[CI] Use sgl-eval for GLM-5.3-Flash GSM8K tests (#43975)

Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
This commit is contained in:
Mohammad Miadh Angkad
2026-10-13 02:26:50 -07:00
committed by GitHub
co-authored by Mohammad Angkad
parent e0c957e222
commit 195cb7afb4
3 changed files with 45 additions and 22 deletions
+5 -1
View File
@@ -94,6 +94,7 @@ def _run_sgl_eval(
chat_template_kwargs: Optional[dict] = None,
max_tokens: Optional[int] = None,
temperature: Optional[float] = None,
top_p: Optional[float] = None,
accept_length_thres: Optional[float] = None,
):
"""Shared sgl-eval driver for the MMLU sanity gate and the GSM8K ``sgl_eval`` backend.
@@ -118,6 +119,7 @@ def _run_sgl_eval(
num_threads=num_threads,
max_tokens=max_tokens,
temperature=temperature,
top_p=top_p,
chat_template_kwargs=(
chat_template_kwargs
if chat_template_kwargs is not None
@@ -176,7 +178,7 @@ class GSM8KMixin:
gsm8k_backend: str = "run_eval" # "run_eval" | "sgl_eval"
gsm8k_thinking: bool = False # sgl_eval backend
gsm8k_max_tokens: Optional[int] = None # sgl_eval backend
# None keeps run_eval's greedy default; set both to route the run through
# None keeps the backend's default; set both to route the run through
# the sampling path.
gsm8k_temperature: Optional[float] = None
gsm8k_top_p: Optional[float] = None
@@ -200,6 +202,8 @@ class GSM8KMixin:
num_threads=self.gsm8k_num_threads,
thinking=self.gsm8k_thinking,
max_tokens=self.gsm8k_max_tokens,
temperature=self.gsm8k_temperature,
top_p=self.gsm8k_top_p,
accept_length_thres=self.gsm8k_accept_length_thres,
)
else: