litetune

Fine-tune a small model, convert it to run on a phone, and know what the conversion cost you.

Getting from a Hugging Face checkpoint to a model that works inside your app is a long road — LoRA, merge, export to .litertlm, bundle metadata — and a mistake at any step produces a file of the right size that loads without error and is broken while every check stays green. litetune walks that road and knows the traps on it.

The output is a .litertlm bundle: what LiteRT-LM loads, and what the flutter_gemma plugin runs on Android and iOS.

Any task shaped as prompt → completion works. What "correct" means is the one thing you choose: --scorer tool-call for function calling, where the operation name and every argument value must match, or --scorer exact-text where there is one right string. Everything after scoring — the paired comparison, the intervals, whether a difference resolves, the exit code — reads only whether each example was right, so it does not know or care which task you brought.

• prepare — split the data, and reject rows that cannot be scored
• tune — LoRA or full fine-tuning, with the wiring the export needs
• convert — checkpoint to .litertlm, across quantization recipes
• verify — measure what the conversion cost, before you ship
• bundle — package the artifact with what was measured about it

litetune env shows the environments the stages cached, and --clean removes them.

Everything runs on CPU, which is workable at 270M and the first thing you will want to change above about 1B. Bring your own checkpoint and skip the first two steps, or bring a .litertlm and its float checkpoint and run only verify.

Alpha. Measured end to end on google/functiongemma-270m-it and function calling only. The other scorer is tested but no task has been measured through it end to end. Gemma 3, Gemma 4 and Qwen3.5 export — litetune carries their required flags — but no quality number has been established for them. Try it on yours and open an issue.

Install

pip install litetune

Or through Homebrew, which brings its own Python 3.12:

brew install DenisovAV/tap/litetune

Linux or macOS, Python 3.10–3.12. On Linux you also need libvulkan1 — litert-lm dlopen()s a Vulkan-linked library even for the CPU backend, and without it every invocation, --help included, dies in under a second:

sudo apt-get install -y libvulkan1     # Debian/Ubuntu

macOS needs nothing extra; Colab works out of the box; Windows is untried.

Python 3.13 runs tune, prepare and bundle but not convert or verify: each stage builds its own environment from the interpreter you launched, and numpy==2.0.2 — pinned by the export toolchain — stops publishing wheels after 3.12. Past a ceiling the command refuses and names the pin that set it.

Each of tune, convert and verify builds its environment on first use and caches it. Measured on macOS:

On Linux the training environment is larger: the torch wheel pulls its CUDA dependencies, several hundred megabytes each.

litetune env shows what is on disk and litetune env --clean removes it; the next stage that needs one rebuilds it. Worth knowing because a provision that died halfway leaves a directory that looks like a working one from the outside, and because the cache key includes the interpreter — running litetune under two Pythons builds two sets.

Your data

One JSON object per line. Scoring rows need a prompt and a target; training rows add the completion text the model should produce, or let prepare derive it from the target.

The shape of the target is how you declare the task. An object with a name is a tool call:

{"prompt": "set an alarm for 7", "target": {"name": "set_alarm", "args": {"hour": "7"}}}

A bare string is the answer itself:

{"prompt": "classify the sentiment: it was fine", "target": "neutral"}

Two shapes rather than a target plus a --target-kind, because those two could disagree and a shape cannot disagree with itself. Match it with --scorer when you get to verify.

prepare splits one raw file into train.jsonl and heldout.jsonl and rejects what it cannot score: malformed JSON, and rows with no prompt. Given --tokenizer it also reports the token-length distribution, so a row too long for the sequence limit fails before you rent a GPU rather than after.

The held-out half is never trained on. Scoring a model on rows it was fitted to measures memorisation rather than whether it answers new inputs. The split is derived from the file's content hash, so re-running prepare puts the same rows on the same side.

From your data to a shippable bundle

Five commands, in order. Each is separate because each fails differently, and a single run would hide which one you are in.

# 1. Split, and reject rows that cannot be scored. Seconds.
#    Without --tokenizer it cannot measure token lengths, so it splits the file
#    and exits 4 — "could not check" — rather than implying the rows all fit.
litetune prepare --data raw.jsonl --output-dir data --context-length 1024 \
                 --tokenizer google/functiongemma-270m-it

# 2. Fine-tune. On CPU, so size your expectations accordingly.
litetune tune --model google/functiongemma-270m-it --data data/train.jsonl \
              --output-dir tuned --prompt-mode prerendered --method lora

# 3. Convert, sweeping recipes rather than trusting a default.
litetune convert --model tuned/model --output-dir artifacts \
                 --recipe dynamic_wi8_afp32 --recipe weight_only_wi8_afp32

# 4. Measure what the conversion cost, against the float twin.
#    `convert` names the artifact; look the filename up rather than build it.
litetune verify --model artifacts/weight_only_wi8_afp32/<name>.litertlm \
                --reference tuned/model --data data/heldout.jsonl \
                --json > manifest.json

# 5. Package the artifact with what was measured about it.
litetune bundle --output-dir bundle \
                --model artifacts/weight_only_wi8_afp32/<name>.litertlm \
                --declarations tools.json --prompt-mode prerendered \
                --base-model google/functiongemma-270m-it \
                --base-model-revision <commit-sha> \
                --adapter tuned/adapter \
                --train-metrics tuned/metrics.json \
                --verify-manifest manifest.json

Step 3 already gives you something shippable — one .litertlm per recipe, under artifacts/<recipe>/. Steps 1–3 are also the part most tooling makes you assemble by hand; see What it knows for what they do beyond calling the exporter yourself.

