| #!/usr/bin/env bash |
| |
| |
| |
| |
| |
| |
| |
|
|
| set -euo pipefail |
| cd "$(dirname "$0")/.." |
|
|
| MODEL="${1:-Qwen/Qwen3-0.6B}" |
| SUFFIX=$(echo "$MODEL" | grep -oE '[0-9]+\.[0-9]+B' | tr '.' '_' || echo "custom") |
| STAMP="${TRACE_RUN_TIMESTAMP:-$(date -u +%Y%m%dT%H%M%SZ)}" |
| RUN_PREFIX="${TRACE_RUN_PREFIX:-collect}" |
| BATCH_ID="${RUN_PREFIX}-${SUFFIX}-${STAMP}" |
| MBPP_RUN_ID="${BATCH_ID}-mbpp" |
| NL2BASH_RUN_ID="${BATCH_ID}-nl2bash" |
| MBPP_OUTPUT="datasets/mbpp_${SUFFIX}_${STAMP}.jsonl" |
| NL2BASH_OUTPUT="datasets/nl2bash_${SUFFIX}_${STAMP}.jsonl" |
| MBPP_MANIFEST="runs/${MBPP_RUN_ID}.json" |
| NL2BASH_MANIFEST="runs/${NL2BASH_RUN_ID}.json" |
| BATCH_MANIFEST="runs/${BATCH_ID}.json" |
| STARTED_AT="$(date -u +%Y-%m-%dT%H:%M:%SZ)" |
| MODEL_ARTIFACT="${LLAMA_MODEL:-}" |
| CONTEXT_LENGTH="${LLAMA_CTX:-16384}" |
| CHAT_TEMPLATE="${LLAMA_CHAT_TEMPLATE_FILE:-}" |
| JINJA="${LLAMA_JINJA:-1}" |
| FLASH_ATTN="${LLAMA_FLASH_ATTN:-1}" |
| N_GPU_LAYERS="${LLAMA_N_GPU_LAYERS:-}" |
|
|
| shared_args=( |
| --model "$MODEL" |
| --context-length "$CONTEXT_LENGTH" |
| ) |
|
|
| if [ -n "$MODEL_ARTIFACT" ]; then |
| shared_args+=(--model-artifact "$MODEL_ARTIFACT") |
| fi |
| if [ -n "$CHAT_TEMPLATE" ]; then |
| shared_args+=(--chat-template "$CHAT_TEMPLATE") |
| fi |
| if [ -n "$N_GPU_LAYERS" ]; then |
| shared_args+=(--n-gpu-layers "$N_GPU_LAYERS") |
| fi |
| if [[ "$JINJA" =~ ^(1|true|TRUE|yes|YES|on|ON)$ ]]; then |
| shared_args+=(--jinja) |
| else |
| shared_args+=(--no-jinja) |
| fi |
| if [[ "$FLASH_ATTN" =~ ^(1|true|TRUE|yes|YES|on|ON)$ ]]; then |
| shared_args+=(--flash-attn) |
| else |
| shared_args+=(--no-flash-attn) |
| fi |
|
|
| |
| if ! curl -sf "http://127.0.0.1:${LLAMA_PORT:-8080}/health" > /dev/null 2>&1; then |
| echo "ERROR: llama-server not running. Start it with: ./scripts/llama-server.sh start" |
| exit 1 |
| fi |
|
|
| echo "=== Collection run: $MODEL ===" |
| echo " Batch ID: $BATCH_ID" |
| echo " MBPP run ID: $MBPP_RUN_ID" |
| echo " MBPP output: $MBPP_OUTPUT" |
| echo " NL2Bash run ID: $NL2BASH_RUN_ID" |
| echo " NL2Bash output: $NL2BASH_OUTPUT" |
| echo " Started: $STARTED_AT" |
| echo "" |
|
|
| echo "[1/2] MBPP (500 prompts)..." |
| uv run python collect.py \ |
| --dataset mbpp \ |
| -n 500 \ |
| --run-id "$MBPP_RUN_ID" \ |
| "${shared_args[@]}" \ |
| --output "$MBPP_OUTPUT" |
| echo "" |
|
|
| echo "[2/2] NL2Bash (200 prompts)..." |
| uv run python collect.py \ |
| --dataset nl2bash \ |
| -n 200 \ |
| --run-id "$NL2BASH_RUN_ID" \ |
| "${shared_args[@]}" \ |
| --output "$NL2BASH_OUTPUT" |
| echo "" |
|
|
| FINISHED_AT="$(date -u +%Y-%m-%dT%H:%M:%SZ)" |
| mkdir -p runs |
| python3 - "$BATCH_ID" "$MODEL" "$STARTED_AT" "$FINISHED_AT" "$MBPP_RUN_ID" "$MBPP_OUTPUT" "$MBPP_MANIFEST" "$NL2BASH_RUN_ID" "$NL2BASH_OUTPUT" "$NL2BASH_MANIFEST" "$BATCH_MANIFEST" <<'PY' |
| import json |
| import sys |
| from pathlib import Path |
|
|
| ( |
| batch_id, |
| model, |
| started_at, |
| finished_at, |
| mbpp_run_id, |
| mbpp_output, |
| mbpp_manifest, |
| nl2bash_run_id, |
| nl2bash_output, |
| nl2bash_manifest, |
| batch_manifest, |
| ) = sys.argv[1:] |
|
|
| payload = { |
| "batch_id": batch_id, |
| "model": model, |
| "started_at_utc": started_at, |
| "finished_at_utc": finished_at, |
| "runs": [ |
| { |
| "dataset": "mbpp", |
| "run_id": mbpp_run_id, |
| "output_path": mbpp_output, |
| "manifest_path": mbpp_manifest, |
| }, |
| { |
| "dataset": "nl2bash", |
| "run_id": nl2bash_run_id, |
| "output_path": nl2bash_output, |
| "manifest_path": nl2bash_manifest, |
| }, |
| ], |
| } |
|
|
| path = Path(batch_manifest) |
| path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") |
| PY |
|
|
| echo "=== Done: $FINISHED_AT ===" |
| echo "Batch manifest: $BATCH_MANIFEST" |
| echo "Analyzing results..." |
| uv run python analyze.py "$MBPP_OUTPUT" |
| echo "" |
| uv run python analyze.py "$NL2BASH_OUTPUT" |
|
|
