[CI][Bugfix] Correct the Laguna DFlash acceptance-length reference (#57647)
Signed-off-by: Oxana Korzh <okorzh@amd.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com>
Oxana Korzh committed
Sep 18, 2026 at 19:44 UTC
038cb14bd7874030e98bb6edc4b6f4efe649fff7
1 file changed
+1
-1
tests/v1/e2e/spec_decode/acceptance_rates/dflash/test_dflash.py
+1
-1
@@ -53,7 +53,7 @@ LAGUNA_DFLASH_NVFP4 = DFlashCorrectnessConfig(
53
model="poolside/Laguna-XS-2.1-NVFP4",
54
draft_model="poolside/Laguna-XS-2.1-DFlash-NVFP4",
55
expected_accuracy=0.7, # Standard GSM8K sanity floor.
56
- expected_acceptance_len=3.55 * 0.9,
56
+ expected_acceptance_len=3.05 * 0.9,
57
num_speculative_tokens=15,
58
max_model_len=8192,
59
max_num_seqs=32,