Steps 4 and 5 are what makes it trustworthy. --reference is the float twin: the same weights before conversion. That is what makes the difference between the two the conversion cost rather than a mixture of that and whatever training did. Point it at a different checkpoint — an untuned base, say — and pass --reference-role untuned_base, and both the training gain and the conversion cost come back unavailable, because one number cannot separate two effects.

If you already have a .litertlm and the checkpoint it came from, step 4 runs on its own.

Recipes

litetune knows four, and has measured two:

--recipe has no default. A sweep of one is not a comparison.

Other flags that decide something

What it knows that a shell script does not

Each of these was paid for once, by an artifact that looked fine and was not.

Export flags keyed on model identity. They are in no documentation, and config.json does not contain enough to derive them:

Without the first, FunctionGemma exports as a generic model — its config.json says gemma3_text, which the exporter does not recognise, so it falls through a silent catch-all. The runtime then builds no tool-call channel at all. An app that parses the response text sees nothing wrong; an app that passes tools natively receives no calls. The export succeeds, the file is the right size, every liveness check is green.

A trained checkpoint is told what it came from. tune writes litetune.json beside the model, and convert reads it. This is not bookkeeping: the per-family export flags key on the model's name, a directory has none, and transformers 5.x deletes _name_or_path from config.json on save — so without it a checkpoint this tool produced would export with none of the flags its family requires, successfully and silently. config.json cannot stand in: FunctionGemma and Gemma 3 270M/1B all declare model_type: gemma3_text and need different values. A checkpoint from elsewhere says so with --base-model or --train-metrics; one that says nothing is refused rather than guessed at, because guessing wrong ships a bundle with no tool-call channel that passes every check.

The prompt template has to be one the device can execute. FunctionGemma's own template uses macro and dictsort; LiteRT-LM renders with MiniJinja, which supports neither. A bundle carrying it exports cleanly, is the right size, passes every liveness check, and answers the text path flutter_gemma uses — then fails the native tool path with litert_lm_conversation_send_message_stream failed, which is the whole error the caller gets. litetune ships a template the runtime can run and passes it on export. Measured on the same checkpoint: with the override the runtime answers [tool_call] set_alarm{hour:7}; without it, INTERNAL: Failed to apply template.

The terminator comes from the chat template, not from eos_token_id. They are not always the same token, and a model trained to emit the wrong one never closes its turn — on a device it emits call after call, and a consumer that delimits the reply by the turn marker cannot find the end of one.

The adapter is saved before the merge. A merged checkpoint cannot be un-merged, so the rank-16 delta is the only form you can re-apply to a different base, inspect, or ship on its own.

tokenizer.model is carried back. transformers 5.x stopped writing it and the tokenizer classes stopped exposing vocab_file, so the exporter's SentencePiece branch never fires and the bundle silently gets an HF tokenizer section — losing FST-constrained decoding, which is SentencePiece-only. tune copies the file back and records in metrics.json whether it managed to.

Minimum transformers per family. Gemma 4 and Qwen3.5 fail at tokenizer load on every 4.x release, and Gemma 4 needs 5.5.0 for AutoConfig to recognise the architecture. litetune refuses with the version rather than letting you find out from an AttributeError.

One environment per stage. The training stack and the export toolchain pin incompatible dependencies and cannot share an interpreter.

Results

functiongemma-270m-it, LoRA on google/mobile-actions, scored on 640 examples the model never trained on:

Your gain from fine-tuning depends on your data. What this table is here to show is the last row: conversion cost something, it was small, and one of the two figures is not distinguishable from zero at this sample size.

The two artifacts are 0.04% apart in bytes. Nothing in file size, exit code or logs separates them — running both against held-out data is the only thing that does.

MEASUREMENTS.md has the intervals, three runs of the same configuration and what they disagree about, and which published claims were withdrawn after re-measurement.

Limitations

Known to be broken

• Peak memory is not bounded. Training this model on 8,693 examples was OOM-killed at 32 GiB more than once. There is no preflight check; a death with no Python traceback is probably this.

Limits on the numbers

• Measured on one model. functiongemma-270m-it. Other families export but have no quality figure.
• Measurement runs on CPU; your users run on a phone. The GPU backend is reported to score below CPU on identical artifacts, and litetune has not measured that gap, so every number here is an optimistic estimate.
• Two prompt renderings are in the field for the same model, and they disagree for every declaration with more than one property. Costly on a base checkpoint, near-free after fine-tuning; contract.json records which you used. See MEASUREMENTS.md.

Not built yet

• Decoding parameters reach only one side. litetune passes none to the device, so the reference is held to an explicit token limit while the device runs to the runtime's own. The manifest says so and counts unterminated generations.
• Evaluation is slower than it needs to be — one subprocess per prompt. A persistent litert-lm serve client is worth roughly thirtyfold.
• .litertlm only, no library API, and no single run command.

Exit codes

The verdict is the exit code; the printed summary renders it. A run that completed and a run that could not be judged must never be confused, which is why there are five and not two.

4 is not a failure of the model. It means no answer could be obtained: a missing shared library, a malformed input, a killed process — or a command line litetune would not run. That last one is why a usage error exits 4 and not argparse's usual 2: here 2 means inconclusive, which is a claim about a measurement, and a typo should not produce one.

bundle carries a verdict rather than producing one, so it returns whatever --status or --verify-manifest gave it. With neither it returns 2: bundling re-measures nothing.

Wiring || exit 1 on anything non-zero throws all of this away.

Contributing

Issues and pull requests welcome, particularly measurements on models other than the one above — that is the gap this alpha most needs closed.

Run the checks with pytest, ruff check, ruff format --check and mypy src.

License

Apache 2.0.