A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 4, 11:15 UTC
[CI] Add temperature to bfcl eval, default greedy (#41059)
Signed-off-by: Yifan Zong <yzong@redhat.com>
yzong-rh committed
Apr 29, 2026 at 17:01 UTC
93da1fe97abf71ac81e7daea21547292f9b39aa4
1 file changed
+6
-2
.buildkite/scripts/tool_call/run-bfcl-eval.sh
+6
-2
@@ -28,6 +28,7 @@
28
# BFCL_MAX_MODEL_LEN - Max model length (default: 4096)
29
# BFCL_PORT - Server port (default: 8000)
30
# BFCL_REASONING_PARSER - Reasoning parser name (default: disabled)
31
+# BFCL_TEMPERATURE - Temperature (default: 0.0)
32
# BFCL_EXTRA_ARGS - Additional vLLM server args
33
34
set -euo pipefail
@@ -43,6 +44,7 @@ TP_SIZE="${BFCL_TP_SIZE:-1}"
44
MAX_MODEL_LEN="${BFCL_MAX_MODEL_LEN:-4096}"
45
PORT="${BFCL_PORT:-8000}"
46
REASONING_PARSER="${BFCL_REASONING_PARSER:-}"
47
+TEMPERATURE="${BFCL_TEMPERATURE:-0.0}"
48
EXTRA_ARGS="${BFCL_EXTRA_ARGS:-}"
49
50
# Set up output directory
@@ -139,7 +141,7 @@ echo "vLLM server is ready. (started in ${SECONDS_WAITED}s)"
141
# be patched in-process so BFCL knows to use the OpenAI-compatible handler
142
# against our local vLLM server.
143
bfcl_exit_code=0
142
-python3 - "$MODEL" "$TEST_CATEGORY" "$NUM_THREADS" "$PORT" "$API_TYPE" "$OUTPUT_DIR" << 'PYEOF' || bfcl_exit_code=$?
144
+python3 - "$MODEL" "$TEST_CATEGORY" "$NUM_THREADS" "$PORT" "$API_TYPE" "$TEMPERATURE" "$OUTPUT_DIR" << 'PYEOF' || bfcl_exit_code=$?
145
import os
146
import sys
147
@@ -148,7 +150,8 @@ test_category = sys.argv[2]
150
num_threads = int(sys.argv[3])
151
port = sys.argv[4]
152
api_type = sys.argv[5]
151
-output_dir = sys.argv[6] if len(sys.argv) > 6 and sys.argv[6] else os.getcwd()
153
+temperature = float(sys.argv[6])
154
+output_dir = sys.argv[7] if len(sys.argv) > 7 and sys.argv[7] else os.getcwd()
155
156
os.environ["OPENAI_BASE_URL"] = f"http://localhost:{port}/v1"
157
os.environ["OPENAI_API_KEY"] = "dummy"
@@ -204,6 +207,7 @@ gen_kwargs["model"] = [model]
207
gen_kwargs["test_category"] = [c.strip() for c in test_category.split(",")]
208
gen_kwargs["skip_server_setup"] = True
209
gen_kwargs["num_threads"] = num_threads
210
+gen_kwargs["temperature"] = temperature
211
generate(**gen_kwargs)
212
213
# ---- evaluate ----