Hugging Face's logo

wsgpolar
/
Anura-R1-Ultra

Image-Text-to-Text
MLX
Safetensors
qwen3_5_moe
anura
genysis-iq
agentic
coding
vision
tool-use
conversational
4-bit precision
Model card Files
xet
 Community
wsgpolar commited on
Commit
fcbae76
·
verified ·
1 Parent(s): da9d400

Back up training/adapters

Browse files
Files changed (40) hide show
  1. training/adapters/anura-r1-ultra-lora-r2/adapter_config.json +111 -0
  2. training/adapters/anura-r1-ultra-lora-r2/checkpoint-00030.safetensors +3 -0
  3. training/adapters/anura-r1-ultra-lora-r2/checkpoint-00060.safetensors +3 -0
  4. training/adapters/anura-r1-ultra-lora-r2/checkpoint-00090.safetensors +3 -0
  5. training/adapters/anura-r1-ultra-lora-r2/checkpoint-00120.safetensors +3 -0
  6. training/adapters/anura-r1-ultra-lora-r2/train_args.json +30 -0
  7. training/adapters/anura-r1-ultra-lora-r3/adapter_config.json +111 -0
  8. training/adapters/anura-r1-ultra-lora-r3/adapters.safetensors +3 -0
  9. training/adapters/anura-r1-ultra-lora-r3/checkpoint-00040.safetensors +3 -0
  10. training/adapters/anura-r1-ultra-lora-r3/checkpoint-00080.safetensors +3 -0
  11. training/adapters/anura-r1-ultra-lora-r3/checkpoint-00120.safetensors +3 -0
  12. training/adapters/anura-r1-ultra-lora-r3/final.safetensors +3 -0
  13. training/adapters/anura-r1-ultra-lora-r3/train_args.json +30 -0
  14. training/adapters/anura-r1-ultra-lora/adapter_config.json +111 -0
  15. training/adapters/anura-r1-ultra-lora/adapters.safetensors +3 -0
  16. training/adapters/anura-r1-ultra-lora/checkpoint-00040.safetensors +3 -0
  17. training/adapters/anura-r1-ultra-lora/checkpoint-00080.safetensors +3 -0
  18. training/adapters/anura-r1-ultra-lora/checkpoint-00120.safetensors +3 -0
  19. training/adapters/anura-r1-ultra-lora/checkpoint-00160.safetensors +3 -0
  20. training/adapters/anura-r1-ultra-lora/checkpoint-00200.safetensors +3 -0
  21. training/adapters/anura-r1-ultra-lora/checkpoint-00240.safetensors +3 -0
  22. training/adapters/anura-r1-ultra-lora/checkpoint-00280.safetensors +3 -0
  23. training/adapters/anura-r1-ultra-lora/checkpoint-00320.safetensors +3 -0
  24. training/adapters/anura-r1-ultra-lora/checkpoint-00360.safetensors +3 -0
  25. training/adapters/anura-r1-ultra-lora/step80-backup.safetensors +3 -0
  26. training/adapters/anura-r1-ultra-lora/train_args.json +30 -0
  27. training/adapters/dev-lw/adapter_config.json +61 -0
  28. training/adapters/dev-lw/adapters.safetensors +3 -0
  29. training/adapters/dev-lw/checkpoint-00020.safetensors +3 -0
  30. training/adapters/dev-lw/checkpoint-00040.safetensors +3 -0
  31. training/adapters/dev-lw/train_args.json +28 -0
  32. training/adapters/dev-test/adapter_config.json +61 -0
  33. training/adapters/dev-test/adapters.safetensors +3 -0
  34. training/adapters/dev-test/checkpoint-00020.safetensors +3 -0
  35. training/adapters/dev-test/checkpoint-00040.safetensors +3 -0
  36. training/adapters/dev-test/train_args.json +25 -0
  37. training/adapters/r2-step120/adapter_config.json +111 -0
  38. training/adapters/r2-step120/adapters.safetensors +3 -0
  39. training/adapters/r3-final/adapter_config.json +111 -0
  40. training/adapters/r3-final/adapters.safetensors +3 -0
training/adapters/anura-r1-ultra-lora-r2/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.39.mlp.shared_expert.up_proj",
7
+ "language_model.model.layers.39.mlp.shared_expert.down_proj",
8
+ "language_model.model.layers.39.mlp.shared_expert.gate_proj",
9
+ "language_model.model.layers.39.self_attn.o_proj",
10
+ "language_model.model.layers.39.self_attn.v_proj",
11
+ "language_model.model.layers.39.self_attn.k_proj",
12
+ "language_model.model.layers.39.self_attn.q_proj",
13
+ "language_model.model.layers.38.mlp.shared_expert.up_proj",
14
+ "language_model.model.layers.38.mlp.shared_expert.down_proj",
15
+ "language_model.model.layers.38.mlp.shared_expert.gate_proj",
16
+ "language_model.model.layers.38.linear_attn.out_proj",
17
+ "language_model.model.layers.38.linear_attn.in_proj_z",
18
+ "language_model.model.layers.38.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.37.mlp.shared_expert.up_proj",
20
+ "language_model.model.layers.37.mlp.shared_expert.down_proj",
21
+ "language_model.model.layers.37.mlp.shared_expert.gate_proj",
22
+ "language_model.model.layers.37.linear_attn.out_proj",
23
+ "language_model.model.layers.37.linear_attn.in_proj_z",
24
+ "language_model.model.layers.37.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.36.mlp.shared_expert.up_proj",
26
+ "language_model.model.layers.36.mlp.shared_expert.down_proj",
27
+ "language_model.model.layers.36.mlp.shared_expert.gate_proj",
28
+ "language_model.model.layers.36.linear_attn.out_proj",
29
+ "language_model.model.layers.36.linear_attn.in_proj_z",
30
+ "language_model.model.layers.36.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.35.mlp.shared_expert.up_proj",
32
+ "language_model.model.layers.35.mlp.shared_expert.down_proj",
33
+ "language_model.model.layers.35.mlp.shared_expert.gate_proj",
34
+ "language_model.model.layers.35.self_attn.o_proj",
35
+ "language_model.model.layers.35.self_attn.v_proj",
36
+ "language_model.model.layers.35.self_attn.k_proj",
37
+ "language_model.model.layers.35.self_attn.q_proj",
38
+ "language_model.model.layers.34.mlp.shared_expert.up_proj",
39
+ "language_model.model.layers.34.mlp.shared_expert.down_proj",
40
+ "language_model.model.layers.34.mlp.shared_expert.gate_proj",
41
+ "language_model.model.layers.34.linear_attn.out_proj",
42
+ "language_model.model.layers.34.linear_attn.in_proj_z",
43
+ "language_model.model.layers.34.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.33.mlp.shared_expert.up_proj",
45
+ "language_model.model.layers.33.mlp.shared_expert.down_proj",
46
+ "language_model.model.layers.33.mlp.shared_expert.gate_proj",
47
+ "language_model.model.layers.33.linear_attn.out_proj",
48
+ "language_model.model.layers.33.linear_attn.in_proj_z",
49
+ "language_model.model.layers.33.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.32.mlp.shared_expert.up_proj",
51
+ "language_model.model.layers.32.mlp.shared_expert.down_proj",
52
+ "language_model.model.layers.32.mlp.shared_expert.gate_proj",
53
+ "language_model.model.layers.32.linear_attn.out_proj",
54
