NeuronAI-2B

NeuronAI-2B is an Uzbek-first, bilingual assistant model built from Qwen3.5-2B-Base. It combines an Uzbek tokenizer retrofit, continued pretraining, annealing, and assistant-only supervised fine-tuning. The published weights are fully merged—no LoRA adapter is needed.

License: Apache License 2.0. Commercial and non-commercial use are permitted under the license terms. This differs from the NeuronAI-4B release, which is licensed for non-commercial use.

Quick start

Install a recent Transformers build with Qwen3.5 support:

pip install -U "transformers>=5.1" accelerate torch

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "NeuronUz/NeuronAI-2B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map={"": 0},
).eval()

messages = [
    {"role": "system", "content": "Siz foydali va aniq AI yordamchisiz."},
    {"role": "user", "content": "Alisher Navoiy haqida qisqacha aytib bering."},
]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    enable_thinking=False,
    return_tensors="pt",
    return_dict=True,
).to(model.device)

with torch.inference_mode():
    output = model.generate(
        **inputs,
        max_new_tokens=1024,
        do_sample=True,
        temperature=0.7,
        top_p=0.8,
        top_k=20,
        min_p=0.0,
        repetition_penalty=1.0,
        use_cache=True,
    )

reply = tokenizer.decode(
    output[0, inputs["input_ids"].shape[1]:],
    skip_special_tokens=True,
).strip()
print(reply)

This is the recommended quality-oriented preset for general assistant use: non-thinking mode with Qwen3.5's instruct sampling settings. Greedy decoding can cause repetition and lower response quality; reserve do_sample=False for deterministic evaluation or classification. The generation metadata already registers <|im_end|> and <|endoftext|> as end-of-sequence tokens. Keep the combined prompt and output within the validated 4,096-token serving limit.

Serve with vLLM

pip install -U vllm
vllm serve NeuronUz/NeuronAI-2B \
  --dtype bfloat16 \
  --max-model-len 4096 \
  --tensor-parallel-size 1 \
  --generation-config vllm \
  --default-chat-template-kwargs '{"enable_thinking":false}' \
  --language-model-only \
  --enable-prefix-caching \
  --mamba-block-size 16 \
  --mamba-cache-mode align

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "NeuronUz/NeuronAI-2B",
    "messages": [
      {"role": "user", "content": "O‘zbekiston haqida uchta fakt ayting."}
    ],
    "max_tokens": 1024,
    "temperature": 0.7,
    "top_p": 0.8,
    "top_k": 20,
    "min_p": 0.0,
    "presence_penalty": 1.5,
    "repetition_penalty": 1.0,
    "chat_template_kwargs": {"enable_thinking": false}
  }'

Benchmarks

All five model result sets below cover the same full eight-task suite. Classification and multiple-choice tasks use accuracy; FLORES+ translation uses COMET. The weighted score is normalized by the 0.95 sum of the published task weights. All eight NeuronAI-2B tasks completed and passed the invalid-output gate.

Benchmark	Metric	Weight	NeuronAI-2B	Qwen3.5-2B	alloma-8B	alloma-3B	alloma-1B
UzLiB	accuracy	0.20	49.60%	28.69%	42.40%	32.08%	23.32%
TUMLU-Uzbek	accuracy	0.20	32.57%	31.29%	20.71%	27.71%	22.00%
FLORES+ en→uz	COMET	0.15	0.8762	0.7010	0.8779	0.8673	0.7383
Uzbek news	accuracy	0.10	78.55%	36.75%	57.77%	13.60%	25.41%
MMLU English	accuracy	0.10	54.07%	52.39%	53.47%	38.73%	21.98%
MMLU Uzbek	accuracy	0.10	46.85%	37.10%	40.04%	32.74%	21.11%
FLORES+ uz→en	COMET	0.05	0.8535	0.8072	0.8713	0.7954	0.7636
Uzbek sentiment	accuracy	0.05	95.50%	76.87%	79.94%	38.85%	79.54%
Normalized weighted score		1.00	0.5954	0.4528	0.5187	0.4147	0.3661

Alloma runs used the APST apostrophe preprocessing required by their model cards; NeuronAI and stock Qwen did not. The alloma-8B column combines its full model-card-protocol evaluation with separately archived full UzLiB, TUMLU-Uzbek, and MMLU-Uzbek runs. Exact source files, scores, and run IDs are included in benchmark_results.json.

