Edit mixtral quantization config file (#114) (#1739)

huggingface · Feb 5, 2025 · df43900 · df43900
1 parent 183a80d
commit df43900
Showing 1 changed file with 2 additions and 5 deletions.
diff --git a/examples/text-generation/quantization_config/maxabs_quant_mixtral.json b/examples/text-generation/quantization_config/maxabs_quant_mixtral.json
@@ -3,10 +3,7 @@
     "mode": "QUANTIZE",
     "observer": "maxabs",
     "scale_method": "maxabs_hw",
-    "allowlist": {"types": [], "names":  ["gate","w1","w3","w2"]},
-    "blocklist": {"types": [], "names":  [
-        "model.layers.1.block_sparse_moe.experts.(3|4).w2",
-        "model.layers.[29-31].block_sparse_moe.experts.[0-7].w2"
-    ]},
+    "allowlist": {"types": [], "names":  []},
+    "blocklist": {"types": [], "names":  ["self_attn"]},
     "dump_stats_path": "./hqt_output/measure"
 }