julia-routing-strix-halo-multilingual by Richard: Open Model

Runs On

What it takes to serve julia-routing-strix-halo-multilingual (144M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

Precision	Weights	Memory needed	Cheapest setup	Per hour	Also fits
16-bit	0.3 GB	0.3 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
8-bit	0.1 GB	0.2 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
4-bit	0.1 GB	0.1 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 1, 2026.

julia-routing-strix-halo-multilingual on every accelerator the SAVRN Index prices, at every precision

Model Card

By Richard, published under apache-2.0, revision 3893bf610310.

Experimental multilingual fine-tune of Julia-1 for fast bounded-decision routing on AMD Strix Halo-class local machines. This is an experimental v0.2 multilingual candidate, not a replacement for the English-focused v0.1 checkpoint. - task routing over fixed options - documentation-update triage - context-management metadata - non-authoritative tool-policy hints Do not use this model as the final authority for destructive commands, credential handling, production deploys, security replay safety, durable memory writes, summarization, or final prose. By language on the large multilingual holdout: All latency numbers in development were measured on CPU. NPU acceleration has not been validated.…

Read Richard's full model card

Julia Routing Strix Halo Multilingual Experimental

Experimental multilingual fine-tune of Julia-1 for fast bounded-decision routing on AMD Strix Halo-class local machines.

Runtime status:

CPU: tested
GPU: expected to work through the Julia/PyTorch runtime, not separately benchmarked for this release
NPU: not validated yet; AMD/XDNA NPU packaging and execution are future work

This is an experimental v0.2 multilingual candidate, not a replacement for the English-focused v0.1 checkpoint.

Intended use

Good candidates:

task routing over fixed options
eval-case routing: category and deterministic grader type
documentation-update triage
context-management metadata
non-authoritative tool-policy hints

Do not use this model as the final authority for destructive commands, credential handling, production deploys, security replay safety, durable memory writes, summarization, or final prose.

Training summary

Base model: SupersonicLabs/Julia-1
Starting checkpoint: julia-routing-strix-halo v0.1 tuned routing model
Fine-tuning method: native partial fine-tune, no LoRA
Trainable parameters: 7,634,563
Unfrozen modules: decision head plus last 2 encoder layers
Learning rate: 1e-5
Epochs: 1
Languages: Spanish, French, German, Portuguese, Chinese, Japanese, Arabic
Labels remain canonical English option IDs

Local eval summary

Eval	Score	Accuracy
Multilingual holdout small	`187/224`	`83.5%`
Multilingual holdout large	`574/672`	`85.4%`
English holdout	`98/126`	`77.8%`

By language on the large multilingual holdout:

Language	Accuracy
Portuguese	`93.8%`
French	`90.6%`
Spanish	`88.5%`
German	`85.4%`
Japanese	`82.3%`
Chinese	`79.2%`
Arabic	`78.1%`

All latency numbers in development were measured on CPU. NPU acceleration has not been validated.

Usage

Install the Julia runtime from the base model package, then load this checkpoint:

from julia import load_model

engine = load_model(
    "./julia-routing-strix-halo-multilingual",
    device="cpu",
    strict_encoding=True,
    max_length=1024,
    head_length=512,
)

result = engine.predict(
    state={"language": "es", "request": "Corrige la prueba TypeScript que falla y ejecuta la suite."},
    questions={
        "route": {
            "type": "choice",
            "instructions": "Which worker route should handle this request?",
            "criteria": {
                "code_edit": "modify repository files",
                "browser_qa": "interact with a web UI",
                "research": "gather information only",
                "chat": "answer conversationally",
            },
        }
    },
)
print(result)

Files

model.safetensors — fine-tuned Julia decision weights
julia_config.json — Julia runtime configuration
encoder/config.json — encoder configuration
tokenizer/ — tokenizer files

Limitations

Experimental multilingual candidate.
Specialized for bounded decisions and fixed option sets.
Not a general-purpose embedding model.
Not a generative model.
Can be confidently wrong outside the routing distributions represented in training.
NPU execution has not been validated yet.

Identity and Version

Repository: precisionailabs/julia-routing-strix-halo-multilingual
Publisher: Richard
Task: Text classification
Modality: Text
Library: julia
Parameters: 144M parameters
Languages: cpu, gpu
Revision: 3893bf610310f850f12fb878acb5279d36a86c84
First published: 2026-09-29
Last updated: 2026-09-29

Files and Weights

8 files, 611.6 MB in total. The weights are 1 file totalling 577.2 MB in safetensors.

Weights1 file · 577.2 MB

Configuration2 files · 2.1 KB

Tokenizer2 files · 34.4 MB

Documentation2 files · 4.8 KB

Repository1 file · 1.6 KB

Every file

File	Type	Size	SHA-256
model.safetensors	Weights	577.2 MB	358ddc10d57b
encoder/config.json	Configuration	1.9 KB	—
julia_config.json	Configuration	147 B	—
README.md	Documentation	3.5 KB	—
UPLOAD.md	Documentation	1.2 KB	—
.gitattributes	Repository	1.6 KB	—
tokenizer/tokenizer.json	Tokenizer	34.4 MB	609d8f4c067c
tokenizer/tokenizer_config.json	Tokenizer	652 B	—

License and Download

License: apache-2.0
Access: Open weights, no gate
Download size: 577.2 MB

Download from Richard

Released by Richard through its official repository on Hugging Face. Read the license.

