submission.env.example

# ─────────────────────────────────────────────────────────────────────────────
# Submission runner config. Copy to `submission.env`, fill in, then run:
#
#   docker run --rm --env-file submission.env -v "$PWD/out:/out" \
#     problem-reductions-runner:v0.6.0
#   # → ./out/submission.json
#
# Most runs only need the two REQUIRED lines. Everything below them has a sane
# default — leave a line commented out to keep the default.
# (submission.env is gitignored — it may hold your API key; never commit it.)
# ─────────────────────────────────────────────────────────────────────────────

# ── REQUIRED ─────────────────────────────────────────────────────────────────
# Any LiteLLM-routable model + its matching key. Examples of MODEL_NAME:
#   anthropic/claude-...   openai/gpt-...   openrouter/...   gemini/...
#   openai/<model> together with API_BASE for an OpenAI-compatible endpoint
MODEL_NAME=openai/gpt-5.4
API_KEY=...          # generic key — works for any provider
# Prefer the provider's own env var? Use it INSTEAD of API_KEY, e.g.:
#   OPENAI_API_KEY=...   ANTHROPIC_API_KEY=...   OPENROUTER_API_KEY=...   GEMINI_API_KEY=...

# Price is REQUIRED — there is no built-in table (a stale default would silently
# mis-meter the $20 cap). Spend is computed as token_usage × your price, so YOU
# declare the rate (USD / 1M tokens). Look it up from your provider's pricing page.
PRICE_IN=3.0         # USD / 1M input tokens
PRICE_OUT=15.0       # USD / 1M output tokens
# PRICE_CACHE_READ=0.30    # set these too for prompt-caching models
# PRICE_CACHE_WRITE=3.75

# ── NON-STANDARD PROVIDER (OpenRouter / gateway / local vLLM / Azure …) ───────
# Nothing is hardcoded to anthropic/openai. Reach any model with these:
# API_BASE=https://my-gateway.example/v1
# API_KEY=...          # generic key when no PROVIDER_API_KEY name fits
# MODEL_KWARGS={"custom_llm_provider":"openai","extra_headers":{"X-Org":"acme"}}
#   # ^ JSON of extra litellm.completion kwargs (api_version, temperature, …)

# ── BUDGET (defaults are the ranked configuration — change only for smoke runs) ─
# BUDGET_USD=20        # must be 20 to be ranked
# PER_RULE_BUDGET=0.5  # per-rule cost cap
# SAFETY_MARGIN=1      # USD held back so the budget-crossing call stays under cap
# MAX_TOKENS=8192      # per-call output ceiling
# MAX_RULES=5          # cap rules attempted (smoke runs only; omit for a real run)

# ── AGENT MODE ────────────────────────────────────────────────────────────────
# AGENT_MODE=per-rule  # per-rule (default): one isolated session per rule, budget split
#                      # evenly. whole-repo: ONE session over the whole library — the agent
#                      # enumerates and triages the rules itself under a single budget.

# ── CUSTOM AGENT PROMPT (mount the files too: -v "$PWD/cfg:/cfg") ─────────────
# AGENT_STRATEGY_FILE=/cfg/strategy.md   # extra hints into the prompt's {{strategy}} slot
# AGENT_CONFIG=/cfg/config.yaml          # full prompt-config replacement

# ── VERSION PINS (baked into the image from PR_REF; override only to debug) ────
# EXPECTED_PRED_VERSION=    # empty string disables the version check
# EXPECTED_PRED_COMMIT=

# ── METADATA ─────────────────────────────────────────────────────────────────
# SUBMITTED_BY=your-hf-handle   # also fillable at submit time on the Space