llava

🌋 LLaVA is a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding. Updated to version 1.6.

Vision 7B 13B 34B

1.4M Pulls Updated 8 months ago

llava:7b-v1.6-mistral-q4_K_M ... /

projector

72d6f08a42f6 · 624MB

Metadata

general.architecture

clip
general.file_type

F16
clip.has_llava_projector

true
clip.has_text_encoder

false
clip.has_vision_encoder

true
clip.projector_type

mlp
clip.use_gelu

false
clip.vision.attention.head_count

16
clip.vision.attention.layer_norm_epsilon

1e-05
clip.vision.block_count

23
clip.vision.embedding_length

1024
clip.vision.feed_forward_length

4096
clip.vision.image_mean

0.48145467
clip.vision.image_size

336
clip.vision.image_std

0.26862955
clip.vision.patch_size

14
clip.vision.projection_dim

768

Tensors

Name

Type

Shape
v.class_embd

F32

[1024]
v.patch_embd.weight

F16

[14 14 3 1024]
v.position_embd.weight

F16

[1024 577]

v.blk.0

v.blk.0.attn_k.weight

F16

[1024 1024]
v.blk.0.attn_k.bias

F32

[1024]
v.blk.0.attn_v.weight

F16

[1024 1024]
v.blk.0.attn_v.bias

F32

[1024]
v.blk.0.attn_q.weight

F16

[1024 1024]
v.blk.0.attn_q.bias

F32

[1024]
v.blk.0.attn_out.weight

F16

[1024 1024]
v.blk.0.attn_out.bias

F32

[1024]
v.blk.0.ln1.weight

F32

[1024]
v.blk.0.ln1.bias

F32

[1024]
v.blk.0.ffn_down.weight

F16

[1024 4096]
v.blk.0.ffn_down.bias

F32

[4096]
v.blk.0.ffn_up.weight

F16

[4096 1024]
v.blk.0.ffn_up.bias

F32

[1024]
v.blk.0.ln2.weight

F32

[1024]
v.blk.0.ln2.bias

F32

[1024]

v.blk.1

v.blk.1.attn_k.weight

F16

[1024 1024]
v.blk.1.attn_k.bias

F32

[1024]
v.blk.1.attn_v.weight

F16

[1024 1024]
v.blk.1.attn_v.bias

F32

[1024]
v.blk.1.attn_q.weight

F16

[1024 1024]
v.blk.1.attn_q.bias

F32

[1024]
v.blk.1.attn_out.weight

F16

[1024 1024]
v.blk.1.attn_out.bias

F32

[1024]
v.blk.1.ln1.weight

F32

[1024]
v.blk.1.ln1.bias

F32

[1024]
v.blk.1.ffn_down.weight

F16

[1024 4096]
v.blk.1.ffn_down.bias

F32

[4096]
v.blk.1.ffn_up.weight

F16

[4096 1024]
v.blk.1.ffn_up.bias

F32

[1024]
v.blk.1.ln2.weight

F32

[1024]
v.blk.1.ln2.bias

F32

[1024]

v.blk.2

v.blk.2.attn_k.weight

F16

[1024 1024]
v.blk.2.attn_k.bias

F32

[1024]
v.blk.2.attn_v.weight

F16

[1024 1024]
v.blk.2.attn_v.bias

F32

[1024]
v.blk.2.attn_q.weight

F16

[1024 1024]
v.blk.2.attn_q.bias

F32

[1024]
v.blk.2.attn_out.weight

F16

[1024 1024]
v.blk.2.attn_out.bias

F32

[1024]
v.blk.2.ln1.weight

F32

[1024]
v.blk.2.ln1.bias

F32

[1024]
v.blk.2.ffn_down.weight

F16

[1024 4096]
v.blk.2.ffn_down.bias

F32

[4096]
v.blk.2.ffn_up.weight

F16

[4096 1024]
v.blk.2.ffn_up.bias

F32

[1024]
v.blk.2.ln2.weight

F32

[1024]
v.blk.2.ln2.bias

F32

[1024]