+ "language_model.model.layers.32.linear_attn.in_proj_z",
55
+ "language_model.model.layers.32.linear_attn.in_proj_qkv",
56
+ "language_model.model.layers.31.mlp.shared_expert.up_proj",
57
+ "language_model.model.layers.31.mlp.shared_expert.down_proj",
58
+ "language_model.model.layers.31.mlp.shared_expert.gate_proj",
59
+ "language_model.model.layers.31.self_attn.o_proj",
60
+ "language_model.model.layers.31.self_attn.v_proj",
61
+ "language_model.model.layers.31.self_attn.k_proj",
62
+ "language_model.model.layers.31.self_attn.q_proj",
63
+ "language_model.model.layers.30.mlp.shared_expert.up_proj",
64
+ "language_model.model.layers.30.mlp.shared_expert.down_proj",
65
+ "language_model.model.layers.30.mlp.shared_expert.gate_proj",
66
+ "language_model.model.layers.30.linear_attn.out_proj",
67
+ "language_model.model.layers.30.linear_attn.in_proj_z",
68
+ "language_model.model.layers.30.linear_attn.in_proj_qkv",
69
+ "language_model.model.layers.29.mlp.shared_expert.up_proj",
70
+ "language_model.model.layers.29.mlp.shared_expert.down_proj",
71
+ "language_model.model.layers.29.mlp.shared_expert.gate_proj",
72
+ "language_model.model.layers.29.linear_attn.out_proj",
73
+ "language_model.model.layers.29.linear_attn.in_proj_z",
74
+ "language_model.model.layers.29.linear_attn.in_proj_qkv",
75
+ "language_model.model.layers.28.mlp.shared_expert.up_proj",
76
+ "language_model.model.layers.28.mlp.shared_expert.down_proj",
77
+ "language_model.model.layers.28.mlp.shared_expert.gate_proj",
78
+ "language_model.model.layers.28.linear_attn.out_proj",
79
+ "language_model.model.layers.28.linear_attn.in_proj_z",
80
+ "language_model.model.layers.28.linear_attn.in_proj_qkv",
81
+ "language_model.model.layers.27.mlp.shared_expert.up_proj",
82
+ "language_model.model.layers.27.mlp.shared_expert.down_proj",
83
+ "language_model.model.layers.27.mlp.shared_expert.gate_proj",
84
+ "language_model.model.layers.27.self_attn.o_proj",
85
+ "language_model.model.layers.27.self_attn.v_proj",
86
+ "language_model.model.layers.27.self_attn.k_proj",
87
+ "language_model.model.layers.27.self_attn.q_proj",
88
+ "language_model.model.layers.26.mlp.shared_expert.up_proj",
89
+ "language_model.model.layers.26.mlp.shared_expert.down_proj",
90
+ "language_model.model.layers.26.mlp.shared_expert.gate_proj",
91
+ "language_model.model.layers.26.linear_attn.out_proj",
92
+ "language_model.model.layers.26.linear_attn.in_proj_z",
93
+ "language_model.model.layers.26.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.25.mlp.shared_expert.up_proj",
95
+ "language_model.model.layers.25.mlp.shared_expert.down_proj",
96
+ "language_model.model.layers.25.mlp.shared_expert.gate_proj",
97
+ "language_model.model.layers.25.linear_attn.out_proj",
98
+ "language_model.model.layers.25.linear_attn.in_proj_z",
99
+ "language_model.model.layers.25.linear_attn.in_proj_qkv",
100
+ "language_model.model.layers.24.mlp.shared_expert.up_proj",
101
+ "language_model.model.layers.24.mlp.shared_expert.down_proj",
102
+ "language_model.model.layers.24.mlp.shared_expert.gate_proj",
103
+ "language_model.model.layers.24.linear_attn.out_proj",
104
+ "language_model.model.layers.24.linear_attn.in_proj_z",
105
+ "language_model.model.layers.24.linear_attn.in_proj_qkv"
106
+ ],
107
+ "rank": 16,
108
+ "scale": 2.0
109
+ },
110
+ "num_layers": -1
111
+ }
training/adapters/anura-r1-ultra-lora-r2/checkpoint-00030.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1195ef4c2dc0d47b7caf39877c7668edea24a850816caec78b8b388e9510b45c
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r2/checkpoint-00060.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:967e6a9fda514395fd884d76bedaae5ce301d1e26cedae01487eadcdd1dc8283
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r2/checkpoint-00090.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:072746f219cf17584d557ed96bfd3b8a2b604bc12dd3a663cfc229dff3b29be0
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r2/checkpoint-00120.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87dd5bdb3d518a34063a40ebbf15fa1a438412c3d4b04f8462f394e4370a34c4
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r2/train_args.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "models/base/Qwen3.6-35B-A3B-4bit",
3
+ "data": "data/train_r2.jsonl",
4
+ "valid": "data/valid.jsonl",
5
+ "out": "models/adapters/anura-r1-ultra-lora-r2",
6
+ "rank": 16,
7
+ "lora_layers": 16,
8
+ "alpha": 32.0,
9
+ "dropout": 0.0,
10
+ "lr": 5e-05,
11
+ "min_lr": 5e-06,
12
+ "epochs": 1.0,
13
+ "warmup": 10,
14
+ "batch_tokens": 5120,
15
+ "max_len": 5120,
16
+ "grad_accum": 1,
17
+ "grad_clip": 1.0,
18
+ "grad_checkpoint": false,
19
+ "gdn_chunk": 32,
20
+ "loss_chunk": 512,
21
+ "attn_chunk": 1024,
22
+ "layerwise": true,
23
+ "eval_every": 30,
24
+ "save_every": 30,
25
+ "seed": 2,
26
+ "resume": "models/adapters/anura-r1-ultra-lora",
27
+ "start_step": 0,
28
+ "cache_limit_gb": -1,
29
+ "max_steps": 0
30
+ }
training/adapters/anura-r1-ultra-lora-r3/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.39.mlp.shared_expert.up_proj",
7
+ "language_model.model.layers.39.mlp.shared_expert.down_proj",
8
+ "language_model.model.layers.39.mlp.shared_expert.gate_proj",
9
+ "language_model.model.layers.39.self_attn.o_proj",
10
+ "language_model.model.layers.39.self_attn.v_proj",
11
+ "language_model.model.layers.39.self_attn.k_proj",
12
+ "language_model.model.layers.39.self_attn.q_proj",
13
+ "language_model.model.layers.38.mlp.shared_expert.up_proj",
14
+ "language_model.model.layers.38.mlp.shared_expert.down_proj",
15
+ "language_model.model.layers.38.mlp.shared_expert.gate_proj",
16
+ "language_model.model.layers.38.linear_attn.out_proj",
17
+ "language_model.model.layers.38.linear_attn.in_proj_z",
18
+ "language_model.model.layers.38.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.37.mlp.shared_expert.up_proj",
20
+ "language_model.model.layers.37.mlp.shared_expert.down_proj",
21
+ "language_model.model.layers.37.mlp.shared_expert.gate_proj",
22
+ "language_model.model.layers.37.linear_attn.out_proj",
23
+ "language_model.model.layers.37.linear_attn.in_proj_z",
24
+ "language_model.model.layers.37.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.36.mlp.shared_expert.up_proj",
26
