Upload folder using huggingface_hub

config.json

@@ -2,6 +2,7 @@
   "architectures": [
     "Glm4vMoeForConditionalGeneration"
   ],
+  "dtype": "bfloat16",
   "image_end_token_id": 151340,
   "image_start_token_id": 151339,
   "image_token_id": 151363,
@@ -9,6 +10,7 @@
   "quantization_config": {
     "config_groups": {
       "group_0": {
+        "format": "pack-quantized",
         "input_activations": null,
         "output_activations": null,
         "targets": [
@@ -20,7 +22,7 @@
           "dynamic": false,
           "group_size": 32,
           "num_bits": 4,
-          "observer": "minmax",
+          "observer": "mse",
           "observer_kwargs": {},
           "strategy": "group",
           "symmetric": true,
@@ -155,6 +157,148 @@
       "model.visual.merger.gate_proj",
       "model.visual.merger.up_proj",
       "model.visual.merger.down_proj",
+      "model.language_model.layers.0.self_attn.q_proj",
+      "model.language_model.layers.0.self_attn.k_proj",
+      "model.language_model.layers.0.self_attn.v_proj",
+      "model.language_model.layers.0.self_attn.o_proj",
+      "model.language_model.layers.0.mlp.gate_proj",
+      "model.language_model.layers.0.mlp.up_proj",
+      "model.language_model.layers.0.mlp.down_proj",
+      "model.language_model.layers.1.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.1.mlp.shared_experts.up_proj",
+      "model.language_model.layers.1.mlp.shared_experts.down_proj",
+      "model.language_model.layers.2.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.2.mlp.shared_experts.up_proj",
+      "model.language_model.layers.2.mlp.shared_experts.down_proj",
+      "model.language_model.layers.3.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.3.mlp.shared_experts.up_proj",
+      "model.language_model.layers.3.mlp.shared_experts.down_proj",
+      "model.language_model.layers.4.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.4.mlp.shared_experts.up_proj",
+      "model.language_model.layers.4.mlp.shared_experts.down_proj",
+      "model.language_model.layers.5.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.5.mlp.shared_experts.up_proj",
+      "model.language_model.layers.5.mlp.shared_experts.down_proj",
+      "model.language_model.layers.6.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.6.mlp.shared_experts.up_proj",
+      "model.language_model.layers.6.mlp.shared_experts.down_proj",
+      "model.language_model.layers.7.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.7.mlp.shared_experts.up_proj",
+      "model.language_model.layers.7.mlp.shared_experts.down_proj",
+      "model.language_model.layers.8.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.8.mlp.shared_experts.up_proj",
+      "model.language_model.layers.8.mlp.shared_experts.down_proj",
+      "model.language_model.layers.9.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.9.mlp.shared_experts.up_proj",
+      "model.language_model.layers.9.mlp.shared_experts.down_proj",
+      "model.language_model.layers.10.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.10.mlp.shared_experts.up_proj",
+      "model.language_model.layers.10.mlp.shared_experts.down_proj",
+      "model.language_model.layers.11.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.11.mlp.shared_experts.up_proj",
+      "model.language_model.layers.11.mlp.shared_experts.down_proj",
+      "model.language_model.layers.12.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.12.mlp.shared_experts.up_proj",
+      "model.language_model.layers.12.mlp.shared_experts.down_proj",
+      "model.language_model.layers.13.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.13.mlp.shared_experts.up_proj",
+      "model.language_model.layers.13.mlp.shared_experts.down_proj",
+      "model.language_model.layers.14.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.14.mlp.shared_experts.up_proj",
+      "model.language_model.layers.14.mlp.shared_experts.down_proj",
+      "model.language_model.layers.15.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.15.mlp.shared_experts.up_proj",
+      "model.language_model.layers.15.mlp.shared_experts.down_proj",
+      "model.language_model.layers.16.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.16.mlp.shared_experts.up_proj",
+      "model.language_model.layers.16.mlp.shared_experts.down_proj",
+      "model.language_model.layers.17.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.17.mlp.shared_experts.up_proj",
+      "model.language_model.layers.17.mlp.shared_experts.down_proj",
+      "model.language_model.layers.18.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.18.mlp.shared_experts.up_proj",
+      "model.language_model.layers.18.mlp.shared_experts.down_proj",
+      "model.language_model.layers.19.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.19.mlp.shared_experts.up_proj",
+      "model.language_model.layers.19.mlp.shared_experts.down_proj",
+      "model.language_model.layers.20.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.20.mlp.shared_experts.up_proj",
+      "model.language_model.layers.20.mlp.shared_experts.down_proj",
+      "model.language_model.layers.21.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.21.mlp.shared_experts.up_proj",
+      "model.language_model.layers.21.mlp.shared_experts.down_proj",
+      "model.language_model.layers.22.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.22.mlp.shared_experts.up_proj",
+      "model.language_model.layers.22.mlp.shared_experts.down_proj",
