项目文件夹

文件
2026-07-13 12:24:33 +08:00

25 行
927 B
Bash

#!/usr/bin/env bash
# Run the TTFT benchmark at several context sizes and collate the results.
# ---------------------------------------------------------------
BENCH="openai_chat_completion_client.py" # path to your benchmark script
MASTER_OUT="all_ttft_results.jsonl" # final merged log
CONTEXT_SIZES=(50 1000 2000 8000 16000 24000 32000 64000 96000 128000)
: > "$MASTER_OUT" # truncate / create the final log file
for TOKENS in "${CONTEXT_SIZES[@]}"; do
MAX_CTX=$((TOKENS))
OUTFILE="ttft_${TOKENS}.jsonl"
echo -e "\n▶︎ Running ${TOKENS}-token test (max_ctx_tokens=${MAX_CTX})"
python "$BENCH" --max_ctx_tokens "$MAX_CTX" \
--num_following 1 \
--out "$OUTFILE" \
--model meta-llama/Llama-3.3-70B-Instruct \
cat "$OUTFILE" >> "$MASTER_OUT" # append to the master log
done
echo -e "\n✅ All done – combined results in $MASTER_OUT"