+ "language_model.model.layers.36.mlp.shared_expert.down_proj",
27
+ "language_model.model.layers.36.mlp.shared_expert.gate_proj",
28
+ "language_model.model.layers.36.linear_attn.out_proj",
29
+ "language_model.model.layers.36.linear_attn.in_proj_z",
30
+ "language_model.model.layers.36.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.35.mlp.shared_expert.up_proj",
32
+ "language_model.model.layers.35.mlp.shared_expert.down_proj",
33
+ "language_model.model.layers.35.mlp.shared_expert.gate_proj",
34
+ "language_model.model.layers.35.self_attn.o_proj",
35
+ "language_model.model.layers.35.self_attn.v_proj",
36
+ "language_model.model.layers.35.self_attn.k_proj",
37
+ "language_model.model.layers.35.self_attn.q_proj",
38
+ "language_model.model.layers.34.mlp.shared_expert.up_proj",
39
+ "language_model.model.layers.34.mlp.shared_expert.down_proj",
40
+ "language_model.model.layers.34.mlp.shared_expert.gate_proj",
41
+ "language_model.model.layers.34.linear_attn.out_proj",
42
+ "language_model.model.layers.34.linear_attn.in_proj_z",
43
+ "language_model.model.layers.34.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.33.mlp.shared_expert.up_proj",
45
+ "language_model.model.layers.33.mlp.shared_expert.down_proj",
46
+ "language_model.model.layers.33.mlp.shared_expert.gate_proj",
47
+ "language_model.model.layers.33.linear_attn.out_proj",
48
+ "language_model.model.layers.33.linear_attn.in_proj_z",
49
+ "language_model.model.layers.33.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.32.mlp.shared_expert.up_proj",
51
+ "language_model.model.layers.32.mlp.shared_expert.down_proj",
52
+ "language_model.model.layers.32.mlp.shared_expert.gate_proj",
53
+ "language_model.model.layers.32.linear_attn.out_proj",
54
+ "language_model.model.layers.32.linear_attn.in_proj_z",
55
+ "language_model.model.layers.32.linear_attn.in_proj_qkv",
56
+ "language_model.model.layers.31.mlp.shared_expert.up_proj",
57
+ "language_model.model.layers.31.mlp.shared_expert.down_proj",
58
+ "language_model.model.layers.31.mlp.shared_expert.gate_proj",
59
+ "language_model.model.layers.31.self_attn.o_proj",
60
+ "language_model.model.layers.31.self_attn.v_proj",
61
+ "language_model.model.layers.31.self_attn.k_proj",
62
+ "language_model.model.layers.31.self_attn.q_proj",
63
+ "language_model.model.layers.30.mlp.shared_expert.up_proj",
64
+ "language_model.model.layers.30.mlp.shared_expert.down_proj",
65
+ "language_model.model.layers.30.mlp.shared_expert.gate_proj",
66
+ "language_model.model.layers.30.linear_attn.out_proj",
67
+ "language_model.model.layers.30.linear_attn.in_proj_z",
68
+ "language_model.model.layers.30.linear_attn.in_proj_qkv",
69
+ "language_model.model.layers.29.mlp.shared_expert.up_proj",
70
+ "language_model.model.layers.29.mlp.shared_expert.down_proj",
71
+ "language_model.model.layers.29.mlp.shared_expert.gate_proj",
72
+ "language_model.model.layers.29.linear_attn.out_proj",
73
+ "language_model.model.layers.29.linear_attn.in_proj_z",
74
+ "language_model.model.layers.29.linear_attn.in_proj_qkv",
75
+ "language_model.model.layers.28.mlp.shared_expert.up_proj",
76
+ "language_model.model.layers.28.mlp.shared_expert.down_proj",
77
+ "language_model.model.layers.28.mlp.shared_expert.gate_proj",
78
+ "language_model.model.layers.28.linear_attn.out_proj",
79
+ "language_model.model.layers.28.linear_attn.in_proj_z",
80
+ "language_model.model.layers.28.linear_attn.in_proj_qkv",
81
+ "language_model.model.layers.27.mlp.shared_expert.up_proj",
82
+ "language_model.model.layers.27.mlp.shared_expert.down_proj",
83
+ "language_model.model.layers.27.mlp.shared_expert.gate_proj",
84
+ "language_model.model.layers.27.self_attn.o_proj",
85
+ "language_model.model.layers.27.self_attn.v_proj",
86
+ "language_model.model.layers.27.self_attn.k_proj",
87
+ "language_model.model.layers.27.self_attn.q_proj",
88
+ "language_model.model.layers.26.mlp.shared_expert.up_proj",
89
+ "language_model.model.layers.26.mlp.shared_expert.down_proj",
90
+ "language_model.model.layers.26.mlp.shared_expert.gate_proj",
91
+ "language_model.model.layers.26.linear_attn.out_proj",
92
+ "language_model.model.layers.26.linear_attn.in_proj_z",
93
+ "language_model.model.layers.26.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.25.mlp.shared_expert.up_proj",
95
+ "language_model.model.layers.25.mlp.shared_expert.down_proj",
96
+ "language_model.model.layers.25.mlp.shared_expert.gate_proj",
97
+ "language_model.model.layers.25.linear_attn.out_proj",
98
+ "language_model.model.layers.25.linear_attn.in_proj_z",
99
+ "language_model.model.layers.25.linear_attn.in_proj_qkv",
100
+ "language_model.model.layers.24.mlp.shared_expert.up_proj",
101
+ "language_model.model.layers.24.mlp.shared_expert.down_proj",
102
+ "language_model.model.layers.24.mlp.shared_expert.gate_proj",
103
+ "language_model.model.layers.24.linear_attn.out_proj",
104
+ "language_model.model.layers.24.linear_attn.in_proj_z",
105
+ "language_model.model.layers.24.linear_attn.in_proj_qkv"
106
+ ],
107
+ "rank": 16,
108
+ "scale": 2.0
109
+ },
110
+ "num_layers": -1
111
+ }
training/adapters/anura-r1-ultra-lora-r3/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0dbbf7c6907cecd6b82679eb8675599e28580760143d8a9d84b9ada834c5ac91
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r3/checkpoint-00040.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85e17aa3dcc2f84139b607d759ce73f7ee766824f49a340dce2d01eb15d3f460
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r3/checkpoint-00080.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49a0ca5ee2ae539813faf9e90e67b8e2bfc68fdf372e890ddb8c77c0e9c92950
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r3/checkpoint-00120.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:332e533b81e9acc39568c9c6508b2dc7e1d63515617582e8cc9a04af99b3ab8a
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r3/final.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0dbbf7c6907cecd6b82679eb8675599e28580760143d8a9d84b9ada834c5ac91
3
+ size 30697232
training/adapters/anura-r1-ultra-lora-r3/train_args.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "models/base/Qwen3.6-35B-A3B-4bit",
3
+ "data": "data/train_r3.jsonl",
4
+ "valid": "data/valid.jsonl",
5
+ "out": "models/adapters/anura-r1-ultra-lora-r3",
