Hugging Face's logo

amy-hyunji-lee
/
version_llama3.original_8b.pretrain

llava_llama
Model card Files Community
amy-hyunji-lee commited on
Commit
ff10e8b
·
verified ·
1 Parent(s): 5fec6a0

Add files using upload-large-folder tool

Browse files
Files changed (5) hide show
  1. checkpoint-2180/config.json +49 -0
  2. checkpoint-2180/mm_projector.bin +3 -0
  3. config.json +49 -0
  4. mm_projector.bin +3 -0
  5. trainer_state.json +0 -0
checkpoint-2180/config.json ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_attn_implementation_autoset": true,
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 128000,
9
+ "eos_token_id": 128001,
10
+ "freeze_mm_mlp_adapter": false,
11
+ "head_dim": 128,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 4096,
14
+ "image_aspect_ratio": "square",
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 14336,
17
+ "max_position_embeddings": 131072,
18
+ "mlp_bias": false,
19
+ "mm_hidden_size": 1024,
20
+ "mm_patch_merge_type": "flat",
21
+ "mm_projector_lr": null,
22
+ "mm_projector_type": "mlp2x_gelu",
23
+ "mm_use_im_patch_token": false,
24
+ "mm_use_im_start_end": false,
25
+ "mm_vision_select_feature": "patch",
26
+ "mm_vision_select_layer": -2,
27
+ "mm_vision_tower": "openai/clip-vit-large-patch14-336",
28
+ "model_type": "llava_llama",
29
+ "num_attention_heads": 32,
30
+ "num_hidden_layers": 32,
31
+ "num_key_value_heads": 8,
32
+ "pretraining_tp": 1,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "rope_type": "dynamic",
37
+ "type": "dynamic"
38
+ },
39
+ "rope_theta": 500000.0,
40
+ "tie_word_embeddings": false,
41
+ "tokenizer_model_max_length": 2048,
42
+ "tokenizer_padding_side": "right",
43
+ "torch_dtype": "bfloat16",
44
+ "transformers_version": "4.51.3",
45
+ "tune_mm_mlp_adapter": true,
46
+ "use_cache": false,
47
+ "use_mm_proj": true,
48
+ "vocab_size": 128256
49
+ }
checkpoint-2180/mm_projector.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3190ec658dac244878877f9f09a7b211d2d8b302630f9168a6ac236f6ee041f
3
+ size 41961528
config.json ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_attn_implementation_autoset": true,
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 128000,
9
+ "eos_token_id": 128001,
10
+ "freeze_mm_mlp_adapter": false,
11
+ "head_dim": 128,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 4096,
14
+ "image_aspect_ratio": "square",
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 14336,
17
+ "max_position_embeddings": 131072,
18
+ "mlp_bias": false,
19
+ "mm_hidden_size": 1024,
20
+ "mm_patch_merge_type": "flat",
21
+ "mm_projector_lr": null,
22
+ "mm_projector_type": "mlp2x_gelu",
23
+ "mm_use_im_patch_token": false,
24
+ "mm_use_im_start_end": false,
25
+ "mm_vision_select_feature": "patch",
26
+ "mm_vision_select_layer": -2,
27
+ "mm_vision_tower": "openai/clip-vit-large-patch14-336",
28
+ "model_type": "llava_llama",
29
+ "num_attention_heads": 32,
30
+ "num_hidden_layers": 32,
31
+ "num_key_value_heads": 8,
32
+ "pretraining_tp": 1,
33
+ "rms_norm_eps": 1e-05,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "rope_type": "dynamic",
37
+ "type": "dynamic"
38
+ },
39
+ "rope_theta": 500000.0,
40
+ "tie_word_embeddings": false,
41
+ "tokenizer_model_max_length": 2048,
42
+ "tokenizer_padding_side": "right",
43
+ "torch_dtype": "bfloat16",
44
+ "transformers_version": "4.51.3",
45
+ "tune_mm_mlp_adapter": true,
46
+ "use_cache": true,
47
+ "use_mm_proj": true,
48
+ "vocab_size": 128256
49
+ }
mm_projector.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3190ec658dac244878877f9f09a7b211d2d8b302630f9168a6ac236f6ee041f
3
+ size 41961528
trainer_state.json ADDED
The diff for this file is too large to render. See raw diff