Skip to content

Commit a19b1a1

Browse files
HDCharlesHDCharles
committed
skip gate
Summary Signed-off-by: HDCharles <charlesdavidhernandez@gmail.com>
1 parent 41f4481 commit a19b1a1

File tree

4 files changed

+20
-0
lines changed

4 files changed

+20
-0
lines changed

‎tests/e2e/vLLM/configs/qwen3_fp4_nvfp4.yaml‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,3 +5,5 @@ scheme: NVFP4
55
dataset_id: HuggingFaceH4/ultrachat_200k
66
dataset_split: train_sft
77
num_calibration_samples: 20
8+
recipe: tests/e2e/vLLM/recipes/fp4_nvfp4_recipe_skip_gate.yaml
9+
save_dir: "Qwen3-30B-A3B-NVFP4-skip-gate"

‎tests/e2e/vLLM/configs/qwen3_fp8_dynamic_per_token.yaml‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,3 +2,5 @@ cadence: "nightly"
22
test_type: "regression"
33
model: Qwen/Qwen3-30B-A3B
44
scheme: FP8_DYNAMIC
5+
recipe: tests/e2e/vLLM/recipes/fp8_dynamic_recipe_skip_gate.yaml
6+
save_dir: "Qwen3-30B-A3B-FP8_DYNAMIC-skip-gate"

‎tests/e2e/vLLM/recipes/fp4_nvfp4_recipe_skip_gate.yaml‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,8 @@
1+
quant_stage:
2+
quant_modifiers:
3+
QuantizationModifier:
4+
targets: Linear
5+
scheme: NVFP4
6+
ignore:
7+
- lm_head
8+
- "re:.*mlp.gate.*"

‎tests/e2e/vLLM/recipes/fp8_dynamic_recipe_skip_gate.yaml‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,8 @@
1+
quant_stage:
2+
quant_modifiers:
3+
QuantizationModifier:
4+
targets: Linear
5+
scheme: FP8_DYNAMIC
6+
ignore:
7+
- lm_head
8+
- "re:.*mlp.gate.*"

0 commit comments

Comments
 (0)