Skip to content

gguf_to_q4nx: name the gated linear-attention gate (#2013) #2015

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

Closed
bong-water-water-bong wants to merge 1 commit into from
Closed

gguf_to_q4nx: name the gated linear-attention gate (#2013) #2015

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
7 changes: 7 additions & 0 deletions tools/gguf_to_q4nx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,6 +134,13 @@ def to_q4_1(t):
LAYER = {
"attn_norm": "input_layernorm", "ffn_norm": "post_attention_layernorm",
"attn_q_norm": "self_attn.q_norm", "attn_k_norm": "self_attn.k_norm",
# engine#2013: the GATED linear-attention (GateDeltaNet) families carry a per-layer attention gate under
# `blk.N.attn_gate.weight`; the container's name for it is `self_attn.gate_proj.weight`, which is what the forward
# looks up (npu/forward/src/model.c:1083, in the linear-attn section; q4nx_forward.h's `attn_output_gate`). Measured on
# this box's `bonsai2-unrot-q4_0.gguf`: every `blk.N.attn_gate.weight` is [5120, 6144] Q4_0, and 5120 x 6144 =
# 31,457,280 = 3840 x 8192, a whole number of 32 x 256 tiles, so the tiling check passes and the name is the only gap.
# Only some layers carry it (0, 1, 2, 4, 5, 6, 8, 9 in the first ten), so it is looked up per layer, never assumed.
"attn_gate": "self_attn.gate_proj",
"attn_q": "self_attn.q_proj", "attn_k": "self_attn.k_proj", "attn_v": "self_attn.v_proj",
"attn_output": "self_attn.o_proj",
"ffn_gate": "mlp.gate_proj", "ffn_up": "mlp.up_proj", "ffn_down": "mlp.down_proj",
Expand Down
Loading