+      "model.language_model.layers.23.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.23.mlp.shared_experts.up_proj",
+      "model.language_model.layers.23.mlp.shared_experts.down_proj",
+      "model.language_model.layers.24.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.24.mlp.shared_experts.up_proj",
+      "model.language_model.layers.24.mlp.shared_experts.down_proj",
+      "model.language_model.layers.25.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.25.mlp.shared_experts.up_proj",
+      "model.language_model.layers.25.mlp.shared_experts.down_proj",
+      "model.language_model.layers.26.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.26.mlp.shared_experts.up_proj",
+      "model.language_model.layers.26.mlp.shared_experts.down_proj",
+      "model.language_model.layers.27.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.27.mlp.shared_experts.up_proj",
+      "model.language_model.layers.27.mlp.shared_experts.down_proj",
+      "model.language_model.layers.28.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.28.mlp.shared_experts.up_proj",
+      "model.language_model.layers.28.mlp.shared_experts.down_proj",
+      "model.language_model.layers.29.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.29.mlp.shared_experts.up_proj",
+      "model.language_model.layers.29.mlp.shared_experts.down_proj",
+      "model.language_model.layers.30.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.30.mlp.shared_experts.up_proj",
+      "model.language_model.layers.30.mlp.shared_experts.down_proj",
+      "model.language_model.layers.31.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.31.mlp.shared_experts.up_proj",
+      "model.language_model.layers.31.mlp.shared_experts.down_proj",
+      "model.language_model.layers.32.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.32.mlp.shared_experts.up_proj",
+      "model.language_model.layers.32.mlp.shared_experts.down_proj",
+      "model.language_model.layers.33.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.33.mlp.shared_experts.up_proj",
+      "model.language_model.layers.33.mlp.shared_experts.down_proj",
+      "model.language_model.layers.34.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.34.mlp.shared_experts.up_proj",
+      "model.language_model.layers.34.mlp.shared_experts.down_proj",
+      "model.language_model.layers.35.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.35.mlp.shared_experts.up_proj",
+      "model.language_model.layers.35.mlp.shared_experts.down_proj",
+      "model.language_model.layers.36.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.36.mlp.shared_experts.up_proj",
+      "model.language_model.layers.36.mlp.shared_experts.down_proj",
+      "model.language_model.layers.37.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.37.mlp.shared_experts.up_proj",
+      "model.language_model.layers.37.mlp.shared_experts.down_proj",
+      "model.language_model.layers.38.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.38.mlp.shared_experts.up_proj",
+      "model.language_model.layers.38.mlp.shared_experts.down_proj",
+      "model.language_model.layers.39.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.39.mlp.shared_experts.up_proj",
+      "model.language_model.layers.39.mlp.shared_experts.down_proj",
+      "model.language_model.layers.40.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.40.mlp.shared_experts.up_proj",
+      "model.language_model.layers.40.mlp.shared_experts.down_proj",
+      "model.language_model.layers.41.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.41.mlp.shared_experts.up_proj",
+      "model.language_model.layers.41.mlp.shared_experts.down_proj",
+      "model.language_model.layers.42.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.42.mlp.shared_experts.up_proj",
+      "model.language_model.layers.42.mlp.shared_experts.down_proj",
+      "model.language_model.layers.43.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.43.mlp.shared_experts.up_proj",
+      "model.language_model.layers.43.mlp.shared_experts.down_proj",
+      "model.language_model.layers.44.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.44.mlp.shared_experts.up_proj",
+      "model.language_model.layers.44.mlp.shared_experts.down_proj",
+      "model.language_model.layers.45.mlp.shared_experts.gate_proj",
+      "model.language_model.layers.45.mlp.shared_experts.up_proj",
+      "model.language_model.layers.45.mlp.shared_experts.down_proj",
       "lm_head"
     ],
     "kv_cache_scheme": null,
@@ -162,11 +306,12 @@
     "quantization_status": "compressed",
     "sparsity_config": {},
     "transform_config": {},
-    "version": "0.10.3.dev33+g33c52de"
+    "version": "0.11.1.a20250828"
   },
   "text_config": {
     "attention_bias": true,
     "attention_dropout": 0.0,
+    "dtype": "bfloat16",
     "eos_token_id": [
       151329,
       151336,
@@ -206,13 +351,11 @@
     "rope_theta": 10000.0,
     "routed_scaling_factor": 1.0,
     "topk_group": 1,
-    "torch_dtype": "bfloat16",
     "use_cache": true,
     "use_qk_norm": false,
     "vocab_size": 151552
   },
   "tie_word_embeddings": false,
-  "torch_dtype": "bfloat16",
   "transformers_version": "4.56.0.dev0",
   "video_end_token_id": 151342,
   "video_start_token_id": 151341,
@@ -233,7 +376,6 @@
     "patch_size": 14,
     "rms_norm_eps": 1e-05,
     "spatial_merge_size": 2,
-    "temporal_patch_size": 2,
-    "torch_dtype": "bfloat16"
+    "temporal_patch_size": 2
   }
 }