6
+ "rank": 16,
7
+ "lora_layers": 16,
8
+ "alpha": 32.0,
9
+ "dropout": 0.0,
10
+ "lr": 3e-05,
11
+ "min_lr": 3e-06,
12
+ "epochs": 1.0,
13
+ "warmup": 10,
14
+ "batch_tokens": 5120,
15
+ "max_len": 5120,
16
+ "grad_accum": 1,
17
+ "grad_clip": 1.0,
18
+ "grad_checkpoint": false,
19
+ "gdn_chunk": 32,
20
+ "loss_chunk": 512,
21
+ "attn_chunk": 1024,
22
+ "layerwise": true,
23
+ "eval_every": 40,
24
+ "save_every": 40,
25
+ "seed": 3,
26
+ "resume": "models/adapters/anura-r1-ultra-lora",
27
+ "start_step": 0,
28
+ "cache_limit_gb": -1,
29
+ "max_steps": 0
30
+ }
training/adapters/anura-r1-ultra-lora/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.39.mlp.shared_expert.up_proj",
7
+ "language_model.model.layers.39.mlp.shared_expert.down_proj",
8
+ "language_model.model.layers.39.mlp.shared_expert.gate_proj",
9
+ "language_model.model.layers.39.self_attn.o_proj",
10
+ "language_model.model.layers.39.self_attn.v_proj",
11
+ "language_model.model.layers.39.self_attn.k_proj",
12
+ "language_model.model.layers.39.self_attn.q_proj",
13
+ "language_model.model.layers.38.mlp.shared_expert.up_proj",
14
+ "language_model.model.layers.38.mlp.shared_expert.down_proj",
15
+ "language_model.model.layers.38.mlp.shared_expert.gate_proj",
16
+ "language_model.model.layers.38.linear_attn.out_proj",
17
+ "language_model.model.layers.38.linear_attn.in_proj_z",
18
+ "language_model.model.layers.38.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.37.mlp.shared_expert.up_proj",
20
+ "language_model.model.layers.37.mlp.shared_expert.down_proj",
21
+ "language_model.model.layers.37.mlp.shared_expert.gate_proj",
22
+ "language_model.model.layers.37.linear_attn.out_proj",
23
+ "language_model.model.layers.37.linear_attn.in_proj_z",
24
+ "language_model.model.layers.37.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.36.mlp.shared_expert.up_proj",
26
+ "language_model.model.layers.36.mlp.shared_expert.down_proj",
27
+ "language_model.model.layers.36.mlp.shared_expert.gate_proj",
28
+ "language_model.model.layers.36.linear_attn.out_proj",
29
+ "language_model.model.layers.36.linear_attn.in_proj_z",
30
+ "language_model.model.layers.36.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.35.mlp.shared_expert.up_proj",
32
+ "language_model.model.layers.35.mlp.shared_expert.down_proj",
33
+ "language_model.model.layers.35.mlp.shared_expert.gate_proj",
34
+ "language_model.model.layers.35.self_attn.o_proj",
35
+ "language_model.model.layers.35.self_attn.v_proj",
36
+ "language_model.model.layers.35.self_attn.k_proj",
37
+ "language_model.model.layers.35.self_attn.q_proj",
38
+ "language_model.model.layers.34.mlp.shared_expert.up_proj",
39
+ "language_model.model.layers.34.mlp.shared_expert.down_proj",
40
+ "language_model.model.layers.34.mlp.shared_expert.gate_proj",
41
+ "language_model.model.layers.34.linear_attn.out_proj",
42
+ "language_model.model.layers.34.linear_attn.in_proj_z",
43
+ "language_model.model.layers.34.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.33.mlp.shared_expert.up_proj",
45
+ "language_model.model.layers.33.mlp.shared_expert.down_proj",
46
+ "language_model.model.layers.33.mlp.shared_expert.gate_proj",
47
+ "language_model.model.layers.33.linear_attn.out_proj",
48
+ "language_model.model.layers.33.linear_attn.in_proj_z",
49
+ "language_model.model.layers.33.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.32.mlp.shared_expert.up_proj",
51
+ "language_model.model.layers.32.mlp.shared_expert.down_proj",
52
+ "language_model.model.layers.32.mlp.shared_expert.gate_proj",
53
+ "language_model.model.layers.32.linear_attn.out_proj",
54
+ "language_model.model.layers.32.linear_attn.in_proj_z",
55
+ "language_model.model.layers.32.linear_attn.in_proj_qkv",
56
+ "language_model.model.layers.31.mlp.shared_expert.up_proj",
57
+ "language_model.model.layers.31.mlp.shared_expert.down_proj",
58
+ "language_model.model.layers.31.mlp.shared_expert.gate_proj",
59
+ "language_model.model.layers.31.self_attn.o_proj",
60
+ "language_model.model.layers.31.self_attn.v_proj",
61
+ "language_model.model.layers.31.self_attn.k_proj",
62
+ "language_model.model.layers.31.self_attn.q_proj",
63
+ "language_model.model.layers.30.mlp.shared_expert.up_proj",
64
+ "language_model.model.layers.30.mlp.shared_expert.down_proj",
65
+ "language_model.model.layers.30.mlp.shared_expert.gate_proj",
66
+ "language_model.model.layers.30.linear_attn.out_proj",
67
+ "language_model.model.layers.30.linear_attn.in_proj_z",
68
+ "language_model.model.layers.30.linear_attn.in_proj_qkv",
69
+ "language_model.model.layers.29.mlp.shared_expert.up_proj",
70
+ "language_model.model.layers.29.mlp.shared_expert.down_proj",
71
+ "language_model.model.layers.29.mlp.shared_expert.gate_proj",
72
+ "language_model.model.layers.29.linear_attn.out_proj",
73
+ "language_model.model.layers.29.linear_attn.in_proj_z",
74
+ "language_model.model.layers.29.linear_attn.in_proj_qkv",
75
+ "language_model.model.layers.28.mlp.shared_expert.up_proj",
76
+ "language_model.model.layers.28.mlp.shared_expert.down_proj",
77
+ "language_model.model.layers.28.mlp.shared_expert.gate_proj",
78
+ "language_model.model.layers.28.linear_attn.out_proj",
79
+ "language_model.model.layers.28.linear_attn.in_proj_z",
80
+ "language_model.model.layers.28.linear_attn.in_proj_qkv",
81
+ "language_model.model.layers.27.mlp.shared_expert.up_proj",
82
+ "language_model.model.layers.27.mlp.shared_expert.down_proj",
83
+ "language_model.model.layers.27.mlp.shared_expert.gate_proj",
84
+ "language_model.model.layers.27.self_attn.o_proj",
85
+ "language_model.model.layers.27.self_attn.v_proj",
86
+ "language_model.model.layers.27.self_attn.k_proj",
87
+ "language_model.model.layers.27.self_attn.q_proj",
88
+ "language_model.model.layers.26.mlp.shared_expert.up_proj",
89
+ "language_model.model.layers.26.mlp.shared_expert.down_proj",
90
+ "language_model.model.layers.26.mlp.shared_expert.gate_proj",
91
+ "language_model.model.layers.26.linear_attn.out_proj",
92
+ "language_model.model.layers.26.linear_attn.in_proj_z",
93
+ "language_model.model.layers.26.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.25.mlp.shared_expert.up_proj",
95
+ "language_model.model.layers.25.mlp.shared_expert.down_proj",
