56 lines
3.9 KiB
CSV
56 lines
3.9 KiB
CSV
role,shape,scale,total_blocks,sampled_blocks
|
|
model.language_model.embed_tokens.weight,"(128, 128)",2d-128,31040,256
|
|
model.language_model.embed_tokens.weight,"(64, 64)",2d-64,124160,256
|
|
model.language_model.layers.*.linear_attn.conv1d.weight,"(1, 1, 4)",3d-axis0-slice,110592,256
|
|
model.language_model.layers.*.linear_attn.conv1d.weight,"(4, 1, 4)",3d-volume,27648,256
|
|
model.language_model.layers.*.linear_attn.in_proj_qkv.weight,"(128, 128)",2d-128,13824,256
|
|
model.language_model.layers.*.linear_attn.in_proj_qkv.weight,"(64, 64)",2d-64,55296,256
|
|
model.language_model.layers.*.linear_attn.in_proj_z.weight,"(128, 128)",2d-128,4608,256
|
|
model.language_model.layers.*.linear_attn.in_proj_z.weight,"(64, 64)",2d-64,18432,256
|
|
model.language_model.layers.*.linear_attn.out_proj.weight,"(128, 128)",2d-128,4608,256
|
|
model.language_model.layers.*.linear_attn.out_proj.weight,"(64, 64)",2d-64,18432,256
|
|
model.language_model.layers.*.mlp.down_proj.weight,"(128, 128)",2d-128,18432,256
|
|
model.language_model.layers.*.mlp.down_proj.weight,"(64, 64)",2d-64,73728,256
|
|
model.language_model.layers.*.mlp.gate_proj.weight,"(128, 128)",2d-128,18432,256
|
|
model.language_model.layers.*.mlp.gate_proj.weight,"(64, 64)",2d-64,73728,256
|
|
model.language_model.layers.*.mlp.up_proj.weight,"(128, 128)",2d-128,18432,256
|
|
model.language_model.layers.*.mlp.up_proj.weight,"(64, 64)",2d-64,73728,256
|
|
model.language_model.layers.*.self_attn.k_proj.weight,"(128, 128)",2d-128,384,256
|
|
model.language_model.layers.*.self_attn.k_proj.weight,"(64, 64)",2d-64,1536,256
|
|
model.language_model.layers.*.self_attn.o_proj.weight,"(128, 128)",2d-128,1536,256
|
|
model.language_model.layers.*.self_attn.o_proj.weight,"(64, 64)",2d-64,6144,256
|
|
model.language_model.layers.*.self_attn.q_proj.weight,"(128, 128)",2d-128,3072,256
|
|
model.language_model.layers.*.self_attn.q_proj.weight,"(64, 64)",2d-64,12288,256
|
|
model.language_model.layers.*.self_attn.v_proj.weight,"(128, 128)",2d-128,384,256
|
|
model.language_model.layers.*.self_attn.v_proj.weight,"(64, 64)",2d-64,1536,256
|
|
model.visual.layers.*.attn.proj.weight,"(128, 128)",2d-128,1536,256
|
|
model.visual.layers.*.attn.proj.weight,"(64, 64)",2d-64,6144,256
|
|
model.visual.layers.*.attn.qkv.weight,"(128, 128)",2d-128,4608,256
|
|
model.visual.layers.*.attn.qkv.weight,"(64, 64)",2d-64,18432,256
|
|
model.visual.layers.*.mlp.linear_fc1.weight,"(128, 128)",2d-128,6144,256
|
|
model.visual.layers.*.mlp.linear_fc1.weight,"(64, 64)",2d-64,24576,256
|
|
model.visual.layers.*.mlp.linear_fc2.weight,"(128, 128)",2d-128,6144,256
|
|
model.visual.layers.*.mlp.linear_fc2.weight,"(64, 64)",2d-64,24576,256
|
|
model.visual.merger.linear_fc1.weight,"(128, 128)",2d-128,1024,256
|
|
model.visual.merger.linear_fc1.weight,"(64, 64)",2d-64,4096,256
|
|
model.visual.merger.linear_fc2.weight,"(128, 128)",2d-128,512,256
|
|
model.visual.merger.linear_fc2.weight,"(64, 64)",2d-64,2048,256
|
|
model.visual.pos_embed.weight,"(128, 128)",2d-128,144,144
|
|
model.visual.pos_embed.weight,"(64, 64)",2d-64,576,256
|
|
mtp.fc.weight,"(128, 128)",2d-128,512,256
|
|
mtp.fc.weight,"(64, 64)",2d-64,2048,256
|
|
mtp.layers.*.mlp.down_proj.weight,"(128, 128)",2d-128,768,256
|
|
mtp.layers.*.mlp.down_proj.weight,"(64, 64)",2d-64,3072,256
|
|
mtp.layers.*.mlp.gate_proj.weight,"(128, 128)",2d-128,768,256
|
|
mtp.layers.*.mlp.gate_proj.weight,"(64, 64)",2d-64,3072,256
|
|
mtp.layers.*.mlp.up_proj.weight,"(128, 128)",2d-128,768,256
|
|
mtp.layers.*.mlp.up_proj.weight,"(64, 64)",2d-64,3072,256
|
|
mtp.layers.*.self_attn.k_proj.weight,"(128, 128)",2d-128,64,64
|
|
mtp.layers.*.self_attn.k_proj.weight,"(64, 64)",2d-64,256,256
|
|
mtp.layers.*.self_attn.o_proj.weight,"(128, 128)",2d-128,256,256
|
|
mtp.layers.*.self_attn.o_proj.weight,"(64, 64)",2d-64,1024,256
|
|
mtp.layers.*.self_attn.q_proj.weight,"(128, 128)",2d-128,512,256
|
|
mtp.layers.*.self_attn.q_proj.weight,"(64, 64)",2d-64,2048,256
|
|
mtp.layers.*.self_attn.v_proj.weight,"(128, 128)",2d-128,64,64
|
|
mtp.layers.*.self_attn.v_proj.weight,"(64, 64)",2d-64,256,256
|