Skip to content

tests/zamba2: set n_mamba_heads in the fixture, so the test means the same thing on transformers 4 and 5 (engine #1036) #232

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
bong-water-water-bong wants to merge 1 commit into
base: 1bit/hrx-vulkan-patched
Choose a base branch
from
Open

tests/zamba2: set n_mamba_heads in the fixture, so the test means the same thing on transformers 4 and 5 (engine #1036) #232

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions tests/test-zamba2-convert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,6 +91,8 @@ def make_checkpoint(out: Path, n_blocks: int, attn_adapter: bool, hybrid: list[i
layers_block_type=["hybrid" if i in hybrid else "mamba" for i in range(n_layer)],
num_mem_blocks=n_blocks, use_shared_attention_adapter=attn_adapter, use_shared_mlp_adapter=True,
adapter_rank=RANK, mamba_d_state=16, mamba_d_conv=4, mamba_expand=2, mamba_headdim=16, mamba_ngroups=1,
# set both: transformers 4.x derives headdim from n_mamba_heads (default 8), 5.x keeps headdim and n_mamba_heads unrelated
n_mamba_heads=2 * HIDDEN // 16,
use_mem_rope=True, max_position_embeddings=64, tie_word_embeddings=True,
)
torch.manual_seed(seed)
Expand Down