96
+ "language_model.model.layers.25.mlp.shared_expert.gate_proj",
97
+ "language_model.model.layers.25.linear_attn.out_proj",
98
+ "language_model.model.layers.25.linear_attn.in_proj_z",
99
+ "language_model.model.layers.25.linear_attn.in_proj_qkv",
100
+ "language_model.model.layers.24.mlp.shared_expert.up_proj",
101
+ "language_model.model.layers.24.mlp.shared_expert.down_proj",
102
+ "language_model.model.layers.24.mlp.shared_expert.gate_proj",
103
+ "language_model.model.layers.24.linear_attn.out_proj",
104
+ "language_model.model.layers.24.linear_attn.in_proj_z",
105
+ "language_model.model.layers.24.linear_attn.in_proj_qkv"
106
+ ],
107
+ "rank": 16,
108
+ "scale": 2.0
109
+ },
110
+ "num_layers": -1
111
+ }
training/adapters/anura-r1-ultra-lora/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a07d03fdf9f3f99368166d18034c5fd4857c86e012952e547fc5c91b1be8c197
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00040.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36fd32353d4a1e8e45bdb5db8b8f3debf4a829ceef0cc8eb3fe82ae023e74ec7
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00080.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8c69ccbd3738ad140d8ac37ce66489adcbfb53eedd9d87fea7d078f9f7df1d5b
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00120.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df7668f7be727fd6bbebeae6ccb311242c292f2ed5acdcfaca1ac6a0982617f5
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00160.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f28b62de950c79c9b890c350f2fb98abee66fea264cc3b790b0d129179de6672
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00200.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7586b973c8ca778f4a153b198e6ee87cbe01127a96847f66bd0858e796a482f
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00240.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a07d03fdf9f3f99368166d18034c5fd4857c86e012952e547fc5c91b1be8c197
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00280.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:624988cd1a815ed371d98c09681f18e56ccab032e3d587badac79e4cfe5889ce
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00320.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a08a30de96e79f1279dfbc4c99321a06eff3ae32c0a7e6bc5a99af948928db9f
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/checkpoint-00360.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:858aade37e5e9f670754df3d0685aad198e4f126145dccbf8c7babb3aa41d617
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/step80-backup.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8c69ccbd3738ad140d8ac37ce66489adcbfb53eedd9d87fea7d078f9f7df1d5b
3
+ size 30697232
training/adapters/anura-r1-ultra-lora/train_args.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "models/base/Qwen3.6-35B-A3B-4bit",
3
+ "data": "data/train.jsonl",
4
+ "valid": "data/valid.jsonl",
5
+ "out": "models/adapters/anura-r1-ultra-lora",
6
+ "rank": 16,
7
+ "lora_layers": 16,
8
+ "alpha": 32.0,
9
+ "dropout": 0.0,
10
+ "lr": 0.0001,
11
+ "min_lr": 5e-06,
12
+ "epochs": 2.0,
13
+ "warmup": 20,
14
+ "batch_tokens": 5120,
15
+ "max_len": 5120,
16
+ "grad_accum": 1,
17
+ "grad_clip": 1.0,
18
+ "grad_checkpoint": false,
19
+ "gdn_chunk": 32,
20
+ "loss_chunk": 512,
21
+ "attn_chunk": 1024,
22
+ "layerwise": true,
23
+ "eval_every": 40,
24
+ "save_every": 40,
25
+ "seed": 1,
26
+ "resume": "models/adapters/anura-r1-ultra-lora",
27
+ "start_step": 80,
28
+ "cache_limit_gb": -1,
29
+ "max_steps": 0
30
+ }
training/adapters/dev-lw/adapter_config.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.23.mlp.up_proj",
7
+ "language_model.model.layers.23.mlp.down_proj",
8
+ "language_model.model.layers.23.mlp.gate_proj",
9
+ "language_model.model.layers.23.self_attn.o_proj",
10
+ "language_model.model.layers.23.self_attn.v_proj",
11
+ "language_model.model.layers.23.self_attn.k_proj",
12
+ "language_model.model.layers.23.self_attn.q_proj",
13
+ "language_model.model.layers.22.mlp.up_proj",
14
+ "language_model.model.layers.22.mlp.down_proj",
15
+ "language_model.model.layers.22.mlp.gate_proj",
16
+ "language_model.model.layers.22.linear_attn.out_proj",
17
+ "language_model.model.layers.22.linear_attn.in_proj_z",
18
+ "language_model.model.layers.22.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.21.mlp.up_proj",
20
+ "language_model.model.layers.21.mlp.down_proj",
21
+ "language_model.model.layers.21.mlp.gate_proj",
22
+ "language_model.model.layers.21.linear_attn.out_proj",
23
+ "language_model.model.layers.21.linear_attn.in_proj_z",
24
+ "language_model.model.layers.21.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.20.mlp.up_proj",
26
+ "language_model.model.layers.20.mlp.down_proj",
27
+ "language_model.model.layers.20.mlp.gate_proj",
28
+ "language_model.model.layers.20.linear_attn.out_proj",
29
+ "language_model.model.layers.20.linear_attn.in_proj_z",
30
+ "language_model.model.layers.20.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.19.mlp.up_proj",
32
+ "language_model.model.layers.19.mlp.down_proj",
33
+ "language_model.model.layers.19.mlp.gate_proj",
34
+ "language_model.model.layers.19.self_attn.o_proj",
35
+ "language_model.model.layers.19.self_attn.v_proj",
36
+ "language_model.model.layers.19.self_attn.k_proj",
37
+ "language_model.model.layers.19.self_attn.q_proj",
38
+ "language_model.model.layers.18.mlp.up_proj",
39
+ "language_model.model.layers.18.mlp.down_proj",
40
+ "language_model.model.layers.18.mlp.gate_proj",
41
+ "language_model.model.layers.18.linear_attn.out_proj",
42
+ "language_model.model.layers.18.linear_attn.in_proj_z",
43
+ "language_model.model.layers.18.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.17.mlp.up_proj",
45
+ "language_model.model.layers.17.mlp.down_proj",
46
+ "language_model.model.layers.17.mlp.gate_proj",
47
+ "language_model.model.layers.17.linear_attn.out_proj",
48
+ "language_model.model.layers.17.linear_attn.in_proj_z",
49
+ "language_model.model.layers.17.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.16.mlp.up_proj",
51
+ "language_model.model.layers.16.mlp.down_proj",
52
+ "language_model.model.layers.16.mlp.gate_proj",
53
+ "language_model.model.layers.16.linear_attn.out_proj",
54
