mlx-community
/

Kimi-K2-Thinking-4bit

Text Generation

4-bit precision

Model card Files Files and versions

awni commited on Nov 6, 2025

Commit

e7ddd1c

·

verified ·

1 Parent(s): a5f40d5

Update config.json

Files changed (1) hide show

config.json +2 -2

config.json CHANGED Viewed

@@ -76,12 +76,12 @@
     "qk_nope_head_dim": 128,
     "qk_rope_head_dim": 64,
     "quantization": {
-        "group_size": 64,
         "bits": 4,
         "mode": "affine"
     },
     "quantization_config": {
-        "group_size": 64,
         "bits": 4,
         "mode": "affine"
     },

     "qk_nope_head_dim": 128,
     "qk_rope_head_dim": 64,
     "quantization": {
+        "group_size": 32,
         "bits": 4,
         "mode": "affine"
     },
     "quantization_config": {
+        "group_size": 32,
         "bits": 4,
         "mode": "affine"
     },