Update model files

config.json

@@ -45,20 +45,19 @@
           "dynamic": false,
           "group_size": 128,
           "num_bits": 4,
-          "observer": "memoryless_minmax",
+          "observer": "minmax",
           "observer_kwargs": {},
+          "scale_dtype": null,
           "strategy": "group",
           "symmetric": true,
-          "type": "int"
+          "type": "int",
+          "zp_dtype": null
         }
       }
     },
     "format": "pack-quantized",
     "global_compression_ratio": null,
     "ignore": [
-      "model.layers.0.mlp.gate_proj",
-      "model.layers.1.mlp.gate_proj",
-      "model.layers.2.mlp.gate_proj",
       "lm_head"
     ],
     "kv_cache_scheme": null,
@@ -66,7 +65,7 @@
     "quantization_status": "compressed",
     "sparsity_config": {},
     "transform_config": {},
-    "version": "0.12.3.dev31+gb019b89"
+    "version": "0.12.3.dev31+gb019b89.d20251216"
   },
   "rms_norm_eps": 1e-05,
   "rope_scaling": null,
@@ -78,4 +77,4 @@
   "use_cache": true,
   "use_qk_norm": true,
   "vocab_size": 151552
-}
+}

model-00001-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:571c2a3c42c9503a0605989b1506c30a76be219d501677fe734e3abe662a1c89
-size 4999812360
+oid sha256:e7b2bcc69816d48bdf35fc64f0355b5104e0039d2178f61fea8989d81904e08b
+size 4999471048

model-00002-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01d5440163ab4d8ad9ac2cc50e42e1a6782a6266d8e61a6514f07836eb5921c0
-size 4998300040
+oid sha256:46321ac0ede036ce4363703692b38cf7a1164c0fecd1f1ea41bd8b110b7f7035
+size 4998300240

model-00003-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfb5ef078eb4ba31e33d8f1bce57a473b7e10b2d77b9d6f3e1dbe629535e14f1
-size 4997280968
+oid sha256:9fcce1e7dc93e4ddf79e0684b6d94257e7fee546b13d33c0f96d83621461b9cc
+size 4997280760

model-00004-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:474d1b0da073b1c3dc5a8afda94fcf5acb978c3a4157d2168994b5de030337fe
-size 4998302400
+oid sha256:347052e23801adabcaaadf48c808804b5c868f72e0f776887ed1a2f23418b343
+size 4998302816

model-00005-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:355fa0c5019d27309fae1b99cb648e6331ac8aa94582af1e04d8e8fa3dfb359d
-size 4997284688
+oid sha256:dbaab130517ff7b7fedd42af62fd6cdd6b88739294969eb99691a019c1c0e262
+size 4997284480

model-00006-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a74675b352a254c1b0d090e3105daff3eb4080ceb93ef8ab60e66d0e8012ddc8
-size 4998303456
+oid sha256:0b4624f63de0ac57136e51790888142cf0f18cbcfca2bd390c24dff8fe16e0ec
+size 4998303640

model-00007-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a70c256c38fcbe827bebc5b5df3e739c287b6cefe5c20a7ea2fa697f52dd97f9
-size 4997284808
+oid sha256:41e511f237b34f919c8d8c1e75698f5c0317609ce80f4d03efd5cec05e436d68
+size 4997284600

model-00008-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf92f3f4e013b7230e69efa3e7f7358a9513fc0c2fad498cfbb3b498dcfc299d
-size 4998303432
+oid sha256:bee8d3202c0b6d75ae927c72bee9c02174a0647048777d0635373b0162000c09
+size 4998303528

model-00009-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43e401afa06767167f70db152ff801c5739bf74b329208b5ce3beb9e9267a5e4
-size 4995908568
+oid sha256:f5ab9b2860d51dfee53cf3f424cf16ca04ea25d72147fa2c881121ac3074d3db
+size 4997284712

model-00010-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d1eef0a3c3f974d1049dae46b451eb8b2ef81e0add3ab44b455cbd929ad3636
-size 4999679672
+oid sha256:ff44f85b410734972e8d41daf537f8753a406d4bf9b67293240e1497fa659e1d
+size 4998303456

model-00011-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3a9408199fd30b27608ec1640b48e42a1667221baf98e5c183f666aa1cc56ae
-size 4998303896
+oid sha256:b9faac7e2992e1f17e4e573960387ffd4dd0a50a4fceb3087c8c13b5feea549e
+size 4997284832