Built From

Derived from SupersonicLabs/Julia-1

Memory Requirements

Precision	Weights in memory
As published	577.2 MB
16-bit	0.3 GB
8-bit	0.1 GB
4-bit	0.1 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About julia-routing-strix-halo-multilingual

How much GPU memory does julia-routing-strix-halo-multilingual need?

About 0.3 GB at 16-bit and 0.1 GB at 4-bit: the weights (144M parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run julia-routing-strix-halo-multilingual on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use julia-routing-strix-halo-multilingual commercially?

Yes. julia-routing-strix-halo-multilingual is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Similar Models

Model · Text classification

Julia-1-MLX

Zain Merchant

Julia-1 for Apple silicon. Runs Supersonic Labs' Julia-1 decision model on the Mac's GPU with MLX through the julia-mlx runtime: the same answers as the official PyTorch runtime on its published evaluations, 5–14× faster on the same Mac. The upstream model repository is SupersonicLabs/Julia-1. The files in this repository are Supersonic Labs' Julia-1 checkpoint, unchanged (model.safetensors SHA-256 df853bf7fe424420011f3d0c47a05d7341aa9eefa7fb9f203ea4aada4ad95b72). The runtime maps it into MLX directly, so no converted copy is needed. Precision (dtype="float16") and embedding placement are load-time options rather than separate files. This is an independent project, not affiliated with or…

Open weights apache-2.0 144M parameters mlx

View model

Model · Text classification

prompt-injection-guard-small

Horizon Labs

A fast, multilingual classifier that flags prompt injection and jailbreak attempts, both in user messages (direct) and in untrusted content an AI agent reads: emails, web pages, documents, RAG chunks, and tool/API outputs (indirect). their clean counterparts, so it looks for instructions aimed at the AI, not for scary words. - Low false-alarm rate on look-alike benign text: 89.7% on NotInject, 99.5% on OR-Bench-hard. half the size and the same decisions as fp32 on our checks), transformers.js. Labels: SAFE (0) and INJECTION (1). This is the same convention as protectai/deberta-v3-base-prompt-injection-v2, so the model is a drop-in replacement in code and tools built for that one. INJECTION…

Open weights apache-2.0 141M parameters 8,192 tokens transformers

View model

Model · Text classification

mailroom-modernbert-classifier

Lucius Morningstar

Hierarchical document classifier for the LLM-Mailroom intake pipeline: a fine-tuned ModernBERT-base encoder with a doctype head plus one subclass head per document class. It is the deterministic pre-check in the BERT-coupled intake overhaul (mailroom-issues #85). 8,192-token context, bf16. heads (contract, corporaterecord, correspondence, insuranceclaim, mergeragreement). MLP heads with dropout 0.1. by plurality vote over windows; subclass by plurality over windows whose doctype vote is the winning class. doctype (6): contract, mergeragreement, corporaterecord, correspondence, insuranceclaim, unknown (inference-only abstention — not a trained class) development, distributor, endorsement…

Open weights apache-2.0 149M parameters 8,192 tokens transformers

View model

Model · Text classification

Firebird-ModernBERT-512-RW

Noumenon, Inc.

Firebird-ModernBERT-512-RW is an experimental post-trained variant of It is a ~149M parameter ModernBERT binary classifier for distinguishing: - 0 — HUMAN The maximum sequence length is 512 tokens. This checkpoint was produced through reward-weighted classifier post-training. The original Firebird checkpoint was kept frozen as a reference model. Training examples were scored by the original classifier, difficult examples received larger loss weights, and the post-trained model was constrained against the frozen reference using a KL penalty. L = weightedcrossentropy + beta KL(reference || policy) Hard human examples received greater weighting than ordinary examples because one goal of the…

Open weights 150M parameters 8,192 tokens transformers

View model

Model · Text classification

pyrrho-v2-nano-g1

Yan Fitzner

Pyrrho is a CPU-runnable co-processor for retrieval-augmented generation. It reads a question before retrieval to suggest what evidence to seek, then reads the question with retrieved passages to assess whether those passages support an answer. A surrounding RAG runtime decides whether to answer, retrieve again, or surface a conflict. The broader project is described in the The evidence verdict is SUFFICIENT, DISPUTED, or INSUFFICIENT. These are predictions about the supplied passages. The model does not retrieve sources, generate answers or citations, check external facts, or prove that a corpus has been searched completely. The two passes use the same encoder with different input…

Open weights cc-by-nc-4.0 150M parameters 8,192 tokens transformers

View model

Model · Text classification

distilbert-base-multilingual-cased-sentiments-student

Lik Xun Yuan

This model is distilled from the zero-shot classification pipeline on the Multilingual Sentiment dataset using this script. In reality the multilingual-sentiment dataset is annotated of course, but we'll pretend and ignore the annotations for the sake of example. Result can be reproduce using the following commands: If you are training this model on Colab, make the following code changes to avoid Out-of-memory error message: - Transformers 4.28.1 - Pytorch 2.0.0+cu118 - Datasets 2.11.0 - Tokenizers 0.13.3

Open weights apache-2.0 135M parameters 512 tokens transformers

View model