v.blk.3

v.blk.3.attn_k.weight

F16

[1024 1024]
v.blk.3.attn_k.bias

F32

[1024]
v.blk.3.attn_v.weight

F16

[1024 1024]
v.blk.3.attn_v.bias

F32

[1024]
v.blk.3.attn_q.weight

F16

[1024 1024]
v.blk.3.attn_q.bias

F32

[1024]
v.blk.3.attn_out.weight

F16

[1024 1024]
v.blk.3.attn_out.bias

F32

[1024]
v.blk.3.ln1.weight

F32

[1024]
v.blk.3.ln1.bias

F32

[1024]
v.blk.3.ffn_down.weight

F16

[1024 4096]
v.blk.3.ffn_down.bias

F32

[4096]
v.blk.3.ffn_up.weight

F16

[4096 1024]
v.blk.3.ffn_up.bias

F32

[1024]
v.blk.3.ln2.weight

F32

[1024]
v.blk.3.ln2.bias

F32

[1024]

v.blk.4

v.blk.4.attn_k.weight

F16

[1024 1024]
v.blk.4.attn_k.bias

F32

[1024]
v.blk.4.attn_v.weight

F16

[1024 1024]
v.blk.4.attn_v.bias

F32

[1024]
v.blk.4.attn_q.weight

F16

[1024 1024]
v.blk.4.attn_q.bias

F32

[1024]
v.blk.4.attn_out.weight

F16

[1024 1024]
v.blk.4.attn_out.bias

F32

[1024]
v.blk.4.ln1.weight

F32

[1024]
v.blk.4.ln1.bias

F32

[1024]
v.blk.4.ffn_down.weight

F16

[1024 4096]
v.blk.4.ffn_down.bias

F32

[4096]
v.blk.4.ffn_up.weight

F16

[4096 1024]
v.blk.4.ffn_up.bias

F32

[1024]
v.blk.4.ln2.weight

F32

[1024]
v.blk.4.ln2.bias

F32

[1024]

v.blk.5

v.blk.5.attn_k.weight

F16

[1024 1024]
v.blk.5.attn_k.bias

F32

[1024]
v.blk.5.attn_v.weight

F16

[1024 1024]
v.blk.5.attn_v.bias

F32

[1024]
v.blk.5.attn_q.weight

F16

[1024 1024]
v.blk.5.attn_q.bias

F32

[1024]
v.blk.5.attn_out.weight

F16

[1024 1024]
v.blk.5.attn_out.bias

F32

[1024]
v.blk.5.ln1.weight

F32

[1024]
v.blk.5.ln1.bias

F32

[1024]
v.blk.5.ffn_down.weight

F16

[1024 4096]
v.blk.5.ffn_down.bias

F32

[4096]
v.blk.5.ffn_up.weight

F16

[4096 1024]
v.blk.5.ffn_up.bias

F32

[1024]
v.blk.5.ln2.weight

F32

[1024]
v.blk.5.ln2.bias

F32

[1024]

v.blk.6

v.blk.6.attn_k.weight

F16

[1024 1024]
v.blk.6.attn_k.bias

F32

[1024]
v.blk.6.attn_v.weight

F16

[1024 1024]
v.blk.6.attn_v.bias

F32

[1024]
v.blk.6.attn_q.weight

F16

[1024 1024]
v.blk.6.attn_q.bias

F32

[1024]
v.blk.6.attn_out.weight

F16

[1024 1024]
v.blk.6.attn_out.bias

F32

[1024]
v.blk.6.ln1.weight

F32

[1024]
v.blk.6.ln1.bias

F32

[1024]
v.blk.6.ffn_down.weight

F16

[1024 4096]
v.blk.6.ffn_down.bias

F32

[4096]
v.blk.6.ffn_up.weight

F16

[4096 1024]
v.blk.6.ffn_up.bias

F32

[1024]
v.blk.6.ln2.weight

F32

[1024]
v.blk.6.ln2.bias

F32

[1024]

