[ROCm][CI] Pinning lm-eval version to resolve multi-modal small eval bug (#34038)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

requirements/rocm-test.txt

@@ -58,7 +58,7 @@ schemathesis==3.39.15
     # OpenAI schema test
 
 # Evaluation and benchmarking
-lm-eval[api]>=0.4.9.2
+lm-eval[api]==0.4.9.2
 jiwer==4.0.0
 
 # Required for multiprocessed tests that use spawn method, Datasets and Evaluate Test