model-00012-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:546652a6813605fd624d637be443a5da78fef507a7229a6ecf15086fb46b6995
-size 4997284344
+oid sha256:6a60f99d2c26c6d7e0db0a87286b6114802e4210621dc12228c989a0246924db
+size 4998303408

model-00013-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7847695810a26312e1e3459a54711212d5a5f40529a25736cb7c58159466462d
-size 4998303776
+oid sha256:1cbebf09bc507e43395cb106bccec978ea6c57801071d80b440406815d97f69a
+size 4998303984

model-00014-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fd1142de8c24ddb6618a37243938c009fdd05de287cb74f34714d44f12a647a
-size 4997284464
+oid sha256:7b62048e5ffc061a30e255a236501d6b0764c6c3b7020572991d1b938e059f20
+size 4997284256

model-00015-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3645ee2c6f1cb9f1dd11cd1adf66489ace0b88cb3d4adcb6932d12a3ada3876
-size 4998303664
+oid sha256:1ffc9ad660a751c10bb1e5700193b93f90d962e0d127c3a6daf233085f6f65ab
+size 4998303864

model-00016-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:867539a161b188874e51383f7afe49352b394742c8af3bb2bc1e87970b3280b6
-size 4997284576
+oid sha256:7c35043bff000286afaa7cdfd607a8739f9ce53340d843e532b7ec85aafc1109
+size 4997284376

model-00017-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a399bb4014a31a8e69b8c866fa64e3adc7e98997a1f2408796345f8cbf7d1923
-size 4998303544
+oid sha256:0a9a820ac78fdde795c4378f4a43032e02e2beebd32a8f859318c418fc9a0219
+size 4998303752

model-00018-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:834368cb2f08ce6eb7cacc5abbe127f23429141ffe85609e3decf5e418dfb322
-size 4997284696
+oid sha256:d7a918bd5c40b5a130fafe54347f8b9d26151a960dc1ed542175ad288f4c4cc1
+size 4997284488

model-00019-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7928dfa2f359a73df7c971fcb754e6673c94bd559463ce161c84ed84fbbf6845
-size 4998303456
+oid sha256:8aa286e5975611d52cd58fd0ade3eb2dc44c093815c858191fe7e60751896055
+size 4998303632

model-00020-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:854400debbbdb7a147f62ba9107123d95b5f7e494a05077d76a0ab02178d1c5b
-size 4997284816
+oid sha256:d4379ba66ab0bd562e37599a3213a6184eb6e31b43de32267e042925f5d4ecab
+size 4997284608

model-00021-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e476adbebdf2d87e57e2bcaa2be6d267983849ac199da08eb23057543429dc4
-size 4998303424
+oid sha256:e279032f627c1646cf4ff24bf6f4ebdacbc3c147663a6e592db52a448e48b423
+size 4998303520

model-00022-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a466bda0c478459d335efbdc318ce77f69a54ab9f5d3f9ec964d028b37a8efe4
-size 4999943272
+oid sha256:dcb2941674fbd161c14f510e37a85cf8ce030d7a6088294c9d6e17e39756c682
+size 4997284720

model-00023-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f90f71d57fe3e594567fd2ecd24ad4f0aea06b09477f08f30b4a5d11a8fa154
-size 4999700392
+oid sha256:485dc1746278e5b7aaa9ddec6807aec2261c2838c1d0f2331252879a8ba57fa4
+size 4998303456

model-00024-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7436cbd95917b91dd9691de90e27231f8727bad8e34990bbe7dabba4bc9d65e6
-size 4998303888
+oid sha256:6a3f9d5e2419c2d569b65ed0e6c648a3c28a237ec86ba4360b054dbcef045bcb
+size 4997284840

model-00025-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2294304956dd3ddb9852d43b36e7c1a63e1896cbc6f4cd99b2a0b98d6c27d843
-size 4997284352
+oid sha256:08d77fe6861ba4226da422d61bdf27d9dcc3b77acaf3592721dcbabadc6fe008
+size 4998303400

model-00026-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42411f4ddb6602e03a69a9c6420cc1c87d714a658b542362d8b6f4161c4b40e0
-size 4998303768
+oid sha256:c9773b3cf0311d3b645412370e26df556a5811c2c98395b88bee587926e8b8cb
+size 4998303976

