Fix quantization_config.ignore: use fused module names (qkv_proj/gate_up_proj) for vLLM

config.json

@@ -70,17 +70,13 @@
   "ignore": [
    "lm_head",
    "model.embed_tokens",
-   "*.self_attn.q_proj",
-   "*.self_attn.k_proj",
-   "*.self_attn.v_proj",
+   "*.self_attn.qkv_proj",
    "*.self_attn.o_proj",
-   "*.mlp.shared_mlp.gate_proj",
-   "*.mlp.shared_mlp.up_proj",
+   "*.mlp.gate_up_proj",
+   "*.mlp.down_proj",
+   "*.mlp.shared_mlp.gate_up_proj",
    "*.mlp.shared_mlp.down_proj",
    "*.mlp.router.gate",
-   "*.mlp.gate_proj",
-   "*.mlp.up_proj",
-   "*.mlp.down_proj",
    "*.eh_proj"
   ]
  }