Update README_nvidia.md

docs/mlperf/inference/gpt-j/README_nvidia.md

@@ -17,6 +17,8 @@ cmr "generate-run-cmds inference _find-performance _all-scenarios" \
 ```
 * Use `--division=closed` to run all scenarios for the closed division.
 * Use `--category=datacenter` to run datacenter scenarios
+* Use `--model=gptj-99.9` to run the high-accuracy model
+* Use `--rerun` to force a rerun even when result files (from a previous run) exist
 
 ### Do full accuracy and performance runs for all the scenarios
 
@@ -30,6 +32,8 @@ cmr "generate-run-cmds inference _submission _all-scenarios" --model=gptj-99 \
 * Use `--power=yes` for measuring power. It is ignored for accuracy and compliance runs
 * Use `--division=closed` to run all scenarios for the closed division. No compliance runs are there for gptj. 
 * `--offline_target_qps`, `--server_target_qps`, and `--singlestream_target_latency` can be used to override the determined performance numbers
+* Use `--model=gptj-99.9` to run the high-accuracy model
+* Use `--rerun` to force a rerun even when result files (from a previous run) exist
 
 
 ### Generate and upload MLPerf submission