v.blk.7

v.blk.7.attn_k.weight

F16

[1024 1024]
v.blk.7.attn_k.bias

F32

[1024]
v.blk.7.attn_v.weight

F16

[1024 1024]
v.blk.7.attn_v.bias

F32

[1024]
v.blk.7.attn_q.weight

F16

[1024 1024]
v.blk.7.attn_q.bias

F32

[1024]
v.blk.7.attn_out.weight

F16

[1024 1024]
v.blk.7.attn_out.bias

F32

[1024]
v.blk.7.ln1.weight

F32

[1024]
v.blk.7.ln1.bias

F32

[1024]
v.blk.7.ffn_down.weight

F16

[1024 4096]
v.blk.7.ffn_down.bias

F32

[4096]
v.blk.7.ffn_up.weight

F16

[4096 1024]
v.blk.7.ffn_up.bias

F32

[1024]
v.blk.7.ln2.weight

F32

[1024]
v.blk.7.ln2.bias

F32

[1024]

v.blk.8

v.blk.8.attn_k.weight

F16

[1024 1024]
v.blk.8.attn_k.bias

F32

[1024]
v.blk.8.attn_v.weight

F16

[1024 1024]
v.blk.8.attn_v.bias

F32

[1024]
v.blk.8.attn_q.weight

F16

[1024 1024]
v.blk.8.attn_q.bias

F32

[1024]
v.blk.8.attn_out.weight

F16

[1024 1024]
v.blk.8.attn_out.bias

F32

[1024]
v.blk.8.ln1.weight

F32

[1024]
v.blk.8.ln1.bias

F32

[1024]
v.blk.8.ffn_down.weight

F16

[1024 4096]
v.blk.8.ffn_down.bias

F32

[4096]
v.blk.8.ffn_up.weight

F16

[4096 1024]
v.blk.8.ffn_up.bias

F32

[1024]
v.blk.8.ln2.weight

F32

[1024]
v.blk.8.ln2.bias

F32

[1024]

v.blk.9

v.blk.9.attn_k.weight

F16

[1024 1024]
v.blk.9.attn_k.bias

F32

[1024]
v.blk.9.attn_v.weight

F16

[1024 1024]
v.blk.9.attn_v.bias

F32

[1024]
v.blk.9.attn_q.weight

F16

[1024 1024]
v.blk.9.attn_q.bias

F32

[1024]
v.blk.9.attn_out.weight

F16

[1024 1024]
v.blk.9.attn_out.bias

F32

[1024]
v.blk.9.ln1.weight

F32

[1024]
v.blk.9.ln1.bias

F32

[1024]
v.blk.9.ffn_down.weight

F16

[1024 4096]
v.blk.9.ffn_down.bias

F32

[4096]
v.blk.9.ffn_up.weight

F16

[4096 1024]
v.blk.9.ffn_up.bias

F32

[1024]
v.blk.9.ln2.weight

F32

[1024]
v.blk.9.ln2.bias

F32

[1024]

v.blk.10

v.blk.10.attn_k.weight

F16

[1024 1024]
v.blk.10.attn_k.bias

F32

[1024]
v.blk.10.attn_v.weight

F16

[1024 1024]
v.blk.10.attn_v.bias

F32

[1024]
v.blk.10.attn_q.weight

F16

[1024 1024]
v.blk.10.attn_q.bias

F32

[1024]
v.blk.10.attn_out.weight

F16

[1024 1024]
v.blk.10.attn_out.bias

F32

[1024]
v.blk.10.ln1.weight

F32

[1024]
v.blk.10.ln1.bias

F32

[1024]
v.blk.10.ffn_down.weight

F16

[1024 4096]
v.blk.10.ffn_down.bias

F32

[4096]
v.blk.10.ffn_up.weight

F16

[4096 1024]
v.blk.10.ffn_up.bias

F32

[1024]
v.blk.10.ln2.weight

F32

[1024]
v.blk.10.ln2.bias

F32

[1024]

