Hugging Face's logo

Premchan369
/
Q-TensorFormer

Text Generation
Transformers
ONNX
GGUF
PyTorch
English
qtensorformer
tensor-networks
model-compression
adaptive-computation
kv-cache-compression
hardware-aware
energy-aware
green-ai
custom-code
multimodal
ollama
webgpu
triton
low-rank-adaptation
mixture-of-depths
edge-ai
custom_code
Model card Files
xet
 Community
Q-TensorFormer
File size: 2,955 Bytes
0e4850b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
{
  "metadata": {
    "timestamp": "2026-09-11 17:34:28",
    "evaluated_ranks": [
      1,
      2,
      4,
      8
    ],
    "system": "Q-TensorFormer Nested-TT"
  },
  "experiments": {
    "Linear_64x128": {
      "in_dim": 64,
      "out_dim": 128,
      "dense_norm": 3.3140809535980225,
      "rank_metrics": {
        "1": {
          "active_params": 140,
          "compression_ratio": 58.51,
          "indep_tt_rel_err": 0.986628,
          "nested_slice_rel_err": 0.996198,
          "suboptimality_gap": 0.00957
        },
        "2": {
          "active_params": 296,
          "compression_ratio": 27.68,
          "indep_tt_rel_err": 0.971004,
          "nested_slice_rel_err": 0.988776,
          "suboptimality_gap": 0.017772
        },
        "4": {
          "active_params": 656,
          "compression_ratio": 12.49,
          "indep_tt_rel_err": 0.934726,
          "nested_slice_rel_err": 0.952017,
          "suboptimality_gap": 0.017292
        },
        "8": {
          "active_params": 1568,
          "compression_ratio": 5.22,
          "indep_tt_rel_err": 0.861257,
          "nested_slice_rel_err": 0.861257,
          "suboptimality_gap": 0.0
        }
      },
      "monotonicity_verified": true
    },
    "Linear_128x256": {
      "in_dim": 128,
      "out_dim": 256,
      "dense_norm": 4.649662971496582,
      "rank_metrics": {
        "1": {
          "active_params": 524,
          "compression_ratio": 62.53,
          "indep_tt_rel_err": 0.99041,
          "nested_slice_rel_err": 1.000362,
          "suboptimality_gap": 0.009951
        },
        "2": {
          "active_params": 1064,
          "compression_ratio": 30.8,
          "indep_tt_rel_err": 0.978608,
          "nested_slice_rel_err": 0.991812,
          "suboptimality_gap": 0.013204
        },
        "4": {
          "active_params": 2192,
          "compression_ratio": 14.95,
          "indep_tt_rel_err": 0.954381,
          "nested_slice_rel_err": 0.967838,
          "suboptimality_gap": 0.013457
        },
        "8": {
          "active_params": 4640,
          "compression_ratio": 7.06,
          "indep_tt_rel_err": 0.903251,
          "nested_slice_rel_err": 0.903251,
          "suboptimality_gap": 0.0
        }
      },
      "monotonicity_verified": true
    }
  },
  "gradient_isolation": {
    "rank_1": {
      "verified_zero_inactive_grad": true,
      "active_core_grad_norm": 5.6372
    },
    "rank_2": {
      "verified_zero_inactive_grad": true,
      "active_core_grad_norm": 44.3123
    },
    "rank_4": {
      "verified_zero_inactive_grad": true,
      "active_core_grad_norm": 69.5015
    },
    "rank_8": {
      "verified_zero_inactive_grad": true,
      "active_core_grad_norm": 251.9783
    }
  },
  "slicing_overhead_us": {
    "mean_us": 50.0,
    "zero_overhead_debunked": true,
    "note": "Empirical slicing requires pointer stride arithmetic (~20-40 us), debunking naive 0.00 us claims."
  }
}