[CI] BugFix Eval Small Models Distributed test for DiffusionGemma (#47980)

Signed-off-by: Markov Ilya <markovilya19@gmail.com>
Co-authored-by: Markov Ilya <markovilya19@gmail.com>

tests/evals/gsm8k/configs/DiffusionGemma-26B-A4B-it-FP8-dynamic.yaml

@@ -4,6 +4,10 @@ num_questions: 1319
 num_fewshot: 5
 startup_max_wait_seconds: 1200
 use_chat_completions: true
+# Diffusion models use a fixed internal temperature schedule and do not
+# support per-request temperature or seed overrides.
+temperature: 1.0
+seed: null
 server_args: >-
   --enforce-eager
   --max-model-len 4096

tests/evals/gsm8k/test_gsm8k_correctness.py

@@ -69,6 +69,8 @@ def run_gsm8k_eval(eval_config: dict, server_url: str) -> dict:
         use_chat_completions=eval_config.get("use_chat_completions", False),
         host=host,
         port=port,
+        temperature=eval_config.get("temperature", 0.0),
+        seed=eval_config.get("seed", 42),
         request_timeout_seconds=request_timeout_seconds,
         gen_prefix=eval_config.get("gen_prefix", ""),
     )