+ "language_model.model.layers.16.linear_attn.in_proj_z",
55
+ "language_model.model.layers.16.linear_attn.in_proj_qkv"
56
+ ],
57
+ "rank": 16,
58
+ "scale": 2.0
59
+ },
60
+ "num_layers": -1
61
+ }
training/adapters/dev-lw/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:febc58421b629a432c250cff48a189256879544bc058b9ac53a534e9d79900a5
3
+ size 13643964
training/adapters/dev-lw/checkpoint-00020.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a4c47471c4a8ce8d6120610e7f6429a0a1d420ae12507a839cf0502aad227571
3
+ size 13643964
training/adapters/dev-lw/checkpoint-00040.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:febc58421b629a432c250cff48a189256879544bc058b9ac53a534e9d79900a5
3
+ size 13643964
training/adapters/dev-lw/train_args.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "models/dev/Qwen3.5-0.8B-4bit",
3
+ "data": "data/dev_train.jsonl",
4
+ "valid": "data/dev_valid.jsonl",
5
+ "out": "models/adapters/dev-lw",
6
+ "rank": 16,
7
+ "lora_layers": 8,
8
+ "alpha": 32,
9
+ "dropout": 0.0,
10
+ "lr": 0.0002,
11
+ "min_lr": 5e-06,
12
+ "epochs": 2.0,
13
+ "warmup": 5,
14
+ "batch_tokens": 4096,
15
+ "max_len": 4096,
16
+ "grad_accum": 1,
17
+ "grad_clip": 1.0,
18
+ "grad_checkpoint": false,
19
+ "gdn_chunk": 32,
20
+ "loss_chunk": 512,
21
+ "attn_chunk": 1024,
22
+ "layerwise": true,
23
+ "eval_every": 20,
24
+ "save_every": 20,
25
+ "seed": 0,
26
+ "resume": null,
27
+ "max_steps": 40
28
+ }
training/adapters/dev-test/adapter_config.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.05,
5
+ "keys": [
6
+ "language_model.model.layers.23.mlp.up_proj",
7
+ "language_model.model.layers.23.mlp.down_proj",
8
+ "language_model.model.layers.23.mlp.gate_proj",
9
+ "language_model.model.layers.23.self_attn.o_proj",
10
+ "language_model.model.layers.23.self_attn.v_proj",
11
+ "language_model.model.layers.23.self_attn.k_proj",
12
+ "language_model.model.layers.23.self_attn.q_proj",
13
+ "language_model.model.layers.22.mlp.up_proj",
14
+ "language_model.model.layers.22.mlp.down_proj",
15
+ "language_model.model.layers.22.mlp.gate_proj",
16
+ "language_model.model.layers.22.linear_attn.out_proj",
17
+ "language_model.model.layers.22.linear_attn.in_proj_z",
18
+ "language_model.model.layers.22.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.21.mlp.up_proj",
20
+ "language_model.model.layers.21.mlp.down_proj",
21
+ "language_model.model.layers.21.mlp.gate_proj",
22
+ "language_model.model.layers.21.linear_attn.out_proj",
23
+ "language_model.model.layers.21.linear_attn.in_proj_z",
24
+ "language_model.model.layers.21.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.20.mlp.up_proj",
26
+ "language_model.model.layers.20.mlp.down_proj",
27
+ "language_model.model.layers.20.mlp.gate_proj",
28
+ "language_model.model.layers.20.linear_attn.out_proj",
29
+ "language_model.model.layers.20.linear_attn.in_proj_z",
30
+ "language_model.model.layers.20.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.19.mlp.up_proj",
32
+ "language_model.model.layers.19.mlp.down_proj",
33
+ "language_model.model.layers.19.mlp.gate_proj",
34
+ "language_model.model.layers.19.self_attn.o_proj",
35
+ "language_model.model.layers.19.self_attn.v_proj",
36
+ "language_model.model.layers.19.self_attn.k_proj",
37
+ "language_model.model.layers.19.self_attn.q_proj",
38
+ "language_model.model.layers.18.mlp.up_proj",
39
+ "language_model.model.layers.18.mlp.down_proj",
40
+ "language_model.model.layers.18.mlp.gate_proj",
41
+ "language_model.model.layers.18.linear_attn.out_proj",
42
+ "language_model.model.layers.18.linear_attn.in_proj_z",
43
+ "language_model.model.layers.18.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.17.mlp.up_proj",
45
+ "language_model.model.layers.17.mlp.down_proj",
46
+ "language_model.model.layers.17.mlp.gate_proj",
47
+ "language_model.model.layers.17.linear_attn.out_proj",
48
+ "language_model.model.layers.17.linear_attn.in_proj_z",
49
+ "language_model.model.layers.17.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.16.mlp.up_proj",
51
+ "language_model.model.layers.16.mlp.down_proj",
52
+ "language_model.model.layers.16.mlp.gate_proj",
53
+ "language_model.model.layers.16.linear_attn.out_proj",
54
+ "language_model.model.layers.16.linear_attn.in_proj_z",
55
+ "language_model.model.layers.16.linear_attn.in_proj_qkv"
56
+ ],
57
+ "rank": 16,
58
+ "scale": 2.0
59
+ },
60
+ "num_layers": -1
61
+ }
training/adapters/dev-test/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32752501f1fe26135293442125412ffb4534de5b93987a282216218100003b1e
3
+ size 13643964
training/adapters/dev-test/checkpoint-00020.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f217e486140a385447bf1e83c9b843398a654fb7a45e33190dbe6d751263f7f8
3
+ size 13643964
training/adapters/dev-test/checkpoint-00040.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32752501f1fe26135293442125412ffb4534de5b93987a282216218100003b1e
3
+ size 13643964
training/adapters/dev-test/train_args.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "models/dev/Qwen3.5-0.8B-4bit",
3
+ "data": "data/dev_train.jsonl",
4
+ "valid": "data/dev_valid.jsonl",
5
+ "out": "models/adapters/dev-test",
6
+ "rank": 16,
7
+ "lora_layers": 8,
8
+ "alpha": 32,
9
+ "dropout": 0.05,
10
+ "lr": 0.0002,
11
+ "min_lr": 5e-06,
12
+ "epochs": 2.0,
13
+ "warmup": 5,
14
+ "batch_tokens": 4096,
15
+ "max_len": 4096,
16
+ "grad_accum": 1,
17
+ "grad_clip": 1.0,
18
+ "grad_checkpoint": true,
19
+ "gdn_chunk": 32,
20
+ "eval_every": 20,
21
+ "save_every": 20,
22
+ "seed": 0,
23
+ "resume": null,
24
+ "max_steps": 40
25
+ }
training/adapters/r2-step120/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.39.mlp.shared_expert.up_proj",
7
+ "language_model.model.layers.39.mlp.shared_expert.down_proj",
8
+ "language_model.model.layers.39.mlp.shared_expert.gate_proj",
9
+ "language_model.model.layers.39.self_attn.o_proj",
10
+ "language_model.model.layers.39.self_attn.v_proj",
11
+ "language_model.model.layers.39.self_attn.k_proj",
12
+ "language_model.model.layers.39.self_attn.q_proj",
13
+ "language_model.model.layers.38.mlp.shared_expert.up_proj",
14
+ "language_model.model.layers.38.mlp.shared_expert.down_proj",
15
+ "language_model.model.layers.38.mlp.shared_expert.gate_proj",
16
+ "language_model.model.layers.38.linear_attn.out_proj",
