Fix non-MM multiturn: Use legacy formatting (#1247) #3014

Job	Run time
test-tinystories-executorch (16-core-ubuntu)	19m 18s
torchchat-command-load-test (macos-14)	1m 33s
test-tinystories-executorch (macos-14-xlarge)	17m 25s
compile-gguf (macos-14)	3m 53s
gather-models-cpu	11s
gather-models-gpu	14s
runner-et (16-core-ubuntu)	5m 46s
runner-aoti (16-core-ubuntu)	4m 41s
runner-et (macos-14-xlarge)	5m 4s
runner-aoti (macos-14-xlarge)	3m 11s
test-gguf-util (macos-14)	4m 30s
test-mps (macos-m1-stable)	53s
test-mps-dtype (macos-m1-stable)	1m 22s
test-torchao-experimental (macos-14-xlarge)	5m 42s
test-build-runner-et-android / linux-job	12m 48s
test-cpu-eval-sanity-check-float16 (x86_64, stories15M)	29m 14s
test-cpu-aoti (x86_64, stories15M)	17m 0s
test-cpu-compile (x86_64, stories15M)	15m 54s
test-cpu-eval-sanity-check-float16 (aarch64, stories15M)	6m 14s
test-cpu-eval-sanity-check (x86_64, stories15M)	9m 19s
test-cpu-aoti (aarch64, stories15M)	9m 22s
test-cpu-compile (aarch64, stories15M)	9m 41s
test-cpu-eval-sanity-check (aarch64, stories15M)	9m 28s
test-cpu-eval-sanity-check-float32 (x86_64, stories15M)	36m 49s
test-cpu-eval-sanity-check-float32 (aarch64, stories15M)	7m 59s
test-gpu-compile (cuda, stories15M) / linux-job	23m 22s
test-gpu-aoti-bfloat16 (cuda, stories15M) / linux-job	13m 3s
test-gpu-aoti-float16 (cuda, stories15M) / linux-job	12m 0s
test-gpu-aoti-float32 (cuda, stories15M) / linux-job	12m 6s
test-gpu-eval-sanity-check (cuda, stories15M) / linux-job	11m 13s
	5h 9m 15s

Provide feedback