NGRAMMERGE(1)

✨ NAME ✨

ngrammerge — it's the manual page for ngrammerge version 1.3.17, bestie 💖

---

🌸 DESCRIPTION 🌸

Okay so, real talk: ngrammerge merges n-gram models. That's the whole vibe. You feed it a bunch of FSTs and it smushes them together into one. Iconic behavior honestly 🫶

Here's how you're supposed to call it:

Usage: ngrammerge [--options] -ofile=out.fst in1.fst in2.fst [in3.fst ...]

Like, yes, you do need -ofile for the output, and yes, you can throw in in3.fst and beyond if you're feeling extra. Not me being greedy with input files 💅

🎀 PROGRAM FLAGS 🎀

These are the flags that are like, hers. The main-character ones 👑

• --alpha — type = double, default = 1. The weight for the first FST. It's giving "first FST gets a say" energy.
• --backoff_label — type = int64_t, default = 0. The backoff label, periodt.
• --beta — type = double, default = 1. Weight for the second (and subsequent) FST(s). So yes, everything after the first one runs on beta ✨
• --check_consistency — type = bool, default = false. Checks model consistency. Turn it on if you want that reassurance, bestie.
• --complete — type = bool, default = false. Completes partial models. Healing them, literally 🧚
• --context_pattern — type = std::string, default = "". The context pattern for the second FST.
• --contexts — type = std::string, default = "". Context patterns file (all FSTs). All of them. Every single one 💜
• --max_replace_order — type = int32_t, default = -1. Maximum order to replace in replace_merge, and it's ignored if < 1. So if you set it to 0 or anything negative, it just... doesn't care. Lowkey rude but okay 😭
• --method — type = std::string, default = "count_merge". One of: "context_merge", "count_merge", "model_merge", "bayes_model_merge", "histogram_merge", "replace_merge". Six options, six whole personalities 🌈
• --norm_eps — type = double, default = 0.001. Normalization check epsilon. Tiny number, big job.
• --normalize — type = bool, default = false. Normalizes the resulting model. Get it together, model 💅
• --ofile — type = std::string, default = "". Output file. This is the one you actually came here for.
• --round_to_int — type = bool, default = false. Rounds all merged counts to integers. No decimals allowed, no cap 🚫

📚 LIBRARY FLAGS 📚

These are the borrowed ones — the group chat that came with the library. Still important, still slay.

Flags from: flags.cc

• --help — type = bool, default = false. Shows usage information. The classic. When in doubt, ask for help 🥺
• --helpshort — type = bool, default = false. Shows brief usage information. Help but make it snack-sized.
• --tmpdir — type = std::string, default = "/tmp". Temporary directory. Living in /tmp, rent free 🌸
• --v — type = int32_t, default = 0. Verbosity level. Crank it up if you want it to yap.

Flags from: fst.cc

• --fst_align — type = bool, default = false. Write FST data aligned where appropriate. Alignment, but make it storage ✨
• --fst_default_cache_gc — type = bool, default = true. Enable garbage collection of cache. Cleaning up after itself, we love a self-sufficient icon 🧹
• --fst_default_cache_gc_limit — type = int64_t, default = 1048576. Cache byte size that triggers garbage collection. Hit 1048576 bytes and the GC wakes up 💀
• --fst_read_mode — type = std::string, default = "read". Default file reading mode for mappable files.
• --fst_verify_properties — type = bool, default = false. Verify FST properties queried by TestProperties.
• --save_relabel_ipairs — type = std::string, default = "". Save input relabel pairs to file.
• --save_relabel_opairs — type = std::string, default = "". Save output relabel pairs to file. (Yes, ipairs and opairs are two different flags. Input and output. Don't mix them up bestie 🙈)

Flags from: ngram-output.cc

• --end_symbol — type = std::string, default = "</s>". Class label for sentence end. Goodbye symbol 🫶
• --start_symbol — type = std::string, default = "<s>". Class label for sentence start. Hello symbol 👋

Flags from: symbol-table.cc

• --fst_compat_symbols — type = bool, default = true. Require symbol tables to match when appropriate. Matching outfits, but for symbol tables 👯
• --fst_field_separator — type = std::string, default = " ". Set of characters used as a separator between printed fields. (That default is a whole run of spaces, so like — be aware 💀)

Flags from: util.cc

• --fst_error_fatal — type = bool, default = true. FST errors are fatal; o.w. return objects flagged as bad: e.g., FSTs: kError property set, FST weights: not a Member(). So by default, an FST error is a whole crisis. Turn it off and instead you get objects marked as bad instead of the program just dying 😭
• --ngram_error_fatal — type = bool, default = true. NGram errors are fatal if true; otherwise returns objects flagged as bad: e.g., NGramModel::Error() is true.

⚠️ Bestie listen — both of those _error_fatal flags default to true. Fatal means fatal. If you'd rather get back a flagged-bad object instead of an abrupt goodbye, you gotta flip them yourself, no cap.

Flags from: weight.cc

• --fst_weight_parentheses — type = std::string, default = "". Characters enclosing the first weight of a printed composite weight (e.g., pair weight, tuple weight and derived classes) to ensure proper I/O of nested composite weights; must have size 0 (none) or 2 (open and close parenthesis). Size 1? Not allowed. Size 3? Absolutely not. Zero or two, that's the rule 💅
• --fst_weight_separator — type = std::string, default = ",". Character separator between printed composite weights; must be a single character. One character. Singular. That's it, that's the law 👑

---

And that's the whole page, bestie!! Every flag, every default, every little type signature — all of it, nothing skipped, we ate and left no crumbs 🍓

This was ngrammerge version 1.3.17, from February 2022 — shouting out the source and the date one last time on the way out 💖 Bye pookie, go merge some n-grams 🦋