ngocbh commited on
Commit
aae4dcf
·
1 Parent(s): 5aa0abe
Files changed (2) hide show
  1. config.json +51 -13
  2. trimkv_weights.pth +1 -1
config.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
- "_attn_implementation_autoset": true,
3
  "architectures": [
4
- "Qwen3ForCausalLM"
5
  ],
6
  "attention_bias": false,
7
  "attention_dropout": 0.0,
@@ -12,36 +12,74 @@
12
  "buffer_size": 128,
13
  "compress_memory": true,
14
  "compress_strategy": "alpha",
 
15
  "eos_token_id": 151645,
 
16
  "head_dim": 128,
17
  "hidden_act": "silu",
18
  "hidden_size": 2560,
19
  "initializer_range": 0.02,
20
  "intermediate_size": 9728,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
  "logit_block_size": 8192,
22
  "max_position_embeddings": 40960,
23
  "max_seq_len": 32768,
24
  "max_window_layers": 36,
25
- "memory_size": 512.0,
26
  "model_type": "qwen3",
27
  "num_attention_heads": 32,
28
  "num_hidden_layers": 36,
29
  "num_key_value_heads": 8,
 
 
 
 
 
30
  "rms_norm_eps": 1e-06,
31
  "rope_scaling": null,
32
  "rope_theta": 1000000,
33
- "skip_layers": 0,
34
  "sliding_window": null,
35
  "tie_word_embeddings": true,
36
- "torch_dtype": "bfloat16",
37
  "trainable_params": "self_attn.retention_gate",
38
- "transformers_version": "4.51.0",
39
  "use_cache": false,
40
  "use_sliding_window": false,
41
- "vocab_size": 151936,
42
- "retention_gate": "rg",
43
- "retention_gate_bias_init": 8.0,
44
- "retention_gate_intermediate_size": 512,
45
- "retention_weight": 1.0,
46
- "rg_dropout": 0.0
47
- }
 
1
  {
2
+ "alpha_threshold": 0.0,
3
  "architectures": [
4
+ "TrimKVQwen3ForCausalLM"
5
  ],
6
  "attention_bias": false,
7
  "attention_dropout": 0.0,
 
12
  "buffer_size": 128,
13
  "compress_memory": true,
14
  "compress_strategy": "alpha",
15
+ "dtype": "bfloat16",
16
  "eos_token_id": 151645,
17
+ "floor_budget_ratio": 0.0,
18
  "head_dim": 128,
19
  "hidden_act": "silu",
20
  "hidden_size": 2560,
21
  "initializer_range": 0.02,
22
  "intermediate_size": 9728,
23
+ "layer_types": [
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention",
52
+ "full_attention",
53
+ "full_attention",
54
+ "full_attention",
55
+ "full_attention",
56
+ "full_attention",
57
+ "full_attention",
58
+ "full_attention",
59
+ "full_attention"
60
+ ],
61
  "logit_block_size": 8192,
62
  "max_position_embeddings": 40960,
63
  "max_seq_len": 32768,
64
  "max_window_layers": 36,
65
+ "memory_size": 256.0,
66
  "model_type": "qwen3",
67
  "num_attention_heads": 32,
68
  "num_hidden_layers": 36,
69
  "num_key_value_heads": 8,
70
+ "retention_gate": "rg",
71
+ "retention_gate_bias_init": 18.0,
72
+ "retention_gate_intermediate_size": 512,
73
+ "retention_weight": 1.0,
74
+ "rg_dropout": 0.0,
75
  "rms_norm_eps": 1e-06,
76
  "rope_scaling": null,
77
  "rope_theta": 1000000,
 
78
  "sliding_window": null,
79
  "tie_word_embeddings": true,
 
80
  "trainable_params": "self_attn.retention_gate",
81
+ "transformers_version": "4.57.1",
82
  "use_cache": false,
83
  "use_sliding_window": false,
84
+ "vocab_size": 151936
85
+ }
 
 
 
 
 
trimkv_weights.pth CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b576910d113419a5bdf5a0ad170637ac36f35c548bac783ee87ab4f31f98ea3e
3
  size 94725981
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:825e0c83489aeae0869a3f6957422212d40922c91cdb20af719bd889e66e7b9b
3
  size 94725981