v.blk.11

v.blk.11.attn_k.weight

F16

[1024 1024]
v.blk.11.attn_k.bias

F32

[1024]
v.blk.11.attn_v.weight

F16

[1024 1024]
v.blk.11.attn_v.bias

F32

[1024]
v.blk.11.attn_q.weight

F16

[1024 1024]
v.blk.11.attn_q.bias

F32

[1024]
v.blk.11.attn_out.weight

F16

[1024 1024]
v.blk.11.attn_out.bias

F32

[1024]
v.blk.11.ln1.weight

F32

[1024]
v.blk.11.ln1.bias

F32

[1024]
v.blk.11.ffn_down.weight

F16

[1024 4096]
v.blk.11.ffn_down.bias

F32

[4096]
v.blk.11.ffn_up.weight

F16

[4096 1024]
v.blk.11.ffn_up.bias

F32

[1024]
v.blk.11.ln2.weight

F32

[1024]
v.blk.11.ln2.bias

F32

[1024]

v.blk.12

v.blk.12.attn_k.weight

F16

[1024 1024]
v.blk.12.attn_k.bias

F32

[1024]
v.blk.12.attn_v.weight

F16

[1024 1024]
v.blk.12.attn_v.bias

F32

[1024]
v.blk.12.attn_q.weight

F16

[1024 1024]
v.blk.12.attn_q.bias

F32

[1024]
v.blk.12.attn_out.weight

F16

[1024 1024]
v.blk.12.attn_out.bias

F32

[1024]
v.blk.12.ln1.weight

F32

[1024]
v.blk.12.ln1.bias

F32

[1024]
v.blk.12.ffn_down.weight

F16

[1024 4096]
v.blk.12.ffn_down.bias

F32

[4096]
v.blk.12.ffn_up.weight

F16

[4096 1024]
v.blk.12.ffn_up.bias

F32

[1024]
v.blk.12.ln2.weight

F32

[1024]
v.blk.12.ln2.bias

F32

[1024]

v.blk.13

v.blk.13.attn_k.weight

F16

[1024 1024]
v.blk.13.attn_k.bias

F32

[1024]
v.blk.13.attn_v.weight

F16

[1024 1024]
v.blk.13.attn_v.bias

F32

[1024]
v.blk.13.attn_q.weight

F16

[1024 1024]
v.blk.13.attn_q.bias

F32

[1024]
v.blk.13.attn_out.weight

F16

[1024 1024]
v.blk.13.attn_out.bias

F32

[1024]
v.blk.13.ln1.weight

F32

[1024]
v.blk.13.ln1.bias

F32

[1024]
v.blk.13.ffn_down.weight

F16

[1024 4096]
v.blk.13.ffn_down.bias

F32

[4096]
v.blk.13.ffn_up.weight

F16

[4096 1024]
v.blk.13.ffn_up.bias

F32

[1024]
v.blk.13.ln2.weight

F32

[1024]
v.blk.13.ln2.bias

F32

[1024]

v.blk.14

v.blk.14.attn_k.weight

F16

[1024 1024]
v.blk.14.attn_k.bias

F32

[1024]
v.blk.14.attn_v.weight

F16

[1024 1024]
v.blk.14.attn_v.bias

F32

[1024]
v.blk.14.attn_q.weight

F16

[1024 1024]
v.blk.14.attn_q.bias

F32

[1024]
v.blk.14.attn_out.weight

F16

[1024 1024]
v.blk.14.attn_out.bias

F32

[1024]
v.blk.14.ln1.weight

F32

[1024]
v.blk.14.ln1.bias

F32

[1024]
v.blk.14.ffn_down.weight

F16

[1024 4096]
v.blk.14.ffn_down.bias

F32

[4096]
v.blk.14.ffn_up.weight

F16

[4096 1024]
v.blk.14.ffn_up.bias

F32

[1024]
v.blk.14.ln2.weight

F32

[1024]
v.blk.14.ln2.bias

F32

[1024]

