blakeslee-research/nucleic-purpose-classifier
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge nucleic/plucky-north-vole-sdna into dev

Browse Source
This commit is contained in:
abkslm
2026-08-01 17:39:17 -07:00
parent 8ac00de087
commit 9ecb82c983
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
label_swe_chat_prompts.py
+8 -3
View File
@@ -27,7 +27,8 @@ from purpose_data import DataError, SOURCE_FIELDS, canonical_json, prompt_hash,
SCRIPT_DIR = Path(__file__).resolve().parent
DEFAULT_INPUT = SCRIPT_DIR / ".artifacts" / "swe-chat" / "candidates.jsonl"
DEFAULT_OUTPUT = SCRIPT_DIR / ".artifacts" / "swe-chat" / "labeled-source.jsonl"
MODEL = "gpt-5.6-luna"
DEFAULT_MODEL = "gpt-5.6-luna"
DEFAULT_REASONING_EFFORT = "low"
STATE_SCHEMA_VERSION = 1
DEFAULT_MAX_RESPONSE_CHARS = 48_000
FENCED_CODE_RE = re.compile(r"(?ms)^[ \t]*```[^\n]*\n.*?^[ \t]*```[ \t]*$")
@@ -252,7 +253,7 @@ def invoke_codex(args: argparse.Namespace, batch: Sequence[Candidate]) -> list[t
command += ["exec", "--ephemeral", "--ignore-user-config", "--ignore-rules", "--skip-git-repo-check"]
if isolation != "external":
command += ["--sandbox", "read-only"]
command += ["--model", MODEL, "--config", 'model_reasoning_effort="low"', "--output-schema", str(schema_path), "--output-last-message", str(response_path), "--color", "never", "-"]
command += ["--model", args.model, "--config", f'model_reasoning_effort="{args.reasoning_effort}"', "--output-schema", str(schema_path), "--output-last-message", str(response_path), "--color", "never", "-"]
try:
completed = subprocess.run(command, input=prompt, text=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, cwd=root, timeout=args.timeout_seconds, check=False)
if completed.returncode != 0:
@@ -345,7 +346,7 @@ def run(args: argparse.Namespace) -> dict[str, int]:
pending, batch_size=args.batch_size, batch_chars=args.batch_chars,
max_prompt_chars=args.max_prompt_chars, max_response_chars=args.max_response_chars,
)
print(f"input={len(source)} resumed={len(states)} pending={len(pending)} batches={len(batch_list)} model={MODEL}", flush=True)
print(f"input={len(source)} resumed={len(states)} pending={len(pending)} batches={len(batch_list)} model={args.model} reasoning={args.reasoning_effort}", flush=True)
for number, batch in enumerate(batch_list, 1):
decisions = invoke_codex(args, batch) # type: ignore[arg-type]
newly: list[dict[str, Any]] = []
@@ -373,6 +374,8 @@ def build_parser() -> argparse.ArgumentParser:
parser.add_argument("--state", type=Path)
parser.add_argument("--audit", type=Path)
parser.add_argument("--codex", default="codex")
parser.add_argument("--model", default=DEFAULT_MODEL)
parser.add_argument("--reasoning-effort", default=DEFAULT_REASONING_EFFORT)
parser.add_argument("--codex-isolation", choices=base.CODEX_ISOLATION_CHOICES, default="auto")
parser.add_argument("--batch-size", type=int, default=20)
parser.add_argument("--batch-chars", type=int, default=80_000)
@@ -393,6 +396,8 @@ def main(argv: Sequence[str] | None = None) -> int:
args.audit = args.audit.expanduser().resolve() if args.audit else args.output.with_name(f"{args.output.stem}.audit.jsonl")
if args.batch_size <= 0 or args.batch_chars <= 0 or args.max_prompt_chars < 1_000 or args.max_response_chars < 1_000 or args.timeout_seconds <= 0 or args.max_attempts <= 0 or (args.limit_sessions is not None and args.limit_sessions <= 0):
parser.error("batch sizes, timeout, attempts, and --limit-sessions must be positive; prompt/response limits must be at least 1000")
if not args.model.strip() or not args.reasoning_effort.strip():
parser.error("--model and --reasoning-effort must be non-empty")
try:
metrics = run(args)
except (DataError, OSError, ValueError, subprocess.SubprocessError) as error: