54f6b34a04e47a6c932a99dcbbc5c0d73570f00f2654156b7a4e2bc90814b28d

config.json

@@ -1,6 +1,6 @@
 {
   "_attn_implementation_autoset": true,
-  "_name_or_path": "google/gemma-2-2b",
+  "_name_or_path": "/tmp/models/tmpnthtz7dc/tmp_e9rkptp",
   "architectures": [
     "Gemma2ForCausalLM"
   ],
@@ -23,6 +23,26 @@
   "num_hidden_layers": 26,
   "num_key_value_heads": 4,
   "pad_token_id": 0,
+  "quantization_config": {
+    "quant_config": {
+      "offload_meta": false,
+      "scale_quant_params": null,
+      "weight_quant_params": {
+        "axis": 1,
+        "channel_wise": true,
+        "group_size": 64,
+        "nbits": 4,
+        "optimize": true,
+        "round_zero": true,
+        "view_as_float": false
+      },
+      "zero_quant_params": null
+    },
+    "quant_method": "hqq",
+    "skip_modules": [
+      "lm_head"
+    ]
+  },
   "query_pre_attn_scalar": 256,
   "rms_norm_eps": 1e-06,
   "rope_theta": 10000.0,

smash_config.json

@@ -11,7 +11,7 @@
     "quant_hqq_weight_bits": 4,
     "max_batch_size": 1,
     "device": "cuda",
-    "cache_dir": "/tmp/models/tmp28ewnzjn",
+    "cache_dir": "/tmp/models/tmpnthtz7dc",
     "task": "",
     "save_load_fn": "hqq",
     "save_load_fn_args": {},