v.blk.15

v.blk.15.attn_k.weight

F16

[1024 1024]
v.blk.15.attn_k.bias

F32

[1024]
v.blk.15.attn_v.weight

F16

[1024 1024]
v.blk.15.attn_v.bias

F32

[1024]
v.blk.15.attn_q.weight

F16

[1024 1024]
v.blk.15.attn_q.bias

F32

[1024]
v.blk.15.attn_out.weight

F16

[1024 1024]
v.blk.15.attn_out.bias

F32

[1024]
v.blk.15.ln1.weight

F32

[1024]
v.blk.15.ln1.bias

F32

[1024]
v.blk.15.ffn_down.weight

F16

[1024 4096]
v.blk.15.ffn_down.bias

F32

[4096]
v.blk.15.ffn_up.weight

F16

[4096 1024]
v.blk.15.ffn_up.bias

F32

[1024]
v.blk.15.ln2.weight

F32

[1024]
v.blk.15.ln2.bias

F32

[1024]

v.blk.16

v.blk.16.attn_k.weight

F16

[1024 1024]
v.blk.16.attn_k.bias

F32

[1024]
v.blk.16.attn_v.weight

F16

[1024 1024]
v.blk.16.attn_v.bias

F32

[1024]
v.blk.16.attn_q.weight

F16

[1024 1024]
v.blk.16.attn_q.bias

F32

[1024]
v.blk.16.attn_out.weight

F16

[1024 1024]
v.blk.16.attn_out.bias

F32

[1024]
v.blk.16.ln1.weight

F32

[1024]
v.blk.16.ln1.bias

F32

[1024]
v.blk.16.ffn_down.weight

F16

[1024 4096]
v.blk.16.ffn_down.bias

F32

[4096]
v.blk.16.ffn_up.weight

F16

[4096 1024]
v.blk.16.ffn_up.bias

F32

[1024]
v.blk.16.ln2.weight

F32

[1024]
v.blk.16.ln2.bias

F32

[1024]

v.blk.17

v.blk.17.attn_k.weight

F16

[1024 1024]
v.blk.17.attn_k.bias

F32

[1024]
v.blk.17.attn_v.weight

F16

[1024 1024]
v.blk.17.attn_v.bias

F32

[1024]
v.blk.17.attn_q.weight

F16

[1024 1024]
v.blk.17.attn_q.bias

F32

[1024]
v.blk.17.attn_out.weight

F16

[1024 1024]
v.blk.17.attn_out.bias

F32

[1024]
v.blk.17.ln1.weight

F32

[1024]
v.blk.17.ln1.bias

F32

[1024]
v.blk.17.ffn_down.weight

F16

[1024 4096]
v.blk.17.ffn_down.bias

F32

[4096]
v.blk.17.ffn_up.weight

F16

[4096 1024]
v.blk.17.ffn_up.bias

F32

[1024]
v.blk.17.ln2.weight

F32

[1024]
v.blk.17.ln2.bias

F32

[1024]

v.blk.18

v.blk.18.attn_k.weight

F16

[1024 1024]
v.blk.18.attn_k.bias

F32

[1024]
v.blk.18.attn_v.weight

F16

[1024 1024]
v.blk.18.attn_v.bias

F32

[1024]
v.blk.18.attn_q.weight

F16

[1024 1024]
v.blk.18.attn_q.bias

F32

[1024]
v.blk.18.attn_out.weight

F16

[1024 1024]
v.blk.18.attn_out.bias

F32

[1024]
v.blk.18.ln1.weight

F32

[1024]
v.blk.18.ln1.bias

F32

[1024]
v.blk.18.ffn_down.weight

F16

[1024 4096]
v.blk.18.ffn_down.bias

F32

[4096]
v.blk.18.ffn_up.weight

F16

[4096 1024]
v.blk.18.ffn_up.bias

F32

[1024]
v.blk.18.ln2.weight

F32

[1024]
v.blk.18.ln2.bias

F32

[1024]

