NeMo
Safetensors
llama
mylesgoose commited on
Commit
4a06fcb
1 Parent(s): ee1b39a

Upload 3 files

Browse files
vision/config.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "mylesgoose/Llama-3.1-Minitron-4B-Width-Base",
3
+ "add_faster_video": false,
4
+ "add_time_instruction": false,
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "bos_token_id": 128000,
11
+ "eos_token_id": 128001,
12
+ "faster_token_stride": 10,
13
+ "force_sample": false,
14
+ "head_dim": 128,
15
+ "hidden_act": "silu",
16
+ "hidden_size": 3072,
17
+ "image_aspect_ratio": "square",
18
+ "image_crop_resolution": null,
19
+ "image_grid_pinpoints": null,
20
+ "image_split_resolution": null,
21
+ "initializer_range": 0.02,
22
+ "intermediate_size": 9216,
23
+ "max_position_embeddings": 131072,
24
+ "mlp_bias": false,
25
+ "mm_hidden_size": 1152,
26
+ "mm_newline_position": "grid",
27
+ "mm_patch_merge_type": "flat",
28
+ "mm_projector_lr": null,
29
+ "mm_projector_type": "mlp2x_gelu",
30
+ "mm_resampler_type": null,
31
+ "mm_spatial_pool_mode": "bilinear",
32
+ "mm_spatial_pool_stride": null,
33
+ "mm_tunable_parts": "mm_mlp_adapter",
34
+ "mm_use_im_patch_token": false,
35
+ "mm_use_im_start_end": false,
36
+ "mm_vision_select_feature": "patch",
37
+ "mm_vision_select_layer": -2,
38
+ "mm_vision_tower": "google/siglip-so400m-patch14-384",
39
+ "mm_vision_tower_lr": null,
40
+ "model_type": "llama",
41
+ "num_attention_heads": 32,
42
+ "num_hidden_layers": 32,
43
+ "num_key_value_heads": 8,
44
+ "pos_skipping_range": 4096,
45
+ "pretraining_tp": 1,
46
+ "rms_norm_eps": 1e-05,
47
+ "rope_scaling": {
48
+ "factor": 8.0,
49
+ "high_freq_factor": 4.0,
50
+ "low_freq_factor": 1.0,
51
+ "original_max_position_embeddings": 8192,
52
+ "rope_type": "llama3"
53
+ },
54
+ "rope_theta": 500000.0,
55
+ "tie_word_embeddings": false,
56
+ "tokenizer_model_max_length": 131072,
57
+ "tokenizer_padding_side": "right",
58
+ "torch_dtype": "bfloat16",
59
+ "transformers_version": "4.45.0.dev0",
60
+ "use_cache": true,
61
+ "use_mm_proj": true,
62
+ "use_pos_skipping": false,
63
+ "vision_tower_pretrained": null,
64
+ "vocab_size": 128256
65
+ }
vision/mm_projector.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d027af164131822a93b20d2fda3b49b60b364f51cf21ad548db757326bc70bc2
3
+ size 25966648
vision/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff