LLM-Server: Bench-Image um Task-Extras ergaenzt (math, ifeval)

lm-eval[api,math] plus langdetect und immutabledict: minerva_math500 braucht
sympy/math_verify/antlr4, leaderboard_ifeval langdetect und immutabledict.
Beide Tasks per Smoke-Test gegen die Engine verifiziert (Datensatz-Download,
Generierung, Scoring, Sample-Speicherung).

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
This commit is contained in:
2026-11-03 06:01:37 +01:00
parent fc880e4c8a
commit efb41d229b
@@ -6,7 +6,11 @@ ENV PIP_NO_CACHE_DIR=1 \
HF_HOME=/hf \
PYTHONUNBUFFERED=1
RUN pip install --no-cache-dir "lm-eval[api]=={{ benchmark_lm_eval_version }}"
# Task extras: [api] for API-based models, [math] for minerva/math500
# scoring (sympy, math_verify, antlr4), langdetect + immutabledict for
# ifeval verification.
RUN pip install --no-cache-dir "lm-eval[api,math]=={{ benchmark_lm_eval_version }}" \
&& pip install --no-cache-dir langdetect immutabledict
WORKDIR /work
ENTRYPOINT ["lm_eval"]