Remek

basal-omni-1.0-mini

README

License: apache-2.0

Results

Served with the basal engine on one H100: basal-omni-1.0-mini with --mode eager, bf16, one option order; basal-1.5 (4.5B) as released (--mode fast, both option orders). Same items and the same clients for both.

Table with columns: benchmark, basal-omni-1.0-mini (E4B), basal-1.5 (4.5B)
benchmark	basal-omni-1.0-mini (E4B)	basal-1.5 (4.5B)
Werdykt v1, macro (hidden PL + EN, 10 categories × 2 languages)	0.769	0.721
PL decisions (held-out test of basal-1.0, 7,081 items)	0.932	0.922
PL general (knowledge, exams, reading; 3,079 items)	0.751	0.746
PL-Real (300 real Polish documents)	0.950	0.963
EN external panel (10 public English decision tasks, 3,842 items)	0.740	0.688
Coverage at 1% error (share decided automatically with the shipped threshold)	42.0%¹	51.0%

Werdykt v1: 0.769 (95% interval 0.758–0.778; Polish 0.775, English 0.763), 4.8 points above basal-1.5 (4.5B) and level with basal-1.5-max (11B, 0.772), served with --deterministic (one request per forward; no difference from the default mode on our tests). Weakest categories: rules (0.352; compute exact thresholds and dates in code), contracts (0.664) and long documents (0.686).
English: it is ahead on the English external panel (+5.2 points; BIG-Bench Hard 0.726 vs 0.556, TabFact 0.780 vs 0.647, Circa 0.750 vs 0.633, RAGTruth 0.760 vs 0.653; behind on JudgeBench, 0.537 vs 0.567), and on the English corpus test of 1,479 items (0.675 vs 0.643).
Polish: it is ahead on the PL decisions test (+1.0 points, 95% interval +0.3 to +1.6), and level on Polish general knowledge (+0.4) and PL-Real (−1.3), whose intervals include zero.
Against basal-1.5-max (11B): level on the English panel (+0.3 points); behind on PL decisions (−0.8) and Polish general knowledge (−4.6).
¹ Coverage: per question type, at 0.70% observed error. A type without a certified threshold is deferred: at the 1% target, score questions go to a person. basal-1.5's 51.0% used one threshold pooled over all types; under that rule this model reaches 54.5% at 0.82% error.

Quick start

1. Install into a fresh uv environment:

bash

uv venv --python 3.12 --allow-existing ~/basal-env && source ~/basal-env/bin/activateuv pip install torch==2.11.0 torchvision==0.26.0 --index-url https://download.pytorch.org/whl/cu128uv pip install "basal[media] @ https://github.com/rkinas/basal/archive/refs/tags/v2.0.0.tar.gz"

2. Start the server. It prints basal: model ... ready:

bash

basal-serve --model Remek/basal-omni-1.0-mini --mode eager --media on --port 8000

This release runs in the engine's eager mode only (PyTorch, bf16, about 16 GB of GPU memory). The fast mode (shared prefix, CUDA graphs), the FP8 checkpoints and the MLX / GGUF ports of basal-1.5 come with a later release of this line.

3. Ask questions, exactly as with basal-1.5:

bash

curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{  "state": "Klient: od wczoraj nie mogę zalogować się do bankowości internetowej, system pokazuje błąd hasła.",  "questions": {    "dept":   {"type": "choice", "instructions": "Do którego działu skierować zgłoszenie?",               "criteria": {"cards": "Reklamacje kart", "online": "Wsparcie bankowości elektronicznej", "loans": "Kredyty"}},    "urgent": {"type": "noul", "instructions": "Czy klient nie może korzystać z usługi?"}}}'

Full API (multi, act, option keys, max_error): API.

Images and audio

With --media on, a request may carry images and audio with its state. They are placed where the state text goes, before it, in the order given; the state text may be empty.

bash

IMG=$(base64 < paragon.jpg | tr -d '\n')curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{  "state": "", "lang": "pl",  "media": [{"type": "image", "data": "'"$IMG"'"}],  "questions": {    "card": {"type": "noul", "instructions": "Czy zapłacono kartą?"},    "vat":  {"type": "choice", "instructions": "Jaka jest najwyższa stawka VAT na paragonie?",             "criteria": {"23": "23%", "8": "8%", "5": "5%", "0": "0% lub zwolniony"}}}}'

"media": a list of {"type": "image" | "audio", "data": "<base64>"} (a data: URL works too). Images: any format Pillow reads. Audio: WAV, FLAC, OGG or MP3, any sample rate (mixed to mono, resampled to 16 kHz), at most 30 s a clip.
"detail": "default" (280 image tokens) or "high" (1,120; small print, slower).
"lang": "pl" or "en", the prompt language when the state text does not show it (a state that is only media).
Limits: 8 items and 20 MB each per request, 50 MP per image.
Mixed media: one request may mix images and audio, and every question sees all of them. In our test (20 pairs of a document image and a spoken request), adding the other medium changed 5–15% of the answers. For the steadiest answers, send each question with only the media it is about.
Answers carry : the temperatures and the certified thresholds were fitted on text; validate the thresholds on your own media before automating with them.

