add actorder e2e test

vllm-project · Aug 25, 2024 · cdc8bcd · cdc8bcd
1 parent e4d37a6
commit cdc8bcd
Show file tree

Hide file tree

Showing 2 changed files with 24 additions and 0 deletions.
diff --git a/tests/llmcompressor/transformers/compression/configs/actorder_1.1b.yaml b/tests/llmcompressor/transformers/compression/configs/actorder_1.1b.yaml
@@ -0,0 +1,5 @@
+cadence: "nightly"
+test_type: "regression"
+model_stub: "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T"
+new_recipe: "tests/llmcompressor/transformers/compression/recipes/new_quant_actorder.yaml"
+ppl_threshold: 20
diff --git a/tests/llmcompressor/transformers/compression/recipes/new_quant_actorder.yaml b/tests/llmcompressor/transformers/compression/recipes/new_quant_actorder.yaml
@@ -0,0 +1,19 @@
+test_stage:
+    quant_modifiers:
+        QuantizationModifier:
+            ignore: ["lm_head", "model.layers.0.mlp.down_proj"]
+            config_groups:
+                group_0:
+                    weights:
+                        num_bits: 4
+                        type: "int"
+                        symmetric: False
+                        strategy: "group"
+                        group_size: 128
+                        actorder: True
+                    input_activations: null
+                    output_activations: null
+                    targets: ["Linear"]
+        GPTQModifier:
+            block_size: 128
+            sequential_update: False