model-00001-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4357d60ff461c2acfe7b7cb562a6bc12b8ddfec1ee792ae9636603ca3601458b
+size 4999540600

model-00002-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f51fd8ead335874afab8f236aef173420bc98c4438816d3678e9e4d6fec12ba
+size 4998839240

model-00003-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0376c9975cdc2232128d3c434ddf29906ade0be93039871a8ef594a982b8d78
+size 4998547048

model-00004-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c834b0796060a271197eb813ed0ca783957d325e85d83453750005ba6b06d71b
+size 4998551072

model-00005-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cbb2e318c178bf4f56ad776c272986733d24a67a5a68109613a6f8001a2f2bf
+size 4998551624

model-00006-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cdcf289b9ef58a701030c27dbfcca697e8399ce2cfc44b9ce28a24d53bba3b6
+size 4998843432

model-00007-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b70784b17a8b038c69b1884222fc49066822bdb2292a1879d0724e50d0da3eea
+size 4998551536

model-00008-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3efb29f4544290a6629b244aa2f2ba87ec5bf00459aa6b7bbebe0232b3551a8
+size 4998551536

model-00009-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9bc59075c55df01634507cbb424c4a3de6da79b63a0bd09c8b2faa236aa461b
+size 4998843728

model-00010-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce4f03f6780f2add935254ebba3a76abc37a39ad5227d42e577f95cd2e221351
+size 4998551336

model-00011-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2c67db321f01d649ec7498440b9c64b21f49b2b8308cae42d8a4a0d8d2da7ea
+size 4998551536

model-00012-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:935d6fcb0c230a5574ddf200c4828fdb3f765e8fa6604552e911e19f99f07d7f
+size 4998551552

model-00013-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d7139e7933acf7b77864ce4d4d26c212230a3ad52fa2a1b73d70255fbbbfb8e
+size 3892761408

model-00014-of-00014.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a8d73275d82ef4305b377529c2426ad55ca13f52d208a15947efde7d8684f0e
+size 1241514112

recipe.yaml

@@ -13,12 +13,15 @@ default_stage:
             block_structure: null
             dynamic: false
             actorder: null
-            observer: minmax
+            observer: mse
             observer_kwargs: {}
           input_activations: null
           output_activations: null
+          format: null
       targets: [Linear]
-      ignore: [lm_head, 're:.*visual.*']
+      ignore: [lm_head, 're:.*embed_tokens', 're:.*input_layernorm', 're:.*post_attention_layernorm',
+        model.language_model.norm, 're:.*shared_experts.*', 're:model.language_model.layers.0.*',
+        're:.*mlp.gate', 're:model.visual.*']
       mappings:
       - smooth_layer: re:.*input_layernorm$
         balance_layers: ['re:.*q_proj$', 're:.*k_proj$', 're:.*v_proj$']
@@ -28,4 +31,5 @@ default_stage:
         balance_layers: ['re:.*gate_proj$', 're:.*up_proj$']
       - smooth_layer: re:.*up_proj$
         balance_layers: ['re:.*down_proj$']
+      offload_device: !!python/object/apply:torch.device [cpu]
       duo_scaling: true

video_preprocessor_config.json

@@ -31,6 +31,7 @@
   "processor_class": "Glm4vProcessor",
   "resample": 3,
   "rescale_factor": 0.00392156862745098,
+  "return_metadata": false,
   "size": {
     "longest_edge": 47040000,
     "shortest_edge": 12544