On omni-probe-pl, our small Polish probe (each item asked from the media and from its text):

Table with columns: probe, items, from the media, from the text, untrained Gemma 4 E4B, from the media
probe	items	from the media	from the text	untrained Gemma 4 E4B, from the media
decision from an image of a rendered document	160	0.725	0.831 (the document's text)	0.500
decision about a real photo	150	0.667	0.507 (the question alone)	0.640
intent of a spoken request (MINDS-14)	140	0.921	0.943 (transcript)

Rendered documents are read 11 points behind their text (the untrained model: 0.500), and photos add 16 points over the question alone.
Speech is read well: intent and topic 2–3 points behind the transcript, spoken decisions 11 points behind (basal-omni-1.0-max, 12B: 22–42 points behind).

Speed

One question per request, one option order, on one H100 (--mode eager, bf16, --media on):

Table with columns: measurement, result
measurement	result
sequential, 200 requests	61 ms p50 / 80 ms p95 per question
32 concurrent clients, 1,000 requests	35.2 decisions/s

basal-omni-1.0-max (12B) on the same test: 66 ms p50 and 19.0 decisions/s.
On Werdykt's 5,000 items (longer states, 16 concurrent clients): 8.7 decisions/s, 67 ms p50 / 0.68 s p95 one request at a time (basal-1.5 in the basal engine's fast mode: 8.0 decisions/s).
--orders 2 costs about twice as much per question.
Long states are prefilled in 2,048-token chunks.

Files

model-*.safetensors, tokenizer, chat_template.jinja, processor_config.json: the model.
- Gemma 4 architecture (E4B: KV-shared layers, per-layer embeddings), 262k vocabulary.
- Gemma's image and audio encoders are included and unchanged (this release is trained on text); the engine reads media through them with --media on (processor_config.json).
CALIBRATION.json: per-type temperatures and certified confidence thresholds for 1% / 5% error, fitted on the served engine (--mode eager, bf16, one option order).
CALIBRATION.o2.json: the same for --orders 2 (both option orders averaged).
: prompt format 2 and the readout protocol.

How it works

The model receives a fixed chat prompt with the state, the question and coded options (A–Z, then two-character codes, up to 255). The assistant turn is prefilled with {"answer": ", and the decision is the softmax over the next-token logits of the option codes only: one forward pass, no text generation. The server applies the calibrated temperature of the question type from CALIBRATION.json.

Use the confidence. CALIBRATION.json stores thresholds for a target error of 1% or 5% among accepted decisions. They were chosen on calibration data before testing; on the test data they accept 42.0% / 77.4% of decisions at 0.70% / 3.7% observed error (at the 1% target score questions have no certified threshold and are deferred). Accept decisions above the threshold automatically and route the rest to a person.

Training

Base and method: fine-tuned from google/gemma-4-E4B-it with LoRA (rank 32) on the language-model layers. The vision and audio parts were not trained.
Data: two stages of about 815,000 and 518,000 typed decisions in Polish and English: rules and deadlines, real documents, long documents, retrieval decisions, contracts, routing, ordinal scores, judging answers, agent and tool decisions, and "cannot be determined" answers.
Distillation: basal-1.5 provides soft targets on the decisions it answers correctly.
Schedule: two stages at learning rate 3e-5: one epoch of 815,390 decisions of the basal-1.7 mix (12,741 steps of 64; without its 1,927 items longer than 6,144 tokens), then one epoch of 517,732 decisions of the basal-1.8 mix (8,090 steps of 64; without its 5,376 longer items).

Limitations

Experimental. It is the first model of a new base; validate it on your own documents before relying on it.
Images and audio are zero-shot. Gemma 4's encoders read them, but this model was trained on text decisions only; media answers are weaker than text where the media itself is hard (phone photos, speech), see "Images and audio". Their calibration is the text calibration.
Speed: the eager engine only; the fast mode, FP8 and the MLX / GGUF ports come later.
Exact arithmetic and thresholds are a weak spot of one-pass decisions: compute them in code.
Option order: one order is asked by default; --orders 2 reduces sensitivity to option order at twice the cost.
Review: decisions with serious consequences for people should be reviewed by a person.