model-00027-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f00474353e92b974f23e11e5c4ac2c2dc8ed7688e5c7a0386f4e1c6c51f7e756
-size 4997284472
+oid sha256:45b4b57734d3bfaa2006c7f591365312a92deac3703b11c486545702b09a3840
+size 4997284264

model-00028-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec8a1a352ea2a43334fa0ce54d7027a3d037149e499f0be7ced35984ea8919de
-size 4998303656
+oid sha256:02ab6a0fe1da903f69ea8c51e8657560036eabf2b22c07293d03696f83b6f815
+size 4998303856

model-00029-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7be3bd1f796195b160e7ed0221e719ab18c551767272457e9dea50999382ceaf
-size 4997284584
+oid sha256:25e9cd3dc3feb967804ac140f6a9062e675326dd4e855a13e6116a018a30f721
+size 4997284384

model-00030-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6bced91cfe11707cfae926dc3c6edc7d8a7d7f0b56bc558fc6153e284b82c56
-size 4998303536
+oid sha256:ca401ce3e94a62ec419536857c3a0dd2a22d6fe50ff8ed02d9406e2df87af07b
+size 4998303744

model-00031-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae344c69386cf0da041fdbe5d9834a3d87598a2325073579e971dc8de80829e1
-size 4997284704
+oid sha256:8e9cc10ccc62c247a84d544a204d5a6a081afab2b57c310a34ee315224828f26
+size 4997284496

model-00032-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb40fa4e4196a526885964e63ac4b9d394a8f897d9a666ef6a78dedea24c7783
-size 4998303456
+oid sha256:72f8455a39b70d36c9a9a8f5aa700b6336af558bf08494870031bd5fb0504732
+size 4998303624

model-00033-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:716ff16eb5380e32462d46443347e4cb6009ae05dd6727dbae08cd91ab61bb9e
-size 4997284816
+oid sha256:ec641831537c5b883d06c4c6e42b32104d0ff96c385b30791ae8212d6fd4f2f4
+size 4997284616

model-00034-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83c292fb8cb9ad72c3ab8c1632bc345b9416c734f2e968c88098effbf2b45c8b
-size 4998303424
+oid sha256:2add5e53c6269a7e04d5918faf90ae7729e8e441948c2d653abd4590b7ab83c2
+size 4998303504

model-00035-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6306113b571461b05fab902cf20741500b3152e17e3417b48cdd70bb2159bb60
-size 4998304000
+oid sha256:cb3338fa36ad496b86950f6109c22d04b308845f5d75f8fd40fdc9ebd3de3456
+size 4997284736

model-00036-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4fadbc86aaff1ff97233801d4e2a797e19e2c5ea91813e14263df293547753c9
-size 4997284240
+oid sha256:251a1db220cead58fa9f7efe15ead5a8e22f1a6063446d3d03d4d63186038d9a
+size 4998303456

model-00037-of-00037.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ccdb2ca0459c2e5adaab403ea1b5bf96ad17a35233472e1fecfb844713be155
-size 4695499440
+oid sha256:1216c4476e72be10cf7313494b311053789133c54afbe2835d47c7931548aae8
+size 4419730336

model.safetensors.index.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3113058954ddad3db3bae8114a4b966364c8c5cc5cee20a335a552fd5812070
-size 12668225
+oid sha256:e82645b7359e97d49c9a0918a4d765dee570aafb41168b1a8aedeaf9b1db898f
+size 12668756

recipe.yaml

@@ -1,28 +1,9 @@
 default_stage:
   default_modifiers:
     GPTQModifier:
-      config_groups:
-        group_0:
-          targets: [Linear]
-          weights:
-            num_bits: 4
-            type: int
-            symmetric: true
-            group_size: 128
-            strategy: group
-            block_structure: null
-            dynamic: false
-            actorder: !!python/object/apply:compressed_tensors.quantization.quant_args.ActivationOrdering [
-              static]
-            scale_dtype: null
-            zp_dtype: null
-            observer: memoryless_minmax
-            observer_kwargs: {}
-          input_activations: null
-          output_activations: null
-          format: null
       targets: [Linear]
-      ignore: [lm_head, 're:.*mlp.gate.*']
+      ignore: [lm_head]
+      scheme: W4A16
       block_size: 128
       dampening_frac: 0.01
       actorder: static