JunHill commited on
Commit
7cebc13
·
verified ·
1 Parent(s): 24c2ee1

Upload 2 files

Browse files
Files changed (2) hide show
  1. config.json +82 -0
  2. trimkv_weights.pth +3 -0
config.json ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "TrimKVQwen3VLForConditionalGeneration"
4
+ ],
5
+ "base_loss": "fwkl",
6
+ "base_model": "Qwen/Qwen3-VL-8B-Thinking",
7
+ "dtype": "bfloat16",
8
+ "eos_token_id": 151645,
9
+ "image_token_id": 151655,
10
+ "logit_block_size": 8192,
11
+ "memory_size": 256.0,
12
+ "model_type": "qwen3_vl",
13
+ "pad_token_id": 151643,
14
+ "retention_weight": 1.0,
15
+ "text_config": {
16
+ "attention_bias": false,
17
+ "attention_dropout": 0.0,
18
+ "attn_impl": "trimkv_flex",
19
+ "bos_token_id": 151643,
20
+ "buffer_size": 128,
21
+ "compress_memory": true,
22
+ "compress_strategy": "alpha",
23
+ "dtype": "float32",
24
+ "eos_token_id": 151645,
25
+ "head_dim": 128,
26
+ "hidden_act": "silu",
27
+ "hidden_size": 4096,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 12288,
30
+ "max_position_embeddings": 262144,
31
+ "max_seq_len": 32768,
32
+ "model_type": "qwen3_vl_text",
33
+ "num_attention_heads": 32,
34
+ "num_hidden_layers": 36,
35
+ "num_key_value_heads": 8,
36
+ "retention_gate": "rg",
37
+ "retention_gate_bias_init": 8.0,
38
+ "retention_gate_intermediate_size": 512,
39
+ "rg_dropout": 0.0,
40
+ "rms_norm_eps": 1e-06,
41
+ "rope_scaling": {
42
+ "mrope_interleaved": true,
43
+ "mrope_section": [
44
+ 24,
45
+ 20,
46
+ 20
47
+ ],
48
+ "rope_type": "default"
49
+ },
50
+ "rope_theta": 5000000,
51
+ "use_cache": false,
52
+ "vocab_size": 151936
53
+ },
54
+ "tie_word_embeddings": false,
55
+ "trainable_params": "self_attn.retention_gate",
56
+ "transformers_version": "4.57.0",
57
+ "use_cache": true,
58
+ "video_token_id": 151656,
59
+ "vision_config": {
60
+ "deepstack_visual_indexes": [
61
+ 8,
62
+ 16,
63
+ 24
64
+ ],
65
+ "depth": 27,
66
+ "dtype": "float32",
67
+ "hidden_act": "gelu_pytorch_tanh",
68
+ "hidden_size": 1152,
69
+ "in_channels": 3,
70
+ "initializer_range": 0.02,
71
+ "intermediate_size": 4304,
72
+ "model_type": "qwen3_vl",
73
+ "num_heads": 16,
74
+ "num_position_embeddings": 2304,
75
+ "out_hidden_size": 4096,
76
+ "patch_size": 16,
77
+ "spatial_merge_size": 2,
78
+ "temporal_patch_size": 2
79
+ },
80
+ "vision_end_token_id": 151653,
81
+ "vision_start_token_id": 151652
82
+ }
trimkv_weights.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e303f9b534e5564a4fd789ac31250bc2d621260ae7102fbbfb4a6e8fc90a5818
3
+ size 151351261