Citation

bibtex

@software{kinas2026basal20,  title   = {basal-2.0: typed-decision models and inference engine for Polish and English},  author  = {Kinas, Remigiusz},  year    = {2026},  version = {2.0.0},  url     = {https://github.com/rkinas/basal}} @techreport{kinas2026basal,  title       = {basal-1.0: Reliable, Highly Optimized Typed Decisions for Polish},  author      = {Kinas, Remigiusz},  institution = {ai5},  year        = {2026},  type        = {Technical report},  doi         = {10.5281/zenodo.23022986},  url         = {https://doi.org/10.5281/zenodo.23022986}}

License and attribution

Apache-2.0. Fine-tuned from google/gemma-4-E4B-it (Apache-2.0).

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

Remek

Model Tree

Base

google/gemma-4-E4B-it

Fine-tuned

this model

Input Modalities

TextImage

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Results

Table with columns: benchmark, basal-omni-1.0-mini (E4B), basal-1.5 (4.5B)
benchmark	basal-omni-1.0-mini (E4B)	basal-1.5 (4.5B)
Werdykt v1, macro (hidden PL + EN, 10 categories × 2 languages)	0.769	0.721
PL decisions (held-out test of basal-1.0, 7,081 items)	0.932	0.922
PL general (knowledge, exams, reading; 3,079 items)	0.751	0.746
PL-Real (300 real Polish documents)	0.950	0.963
EN external panel (10 public English decision tasks, 3,842 items)	0.740	0.688
Coverage at 1% error (share decided automatically with the shipped threshold)	42.0%¹	51.0%

Werdykt v1: 0.769 (95% interval 0.758–0.778; Polish 0.775, English 0.763), 4.8 points above basal-1.5 (4.5B) and level with basal-1.5-max (11B, 0.772), served with --deterministic (one request per forward; no difference from the default mode on our tests). Weakest categories: rules (0.352; compute exact thresholds and dates in code), contracts (0.664) and long documents (0.686).
English: it is ahead on the English external panel (+5.2 points; BIG-Bench Hard 0.726 vs 0.556, TabFact 0.780 vs 0.647, Circa 0.750 vs 0.633, RAGTruth 0.760 vs 0.653; behind on JudgeBench, 0.537 vs 0.567), and on the English corpus test of 1,479 items (0.675 vs 0.643).
Polish: it is ahead on the PL decisions test (+1.0 points, 95% interval +0.3 to +1.6), and level on Polish general knowledge (+0.4) and PL-Real (−1.3), whose intervals include zero.
Against basal-1.5-max (11B): level on the English panel (+0.3 points); behind on PL decisions (−0.8) and Polish general knowledge (−4.6).
¹ Coverage: per question type, at 0.70% observed error. A type without a certified threshold is deferred: at the 1% target, score questions go to a person. basal-1.5's 51.0% used one threshold pooled over all types; under that rule this model reaches 54.5% at 0.82% error.

Quick start

1. Install into a fresh uv environment:

bash

uv venv --python 3.12 --allow-existing ~/basal-env && source ~/basal-env/bin/activateuv pip install torch==2.11.0 torchvision==0.26.0 --index-url https://download.pytorch.org/whl/cu128uv pip install "basal[media] @ https://github.com/rkinas/basal/archive/refs/tags/v2.0.0.tar.gz"

2. Start the server. It prints basal: model ... ready:

bash

basal-serve --model Remek/basal-omni-1.0-mini --mode eager --media on --port 8000

3. Ask questions, exactly as with basal-1.5:

bash

curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{  "state": "Klient: od wczoraj nie mogę zalogować się do bankowości internetowej, system pokazuje błąd hasła.",  "questions": {    "dept":   {"type": "choice", "instructions": "Do którego działu skierować zgłoszenie?",               "criteria": {"cards": "Reklamacje kart", "online": "Wsparcie bankowości elektronicznej", "loans": "Kredyty"}},    "urgent": {"type": "noul", "instructions": "Czy klient nie może korzystać z usługi?"}}}'

Full API (multi, act, option keys, max_error): API.

Images and audio

With --media on, a request may carry images and audio with its state. They are placed where the state text goes, before it, in the order given; the state text may be empty.

bash

IMG=$(base64 < paragon.jpg | tr -d '\n')curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{  "state": "", "lang": "pl",  "media": [{"type": "image", "data": "'"$IMG"'"}],  "questions": {    "card": {"type": "noul", "instructions": "Czy zapłacono kartą?"},    "vat":  {"type": "choice", "instructions": "Jaka jest najwyższa stawka VAT na paragonie?",             "criteria": {"23": "23%", "8": "8%", "5": "5%", "0": "0% lub zwolniony"}}}}'

"media": a list of {"type": "image" | "audio", "data": "<base64>"} (a data: URL works too). Images: any format Pillow reads. Audio: WAV, FLAC, OGG or MP3, any sample rate (mixed to mono, resampled to 16 kHz), at most 30 s a clip.
"detail": "default" (280 image tokens) or "high" (1,120; small print, slower).
"lang": "pl" or "en", the prompt language when the state text does not show it (a state that is only media).
Limits: 8 items and 20 MB each per request, 50 MP per image.
Mixed media: one request may mix images and audio, and every question sees all of them. In our test (20 pairs of a document image and a spoken request), adding the other medium changed 5–15% of the answers. For the steadiest answers, send each question with only the media it is about.
Answers carry : the temperatures and the certified thresholds were fitted on text; validate the thresholds on your own media before automating with them.

