all tests passing

Signed-off-by: Brian Dellabetta <[email protected]>
vllm-project · Feb 17, 2025 · 05e6c32 · 05e6c32
1 parent b5fccc7
commit 05e6c32
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 8 deletions.
diff --git a/tests/e2e/vLLM/recipes/INT8/recipe_int8_channel_weight_dynamic_per_token.yaml b/tests/e2e/vLLM/recipes/INT8/recipe_int8_channel_weight_dynamic_per_token.yaml
@@ -3,7 +3,7 @@ quant_stage:
     SmoothQuantModifier:
       smoothing_strength: 0.8
     GPTQModifier:
-      ignore: [lm_head, "re:vision_tower.*", "re:multi_modal_projector.*"]
+      ignore: ["lm_head", "re:vision_tower.*", "re:multi_modal_projector.*", "re:visual.*", "re:vision_model.*"]
       config_groups:
         group_0:
           weights: {num_bits: 8, type: int, symmetric: true, strategy: channel}

diff --git a/tests/e2e/vLLM/recipes/actorder/recipe_w4a16_actorder_weight.yaml b/tests/e2e/vLLM/recipes/actorder/recipe_w4a16_actorder_weight.yaml
@@ -1,7 +1,7 @@
 quant_stage:
   quant_modifiers:
     GPTQModifier:
-      ignore: ["lm_head", "re:vision_tower.*", "re:multi_modal_projector.*"]
+      ignore: ["lm_head", "re:vision_tower.*", "re:multi_modal_projector.*", "re:visual.*", "re:vision_model.*"]
       config_groups:
         group_0:
           weights:

diff --git a/tests/lmeval/configs/vl_int8_w8a8_dynamic_per_token.yaml b/tests/lmeval/configs/vl_int8_w8a8_dynamic_per_token.yaml
@@ -15,4 +15,5 @@ lmeval:
   num_fewshot: 0
   limit: 1000
   metrics:
-    acc,none: 0.233
+    acc,none: 0.233
+  batch_size: 8
diff --git a/tests/lmeval/configs/vl_w4a16_actorder_weight.yaml b/tests/lmeval/configs/vl_w4a16_actorder_weight.yaml
@@ -1,8 +1,6 @@
 cadence: "weekly"
-#TODO llama 3.2 11B keeps hitting OOM errors
-# model: meta-llama/Llama-3.2-11B-Vision-Instruct
-# model_class: TraceableMllamaForConditionalGeneration
-model: microsoft/Phi-3-vision-128k-instruct
+model: Qwen/Qwen2-VL-2B-Instruct
+model_class: TraceableQwen2VLForConditionalGeneration
 recipe: tests/e2e/vLLM/recipes/actorder/recipe_w4a16_actorder_weight.yaml
 dataset_id: lmms-lab/flickr30k
 dataset_split: "test[:512]"
@@ -17,4 +15,5 @@ lmeval:
   num_fewshot: 0
   limit: 1000
   metrics:
-    acc,none: 0.3
+    acc,none: 0.333
+  batch_size: 4