17
+ "language_model.model.layers.38.linear_attn.in_proj_z",
18
+ "language_model.model.layers.38.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.37.mlp.shared_expert.up_proj",
20
+ "language_model.model.layers.37.mlp.shared_expert.down_proj",
21
+ "language_model.model.layers.37.mlp.shared_expert.gate_proj",
22
+ "language_model.model.layers.37.linear_attn.out_proj",
23
+ "language_model.model.layers.37.linear_attn.in_proj_z",
24
+ "language_model.model.layers.37.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.36.mlp.shared_expert.up_proj",
26
+ "language_model.model.layers.36.mlp.shared_expert.down_proj",
27
+ "language_model.model.layers.36.mlp.shared_expert.gate_proj",
28
+ "language_model.model.layers.36.linear_attn.out_proj",
29
+ "language_model.model.layers.36.linear_attn.in_proj_z",
30
+ "language_model.model.layers.36.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.35.mlp.shared_expert.up_proj",
32
+ "language_model.model.layers.35.mlp.shared_expert.down_proj",
33
+ "language_model.model.layers.35.mlp.shared_expert.gate_proj",
34
+ "language_model.model.layers.35.self_attn.o_proj",
35
+ "language_model.model.layers.35.self_attn.v_proj",
36
+ "language_model.model.layers.35.self_attn.k_proj",
37
+ "language_model.model.layers.35.self_attn.q_proj",
38
+ "language_model.model.layers.34.mlp.shared_expert.up_proj",
39
+ "language_model.model.layers.34.mlp.shared_expert.down_proj",
40
+ "language_model.model.layers.34.mlp.shared_expert.gate_proj",
41
+ "language_model.model.layers.34.linear_attn.out_proj",
42
+ "language_model.model.layers.34.linear_attn.in_proj_z",
43
+ "language_model.model.layers.34.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.33.mlp.shared_expert.up_proj",
45
+ "language_model.model.layers.33.mlp.shared_expert.down_proj",
46
+ "language_model.model.layers.33.mlp.shared_expert.gate_proj",
47
+ "language_model.model.layers.33.linear_attn.out_proj",
48
+ "language_model.model.layers.33.linear_attn.in_proj_z",
49
+ "language_model.model.layers.33.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.32.mlp.shared_expert.up_proj",
51
+ "language_model.model.layers.32.mlp.shared_expert.down_proj",
52
+ "language_model.model.layers.32.mlp.shared_expert.gate_proj",
53
+ "language_model.model.layers.32.linear_attn.out_proj",
54
+ "language_model.model.layers.32.linear_attn.in_proj_z",
55
+ "language_model.model.layers.32.linear_attn.in_proj_qkv",
56
+ "language_model.model.layers.31.mlp.shared_expert.up_proj",
57
+ "language_model.model.layers.31.mlp.shared_expert.down_proj",
58
+ "language_model.model.layers.31.mlp.shared_expert.gate_proj",
59
+ "language_model.model.layers.31.self_attn.o_proj",
60
+ "language_model.model.layers.31.self_attn.v_proj",
61
+ "language_model.model.layers.31.self_attn.k_proj",
62
+ "language_model.model.layers.31.self_attn.q_proj",
63
+ "language_model.model.layers.30.mlp.shared_expert.up_proj",
64
+ "language_model.model.layers.30.mlp.shared_expert.down_proj",
65
+ "language_model.model.layers.30.mlp.shared_expert.gate_proj",
66
+ "language_model.model.layers.30.linear_attn.out_proj",
67
+ "language_model.model.layers.30.linear_attn.in_proj_z",
68
+ "language_model.model.layers.30.linear_attn.in_proj_qkv",
69
+ "language_model.model.layers.29.mlp.shared_expert.up_proj",
70
+ "language_model.model.layers.29.mlp.shared_expert.down_proj",
71
+ "language_model.model.layers.29.mlp.shared_expert.gate_proj",
72
+ "language_model.model.layers.29.linear_attn.out_proj",
73
+ "language_model.model.layers.29.linear_attn.in_proj_z",
74
+ "language_model.model.layers.29.linear_attn.in_proj_qkv",
75
+ "language_model.model.layers.28.mlp.shared_expert.up_proj",
76
+ "language_model.model.layers.28.mlp.shared_expert.down_proj",
77
+ "language_model.model.layers.28.mlp.shared_expert.gate_proj",
78
+ "language_model.model.layers.28.linear_attn.out_proj",
79
+ "language_model.model.layers.28.linear_attn.in_proj_z",
80
+ "language_model.model.layers.28.linear_attn.in_proj_qkv",
81
+ "language_model.model.layers.27.mlp.shared_expert.up_proj",
82
+ "language_model.model.layers.27.mlp.shared_expert.down_proj",
83
+ "language_model.model.layers.27.mlp.shared_expert.gate_proj",
84
+ "language_model.model.layers.27.self_attn.o_proj",
85
+ "language_model.model.layers.27.self_attn.v_proj",
86
+ "language_model.model.layers.27.self_attn.k_proj",
87
+ "language_model.model.layers.27.self_attn.q_proj",
88
+ "language_model.model.layers.26.mlp.shared_expert.up_proj",
89
+ "language_model.model.layers.26.mlp.shared_expert.down_proj",
90
+ "language_model.model.layers.26.mlp.shared_expert.gate_proj",
91
+ "language_model.model.layers.26.linear_attn.out_proj",
92
+ "language_model.model.layers.26.linear_attn.in_proj_z",
93
+ "language_model.model.layers.26.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.25.mlp.shared_expert.up_proj",
95
+ "language_model.model.layers.25.mlp.shared_expert.down_proj",
96
+ "language_model.model.layers.25.mlp.shared_expert.gate_proj",
97
+ "language_model.model.layers.25.linear_attn.out_proj",
98
+ "language_model.model.layers.25.linear_attn.in_proj_z",
99
+ "language_model.model.layers.25.linear_attn.in_proj_qkv",
100
+ "language_model.model.layers.24.mlp.shared_expert.up_proj",
101
+ "language_model.model.layers.24.mlp.shared_expert.down_proj",
102
+ "language_model.model.layers.24.mlp.shared_expert.gate_proj",
103
+ "language_model.model.layers.24.linear_attn.out_proj",
104
+ "language_model.model.layers.24.linear_attn.in_proj_z",
105
+ "language_model.model.layers.24.linear_attn.in_proj_qkv"
106
+ ],
107
+ "rank": 16,
108
+ "scale": 2.0
109
+ },
110
+ "num_layers": -1
111
+ }
training/adapters/r2-step120/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87dd5bdb3d518a34063a40ebbf15fa1a438412c3d4b04f8462f394e4370a34c4
3
+ size 30697232
training/adapters/r3-final/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "lora_parameters": {
4
+ "dropout": 0.0,
5
+ "keys": [
6
+ "language_model.model.layers.39.mlp.shared_expert.up_proj",
7
+ "language_model.model.layers.39.mlp.shared_expert.down_proj",
8
+ "language_model.model.layers.39.mlp.shared_expert.gate_proj",
9
+ "language_model.model.layers.39.self_attn.o_proj",
10