v.blk.19

v.blk.19.attn_k.weight

F16

[1024 1024]
v.blk.19.attn_k.bias

F32

[1024]
v.blk.19.attn_v.weight

F16

[1024 1024]
v.blk.19.attn_v.bias

F32

[1024]
v.blk.19.attn_q.weight

F16

[1024 1024]
v.blk.19.attn_q.bias

F32

[1024]
v.blk.19.attn_out.weight

F16

[1024 1024]
v.blk.19.attn_out.bias

F32

[1024]
v.blk.19.ln1.weight

F32

[1024]
v.blk.19.ln1.bias

F32

[1024]
v.blk.19.ffn_down.weight

F16

[1024 4096]
v.blk.19.ffn_down.bias

F32

[4096]
v.blk.19.ffn_up.weight

F16

[4096 1024]
v.blk.19.ffn_up.bias

F32

[1024]
v.blk.19.ln2.weight

F32

[1024]
v.blk.19.ln2.bias

F32

[1024]

v.blk.20

v.blk.20.attn_k.weight

F16

[1024 1024]
v.blk.20.attn_k.bias

F32

[1024]
v.blk.20.attn_v.weight

F16

[1024 1024]
v.blk.20.attn_v.bias

F32

[1024]
v.blk.20.attn_q.weight

F16

[1024 1024]
v.blk.20.attn_q.bias

F32

[1024]
v.blk.20.attn_out.weight

F16

[1024 1024]
v.blk.20.attn_out.bias

F32

[1024]
v.blk.20.ln1.weight

F32

[1024]
v.blk.20.ln1.bias

F32

[1024]
v.blk.20.ffn_down.weight

F16

[1024 4096]
v.blk.20.ffn_down.bias

F32

[4096]
v.blk.20.ffn_up.weight

F16

[4096 1024]
v.blk.20.ffn_up.bias

F32

[1024]
v.blk.20.ln2.weight

F32

[1024]
v.blk.20.ln2.bias

F32

[1024]

v.blk.21

v.blk.21.attn_k.weight

F16

[1024 1024]
v.blk.21.attn_k.bias

F32

[1024]
v.blk.21.attn_v.weight

F16

[1024 1024]
v.blk.21.attn_v.bias

F32

[1024]
v.blk.21.attn_q.weight

F16

[1024 1024]
v.blk.21.attn_q.bias

F32

[1024]
v.blk.21.attn_out.weight

F16

[1024 1024]
v.blk.21.attn_out.bias

F32

[1024]
v.blk.21.ln1.weight

F32

[1024]
v.blk.21.ln1.bias

F32

[1024]
v.blk.21.ffn_down.weight

F16

[1024 4096]
v.blk.21.ffn_down.bias

F32

[4096]
v.blk.21.ffn_up.weight

F16

[4096 1024]
v.blk.21.ffn_up.bias

F32

[1024]
v.blk.21.ln2.weight

F32

[1024]
v.blk.21.ln2.bias

F32

[1024]

v.blk.22

v.blk.22.attn_k.weight

F16

[1024 1024]
v.blk.22.attn_k.bias

F32

[1024]
v.blk.22.attn_v.weight

F16

[1024 1024]
v.blk.22.attn_v.bias

F32

[1024]
v.blk.22.attn_q.weight

F16

[1024 1024]
v.blk.22.attn_q.bias

F32

[1024]
v.blk.22.attn_out.weight

F16

[1024 1024]
v.blk.22.attn_out.bias

F32

[1024]
v.blk.22.ln1.weight

F32

[1024]
v.blk.22.ln1.bias

F32

[1024]
v.blk.22.ffn_down.weight

F16

[1024 4096]
v.blk.22.ffn_down.bias

F32

[4096]
v.blk.22.ffn_up.weight

F16

[4096 1024]
v.blk.22.ffn_up.bias

F32

[1024]
v.blk.22.ln2.weight

F32

[1024]
v.blk.22.ln2.bias

F32

[1024]

mm.0.bias

F32

[4096]
mm.0.weight

F16

[1024 4096]
mm.2.bias

F32

[4096]
mm.2.weight

F16

[4096 4096]
v.pre_ln.weight

F32

[1024]
v.pre_ln.bias

F32

[1024]