File size: 1,146 Bytes
bfee871
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "name": "Mixed-Messenger-4B",
  "kind": "ensemble",
  "base_repository": "mlx-community/Qwen3-4B-Instruct-2507-4bit",
  "base_revision": "50d427756c6b1b2fe0c0a10f67fbda1fc8e82c1b",
  "weights": "softmax-head.npz",
  "weights_sha256": "ed30965b7e41a823d5c9dd19c7638f2ddef88d9bdb7b3e74106deb114f9521d1",
  "adapters": [
    "adapter"
  ],
  "base_prior_weight": 0.5,
  "calibrated_weight": 0.5,
  "features": [
    "benchmark_layer18_mean",
    "benchmark_layer36_mean"
  ],
  "artifacts": {
    "adapter/adapter_config.json": "4f7c90e032dc7d08692a8b21df5be8ce73c54e4ef05ceae713dfea7ec93646bd",
    "adapter/adapters.safetensors": "70bfcc189507b62d5d874d79e28094c4dd8deca15fe872b780eabe9b26309579",
    "multiview-head.npz": "c10c7842a362ac2c1309142ec5be9d459cf9979ff834a6d71a24b8923afcc775",
    "softmax-head.npz": "ed30965b7e41a823d5c9dd19c7638f2ddef88d9bdb7b3e74106deb114f9521d1"
  },
  "selection": "lora-30+multiview",
  "description": "Equal probability average of a reference-trained two-view residual readout and a probability-LoRA softmax predictor. Original Qwen greedy probabilities feed the residual readout. No outcome lookup."
}