Cialtion commited on
Commit
0d1a221
·
verified ·
1 Parent(s): ec21246

Update RT-Qwen3-4B-AWQ with latest interleave_focal checkpoint

Browse files
RT-Qwen3-4B-AWQ/config.json CHANGED
@@ -75,9 +75,9 @@
75
  "observer_kwargs": {},
76
  "scale_dtype": null,
77
  "strategy": "group",
78
- "symmetric": true,
79
  "type": "int",
80
- "zp_dtype": null
81
  }
82
  }
83
  },
 
75
  "observer_kwargs": {},
76
  "scale_dtype": null,
77
  "strategy": "group",
78
+ "symmetric": false,
79
  "type": "int",
80
+ "zp_dtype": "torch.int8"
81
  }
82
  }
83
  },
RT-Qwen3-4B-AWQ/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e00cef750d0332c56eea1ad8c98fb1f88c83cc8db4dcfc7325894113f121f134
3
- size 3427191984
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b04e9724acc9898f19eaf88b9c934689e7eacf4c9df569af62815911e4a6532
3
+ size 3441415440
RT-Qwen3-4B-AWQ/recipe.yaml CHANGED
@@ -3,7 +3,7 @@ default_stage:
3
  AWQModifier:
4
  targets: [Linear]
5
  ignore: [lm_head]
6
- scheme: W4A16
7
  mappings:
8
  - smooth_layer: re:.*input_layernorm
9
  balance_layers: ['re:.*q_proj', 're:.*k_proj', 're:.*v_proj']
 
3
  AWQModifier:
4
  targets: [Linear]
5
  ignore: [lm_head]
6
+ scheme: W4A16_ASYM
7
  mappings:
8
  - smooth_layer: re:.*input_layernorm
9
  balance_layers: ['re:.*q_proj', 're:.*k_proj', 're:.*v_proj']