Run the benchmarks on your computer

The repository includes a portable Alloma-style benchmark runner. It covers FLORES+ (both directions), Uzbek sentiment, Uzbek news, MMLU English, MMLU Uzbek, and TUMLU-Uzbek.

pip install -r https://huggingface.co/NeuronUz/NeuronAI-2B/resolve/main/benchmark-requirements.txt
wget https://huggingface.co/NeuronUz/NeuronAI-2B/resolve/main/benchmark.py
python benchmark.py --limit 200 --output quick-results.json

The quick command uses the same seed on 200 examples per dataset. Run all public examples and add COMET with:

pip install unbabel-comet
python benchmark.py --limit 0 --comet --output full-results.json

Run one task when you only need a short check:

python benchmark.py --tasks mmlu-uz --limit 200 --output mmlu-uz.json
python benchmark.py --tasks flores --limit 200 --output flores.json

--limit 0 means the full dataset. Only full runs are comparable with the table above; 200-example quick runs are sanity checks. COMET downloads the Unbabel/wmt22-comet-da evaluator and needs additional disk/RAM.

Uzbek tokenizer efficiency

The tokenizer is an in-place, primarily Latin-script Uzbek retrofit rather than a vocabulary extension. The initial 20,000-document figure was measured on training-source uz-crawl, so we replaced it with a larger corpus-stratified test: 118,832 held-out-source documents plus a separate 100,000-document training-source control. Documents were selected with deterministic SHA-256 bottom-k sampling (seed 20260825), exact duplicates were excluded from the selected sample, tiny texts were filtered, and raw source text was tokenized without apostrophe normalization.

Corpus	Status	Documents	Words	NeuronAI-2B	Qwen3.5-2B	Reduction (95% CI)
Community OSCAR Uzbek	Held-out web source	100,000	7,618,770	2.0304	3.3639	39.64% (39.57–39.71%)
Uzbek legal corpus	Held-out legal source/domain	18,832	2,534,566	2.3747	2.9705	20.06% (19.55–20.57%)
uz-crawl	Training-source control	100,000	20,825,680	2.3206	3.3224	30.15% (30.02–30.30%)

Across the two held-out sources combined, the tokenizer uses 35.19% fewer tokens overall and 40.90% fewer tokens on Latin-dominant text, matching its intended Latin-Uzbek focus.

The paired intervals use 5,000 bootstrap replicates over 1,000 deterministic document buckets. OSCAR may still have incidental overlap with other public web corpora and was previously checked in a post-hoc weak-token coverage analysis, but it contributed no tokenizer-training rows. The legal corpus does not appear in the tokenizer or training source manifests and is the cleanest source-and-domain holdout in this test. Full results and script/length breakdowns: fertility_large_20260825.json and fertility_large_20260825.md.

Fertility measures tokenization efficiency—not model quality or measured decoding speed. The evaluated 2B and 4B custom tokenizer files are byte-identical, as are their evaluated stock-base tokenizer files; SHA-256 fingerprints are recorded in the JSON result.

Training

Item	Value
Parameters	1,881,825,088 (1.882B)
Prepared train examples	151,968 (152,152 source rows)
Prepared grouped dev examples	1,535 (1,537 source rows)
Train/dev prompt-group overlap	0
Sequence length / packing	2,048 / disabled
Training duration / seed	1 epoch / 42
Batch size	16 micro × 2 accumulation × 1 GPU = 32 effective
Optimizer	Fused AdamW; betas 0.9/0.95; weight decay 0.01; gradient clipping 1.0
Learning-rate schedule	Peak 1e-4; cosine decay; 142 warmup steps (2.99%)
LoRA	rank 64, alpha 128, dropout 0.05; 12 projection types; 67,276,800 trainable parameters
Loss	Fused causal-LM cross-entropy on assistant-response tokens; prompt tokens masked
Precision	bf16 training with TF32; merged embeddings and normalization tensors retained in fp32

The mixture is Uzbek-first and includes general assistant conversations, translation, Uzbek language and literature, spelling, classification, math, and English-retention examples. Training data is not distributed with this model repository.

Intended use

Good fits include Uzbek research, education, commercial and non-commercial prototyping, translation experiments, writing assistance, retrieval-augmented generation, and local/offline applications. Users remain responsible for validating the model for their application and complying with the Apache 2.0 license and applicable law.

Limitations

• This is a public-suite-selected checkpoint. The benchmark results are useful for reproducibility and relative comparison, but they are not a locked, independent estimate of real-world generalization.
• LoRA rank, learning rate, batch size, and dropout were not exhaustively swept; the table reports the released run, not globally optimal hyperparameters.
• TUMLU-Uzbek is the weakest reported Uzbek task and should not be treated as solved at 32.57% accuracy.
• The model can hallucinate, repeat biases in its data, or produce unsafe or outdated content. It has not been comprehensively safety-evaluated.
• Do not rely on it without expert review for medical, legal, financial, public safety, or other high-stakes decisions.
• SFT used sequences up to 2,048 tokens; serving at longer inherited context lengths has not been validated here. The published inference examples use 4,096 tokens.

License

NeuronAI-2B is released under the Apache License 2.0. Commercial and non-commercial use, modification, and distribution are permitted subject to its terms. This summary does not replace the license text; see LICENSE.