nm-autobot commited on
Commit
a5454f8
·
verified ·
1 Parent(s): 457e3d5

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. config.json +5 -19
config.json CHANGED
@@ -60,18 +60,14 @@
60
  "config_group_0": {
61
  "format": "nvfp4-pack-quantized",
62
  "input_activations": {
63
- "actorder": null,
64
- "block_structure": null,
65
  "dynamic": "local",
66
  "group_size": 16,
67
  "num_bits": 4,
68
  "observer": "static_minmax",
69
- "observer_kwargs": {},
70
  "scale_dtype": "torch.float8_e4m3fn",
71
  "strategy": "tensor_group",
72
  "symmetric": true,
73
- "type": "float",
74
- "zp_dtype": null
75
  },
76
  "output_activations": null,
77
  "targets": [
@@ -79,18 +75,13 @@
79
  "re:.*self_attn.*\\.(q|k|v|o)_proj$"
80
  ],
81
  "weights": {
82
- "actorder": null,
83
- "block_structure": null,
84
  "dynamic": false,
85
  "group_size": 16,
86
  "num_bits": 4,
87
- "observer": null,
88
- "observer_kwargs": {},
89
  "scale_dtype": "torch.float8_e4m3fn",
90
  "strategy": "tensor_group",
91
  "symmetric": true,
92
- "type": "float",
93
- "zp_dtype": null
94
  }
95
  }
96
  },
@@ -98,22 +89,17 @@
98
  "global_compression_ratio": null,
99
  "ignore": [],
100
  "kv_cache_scheme": {
101
- "actorder": null,
102
- "block_structure": null,
103
  "dynamic": false,
104
  "group_size": null,
105
  "num_bits": 8,
106
- "observer": null,
107
- "observer_kwargs": {},
108
- "scale_dtype": null,
109
  "strategy": "tensor",
110
  "symmetric": true,
111
- "type": "float",
112
- "zp_dtype": null
113
  },
 
114
  "quant_method": "compressed-tensors",
115
  "quantization_status": "compressed",
116
- "version": "0.19.1a20261014"
117
  },
118
  "rms_norm_eps": 1e-06,
119
  "rope_scaling": null,
 
60
  "config_group_0": {
61
  "format": "nvfp4-pack-quantized",
62
  "input_activations": {
 
 
63
  "dynamic": "local",
64
  "group_size": 16,
65
  "num_bits": 4,
66
  "observer": "static_minmax",
 
67
  "scale_dtype": "torch.float8_e4m3fn",
68
  "strategy": "tensor_group",
69
  "symmetric": true,
70
+ "type": "float"
 
71
  },
72
  "output_activations": null,
73
  "targets": [
 
75
  "re:.*self_attn.*\\.(q|k|v|o)_proj$"
76
  ],
77
  "weights": {
 
 
78
  "dynamic": false,
79
  "group_size": 16,
80
  "num_bits": 4,
 
 
81
  "scale_dtype": "torch.float8_e4m3fn",
82
  "strategy": "tensor_group",
83
  "symmetric": true,
84
+ "type": "float"
 
85
  }
86
  }
87
  },
 
89
  "global_compression_ratio": null,
90
  "ignore": [],
91
  "kv_cache_scheme": {
 
 
92
  "dynamic": false,
93
  "group_size": null,
94
  "num_bits": 8,
 
 
 
95
  "strategy": "tensor",
96
  "symmetric": true,
97
+ "type": "float"
 
98
  },
99
+ "moe_sparsity_config": null,
100
  "quant_method": "compressed-tensors",
101
  "quantization_status": "compressed",
102
+ "version": "0.19.1a20261015"
103
  },
104
  "rms_norm_eps": 1e-06,
105
  "rope_scaling": null,