same here, any updates on this?

I've found that using `upx --best --ultra-brute` on the pkg executable after creation can decrease its size dramatically.

mrocha, I'm having the same result with my current project. I'm now trying to figure out if pkg can be modified so that the two are compatible. Aside from that,...

https://x.com/elder_plinius/status/1952958577867669892/ Seems like there's a functioning jailbreak, it just hasn't been added to the repo yet?