125cfa1d90b1e1cfc8321e65140f6217c0025144a9af712e12d4a048b1137bc1

config.json

@@ -1,6 +1,6 @@
 {
   "_attn_implementation_autoset": true,
-  "_name_or_path": "PowerInfer/SmallThinker-3B-Preview",
+  "_name_or_path": "/tmp/models/tmpdm8gg0sl/tmp0_8ifq2x",
   "architectures": [
     "Qwen2ForCausalLM"
   ],
@@ -17,6 +17,26 @@
   "num_attention_heads": 16,
   "num_hidden_layers": 36,
   "num_key_value_heads": 2,
+  "quantization_config": {
+    "quant_config": {
+      "offload_meta": false,
+      "scale_quant_params": null,
+      "weight_quant_params": {
+        "axis": 1,
+        "channel_wise": true,
+        "group_size": 64,
+        "nbits": 8,
+        "optimize": true,
+        "round_zero": false,
+        "view_as_float": false
+      },
+      "zero_quant_params": null
+    },
+    "quant_method": "hqq",
+    "skip_modules": [
+      "lm_head"
+    ]
+  },
   "rms_norm_eps": 1e-06,
   "rope_scaling": null,
   "rope_theta": 1000000.0,

smash_config.json

@@ -11,7 +11,7 @@
     "quant_hqq_weight_bits": 8,
     "max_batch_size": 1,
     "device": "cuda",
-    "cache_dir": "/tmp/models/tmpeqq5s829",
+    "cache_dir": "/tmp/models/tmpdm8gg0sl",
     "task": "",
     "save_load_fn": "hqq",
     "save_load_fn_args": {},