On omni-probe-pl, our small Polish probe (each item asked from the media and from its text):

Table with columns: probe, items, from the media, from the text, untrained Gemma 4 E4B, from the media
probe	items	from the media	from the text	untrained Gemma 4 E4B, from the media
decision from an image of a rendered document	160	0.725	0.831 (the document's text)	0.500
decision about a real photo	150	0.667	0.507 (the question alone)	0.640
intent of a spoken request (MINDS-14)	140	0.921	0.943 (transcript)

Rendered documents are read 11 points behind their text (the untrained model: 0.500), and photos add 16 points over the question alone.
Speech is read well: intent and topic 2–3 points behind the transcript, spoken decisions 11 points behind (basal-omni-1.0-max, 12B: 22–42 points behind).

Speed

One question per request, one option order, on one H100 (--mode eager, bf16, --media on):

Table with columns: measurement, result
measurement	result
sequential, 200 requests	61 ms p50 / 80 ms p95 per question
32 concurrent clients, 1,000 requests	35.2 decisions/s

basal-omni-1.0-max (12B) on the same test: 66 ms p50 and 19.0 decisions/s.
On Werdykt's 5,000 items (longer states, 16 concurrent clients): 8.7 decisions/s, 67 ms p50 / 0.68 s p95 one request at a time (basal-1.5 in the basal engine's fast mode: 8.0 decisions/s).
--orders 2 costs about twice as much per question.
Long states are prefilled in 2,048-token chunks.

Files

model-*.safetensors, tokenizer, chat_template.jinja, processor_config.json: the model.
- Gemma 4 architecture (E4B: KV-shared layers, per-layer embeddings), 262k vocabulary.
- Gemma's image and audio encoders are included and unchanged (this release is trained on text); the engine reads media through them with --media on (processor_config.json).
CALIBRATION.json: per-type temperatures and certified confidence thresholds for 1% / 5% error, fitted on the served engine (--mode eager, bf16, one option order).
CALIBRATION.o2.json: the same for --orders 2 (both option orders averaged).
: prompt format 2 and the readout protocol.

How it works

Training

Base and method: fine-tuned from google/gemma-4-E4B-it with LoRA (rank 32) on the language-model layers. The vision and audio parts were not trained.
Data: two stages of about 815,000 and 518,000 typed decisions in Polish and English: rules and deadlines, real documents, long documents, retrieval decisions, contracts, routing, ordinal scores, judging answers, agent and tool decisions, and "cannot be determined" answers.
Distillation: basal-1.5 provides soft targets on the decisions it answers correctly.
Schedule: two stages at learning rate 3e-5: one epoch of 815,390 decisions of the basal-1.7 mix (12,741 steps of 64; without its 1,927 items longer than 6,144 tokens), then one epoch of 517,732 decisions of the basal-1.8 mix (8,090 steps of 64; without its 5,376 longer items).

Limitations

Experimental. It is the first model of a new base; validate it on your own documents before relying on it.
Images and audio are zero-shot. Gemma 4's encoders read them, but this model was trained on text decisions only; media answers are weaker than text where the media itself is hard (phone photos, speech), see "Images and audio". Their calibration is the text calibration.
Speed: the eager engine only; the fast mode, FP8 and the MLX / GGUF ports come later.
Exact arithmetic and thresholds are a weak spot of one-pass decisions: compute them in code.
Option order: one order is asked by default; --orders 2 reduces sensitivity to option order at twice the cost.
Review: decisions with serious consequences for people should be reviewed by a person.

Citation

bibtex

@software{kinas2026basal20,  title   = {basal-2.0: typed-decision models and inference engine for Polish and English},  author  = {Kinas, Remigiusz},  year    = {2026},  version = {2.0.0},  url     = {https://github.com/rkinas/basal}} @techreport{kinas2026basal,  title       = {basal-1.0: Reliable, Highly Optimized Typed Decisions for Polish},  author      = {Kinas, Remigiusz},  institution = {ai5},  year        = {2026},  type        = {Technical report},  doi         = {10.5281/zenodo.23022986},  url         = {https://doi.org/10.5281/zenodo.23022986}}

License and attribution

Apache-2.0. Fine-tuned from google/gemma-4-E4B-it (Apache-2.0).