+ "language_model.model.layers.39.self_attn.v_proj",
11
+ "language_model.model.layers.39.self_attn.k_proj",
12
+ "language_model.model.layers.39.self_attn.q_proj",
13
+ "language_model.model.layers.38.mlp.shared_expert.up_proj",
14
+ "language_model.model.layers.38.mlp.shared_expert.down_proj",
15
+ "language_model.model.layers.38.mlp.shared_expert.gate_proj",
16
+ "language_model.model.layers.38.linear_attn.out_proj",
17
+ "language_model.model.layers.38.linear_attn.in_proj_z",
18
+ "language_model.model.layers.38.linear_attn.in_proj_qkv",
19
+ "language_model.model.layers.37.mlp.shared_expert.up_proj",
20
+ "language_model.model.layers.37.mlp.shared_expert.down_proj",
21
+ "language_model.model.layers.37.mlp.shared_expert.gate_proj",
22
+ "language_model.model.layers.37.linear_attn.out_proj",
23
+ "language_model.model.layers.37.linear_attn.in_proj_z",
24
+ "language_model.model.layers.37.linear_attn.in_proj_qkv",
25
+ "language_model.model.layers.36.mlp.shared_expert.up_proj",
26
+ "language_model.model.layers.36.mlp.shared_expert.down_proj",
27
+ "language_model.model.layers.36.mlp.shared_expert.gate_proj",
28
+ "language_model.model.layers.36.linear_attn.out_proj",
29
+ "language_model.model.layers.36.linear_attn.in_proj_z",
30
+ "language_model.model.layers.36.linear_attn.in_proj_qkv",
31
+ "language_model.model.layers.35.mlp.shared_expert.up_proj",
32
+ "language_model.model.layers.35.mlp.shared_expert.down_proj",
33
+ "language_model.model.layers.35.mlp.shared_expert.gate_proj",
34
+ "language_model.model.layers.35.self_attn.o_proj",
35
+ "language_model.model.layers.35.self_attn.v_proj",
36
+ "language_model.model.layers.35.self_attn.k_proj",
37
+ "language_model.model.layers.35.self_attn.q_proj",
38
+ "language_model.model.layers.34.mlp.shared_expert.up_proj",
39
+ "language_model.model.layers.34.mlp.shared_expert.down_proj",
40
+ "language_model.model.layers.34.mlp.shared_expert.gate_proj",
41
+ "language_model.model.layers.34.linear_attn.out_proj",
42
+ "language_model.model.layers.34.linear_attn.in_proj_z",
43
+ "language_model.model.layers.34.linear_attn.in_proj_qkv",
44
+ "language_model.model.layers.33.mlp.shared_expert.up_proj",
45
+ "language_model.model.layers.33.mlp.shared_expert.down_proj",
46
+ "language_model.model.layers.33.mlp.shared_expert.gate_proj",
47
+ "language_model.model.layers.33.linear_attn.out_proj",
48
+ "language_model.model.layers.33.linear_attn.in_proj_z",
49
+ "language_model.model.layers.33.linear_attn.in_proj_qkv",
50
+ "language_model.model.layers.32.mlp.shared_expert.up_proj",
51
+ "language_model.model.layers.32.mlp.shared_expert.down_proj",
52
+ "language_model.model.layers.32.mlp.shared_expert.gate_proj",
53
+ "language_model.model.layers.32.linear_attn.out_proj",
54
+ "language_model.model.layers.32.linear_attn.in_proj_z",
55
+ "language_model.model.layers.32.linear_attn.in_proj_qkv",
56
+ "language_model.model.layers.31.mlp.shared_expert.up_proj",
57
+ "language_model.model.layers.31.mlp.shared_expert.down_proj",
58
+ "language_model.model.layers.31.mlp.shared_expert.gate_proj",
59
+ "language_model.model.layers.31.self_attn.o_proj",
60
+ "language_model.model.layers.31.self_attn.v_proj",
61
+ "language_model.model.layers.31.self_attn.k_proj",
62
+ "language_model.model.layers.31.self_attn.q_proj",
63
+ "language_model.model.layers.30.mlp.shared_expert.up_proj",
64
+ "language_model.model.layers.30.mlp.shared_expert.down_proj",
65
+ "language_model.model.layers.30.mlp.shared_expert.gate_proj",
66
+ "language_model.model.layers.30.linear_attn.out_proj",
67
+ "language_model.model.layers.30.linear_attn.in_proj_z",
68
+ "language_model.model.layers.30.linear_attn.in_proj_qkv",
69
+ "language_model.model.layers.29.mlp.shared_expert.up_proj",
70
+ "language_model.model.layers.29.mlp.shared_expert.down_proj",
71
+ "language_model.model.layers.29.mlp.shared_expert.gate_proj",
72
+ "language_model.model.layers.29.linear_attn.out_proj",
73
+ "language_model.model.layers.29.linear_attn.in_proj_z",
74
+ "language_model.model.layers.29.linear_attn.in_proj_qkv",
75
+ "language_model.model.layers.28.mlp.shared_expert.up_proj",
76
+ "language_model.model.layers.28.mlp.shared_expert.down_proj",
77
+ "language_model.model.layers.28.mlp.shared_expert.gate_proj",
78
+ "language_model.model.layers.28.linear_attn.out_proj",
79
+ "language_model.model.layers.28.linear_attn.in_proj_z",
80
+ "language_model.model.layers.28.linear_attn.in_proj_qkv",
81
+ "language_model.model.layers.27.mlp.shared_expert.up_proj",
82
+ "language_model.model.layers.27.mlp.shared_expert.down_proj",
83
+ "language_model.model.layers.27.mlp.shared_expert.gate_proj",
84
+ "language_model.model.layers.27.self_attn.o_proj",
85
+ "language_model.model.layers.27.self_attn.v_proj",
86
+ "language_model.model.layers.27.self_attn.k_proj",
87
+ "language_model.model.layers.27.self_attn.q_proj",
88
+ "language_model.model.layers.26.mlp.shared_expert.up_proj",
89
+ "language_model.model.layers.26.mlp.shared_expert.down_proj",
90
+ "language_model.model.layers.26.mlp.shared_expert.gate_proj",
91
+ "language_model.model.layers.26.linear_attn.out_proj",
92
+ "language_model.model.layers.26.linear_attn.in_proj_z",
93
+ "language_model.model.layers.26.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.25.mlp.shared_expert.up_proj",
95
+ "language_model.model.layers.25.mlp.shared_expert.down_proj",
96
+ "language_model.model.layers.25.mlp.shared_expert.gate_proj",
97
+ "language_model.model.layers.25.linear_attn.out_proj",
98
+ "language_model.model.layers.25.linear_attn.in_proj_z",
99
+ "language_model.model.layers.25.linear_attn.in_proj_qkv",
100
+ "language_model.model.layers.24.mlp.shared_expert.up_proj",
101
+ "language_model.model.layers.24.mlp.shared_expert.down_proj",
102
+ "language_model.model.layers.24.mlp.shared_expert.gate_proj",
103
+ "language_model.model.layers.24.linear_attn.out_proj",
104
+ "language_model.model.layers.24.linear_attn.in_proj_z",
105
+ "language_model.model.layers.24.linear_attn.in_proj_qkv"
106
+ ],
107
+ "rank": 16,
108
+ "scale": 2.0
109
+ },
110
+ "num_layers": -1
111
+ }
training/adapters/r3-final/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0dbbf7c6907cecd6b82679eb8675599e28580760143d8a9d84b9ada834c5ac91
3
+ size 30697232