Open-Weight Models Under Apache License 2.0

Model

Qwen-Image-2.1-BF16-mthreads-FlagOS

FlagRelease

FlagOS is a fully open-source system software stack for heterogeneous AI chips. It unifies the model–system–chip layers to enable a "develop once, run anywhere" workflow, eliminating the fragmentation among vendor-specific software stacks and substantially lowering the cost of porting AI workloads across accelerators. In this release, Qwen-Image-2.1 leverages the FlagOS software stack to provide direct multi-chip support. By integrating the Triton-based operator library FlagGems via the Torch-FL plugin, FlagOS enables seamless adaptation of the Diffusers library across chip platforms; the usage experience remains identical to that on NVIDIA, requiring zero code modifications. Inference…

This is a historical merged export, NOT the current flagship. The flagship is v1.9a — adapter + merge instructions at qwen3-vl-8b-hebrew-v19a-ckpt. Full 16-bit merge of v1.8b step 700 — the first generation whose vision tower actually trained (patch-embed gradient fix), and the warm-start ancestor of all v1.9 models. 6.5 MP resolution policy ships in the processor config. Superseded by v1.9a for all use. One line per generation — which checkpoint to use and which are historical: \\ CER over substantive attempts only, alignment-based scorer. Benchmarks differ across generations (Talmud print vs Genizah manuscripts) — compare within a row's named benchmark, not across rows.

This repository contains the selected PEFT LoRA adapter for SajiloFit AI, an English-language fitness and general nutrition assistant built on top of Qwen/Qwen2.5-7B-Instruct. This is an adapter-only repository. It does not contain the complete Qwen2.5-7B model weights. To use it, load the Qwen base model first and then attach this adapter with PEFT. During QLoRA training, the base model is loaded in a quantized form to reduce GPU-memory use while the LoRA parameters are trained. The uploaded artifact is a LoRA adapter and can be attached to either a normal BF16/FP16 base model or a compatible 4-bit base-model load. The adapter is intended to improve Qwen's behaviour for SajiloFit tasks…

by AGK FIRE INC A QLoRA fine-tune of Qwen/Qwen3-8B that tutors for the SAT (math + reading & writing). Trained on ~17k SAT-style examples (60% math, 40% reading & writing), one epoch, 2048-token training length, with full step-by-step solutions plus hint-style answers. 160 held-out questions (100 math, 60 reading & writing). Identical prompt and decoding for both: greedy, 400 new tokens, thinking off, same Answer: X format instruction. Strict scoring = unparseable counted wrong. Reports: evalbase.json, evaltuned.json. The tuned adapter also answers far more math questions (97/100 vs 71/100) — base adapter overfit toward math. More RW training data is recommended before the teacher is…

Model · Image and text to text

GameScaling

Tsinghua Spatial Intelligence Lab

GameScaling models are Qwen3.5 vision-language models with continued pre-training on keyboard-and-mouse gameplay. Each call takes the current game frame (plus a short history of past frames and actions) and returns the next 200 ms of keyboard and mouse actions: 6 steps of 33 ms each, plus one mouse movement and scroll for the whole chunk. The model may think before it acts. This repository contains the inference client and the standard system-prompt template; each of the eight checkpoints is its own repository (below, and in the GameScaling collection). Together they are everything needed to serve a checkpoint and run it in a game; there is no training code. Paper results and analysis are…

Model · Audio classification

ls_conformer_small_ge2e

Quan

model.safetensors: Primary model weights in SafeTensors format - flaxmodel.msgpack: Native Flax Msgpack serialized parameters - config.json: FlaxSpeakerConfig model & training architecture configuration - preprocessorconfig.json: WhisperFeatureExtractor-compatible frontend configuration - evalresults.json: Verified multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs) - modelfp32.tflite: Exported FP32 TensorFlow Lite model for edge/mobile CPU inference - modelint8.tflite: Dynamic-range INT8 quantized TensorFlow Lite model (~3.5x smaller)

Model · Image and text to text

Qwen3.5-4B-onwx

Ryu Gyo

Alibaba の Qwen3.5-4B を、onwx 用に変換したものです。onwx は LLM を AMD Ryzen AI の NPU（XDNA2）だけで 動かすアプリで、このリポジトリにはモデルと、このモデル専用の NPU プログラムが入っています（アプリは別にインストールします）。 重みは int4（128 個ごとのスケールと最小値つき）で NPU に載せ、行列計算・DeltaNet・アテンション・LM ヘッドまで 1 トークンの計算をすべて NPU の 32 個の AI エンジンで行います。画像のエンコーダー（ViT）は CPU で動きます。NPU 向けのプログラムはビルド済みなので、PC の上でコンパイルを待つことはありません。 （onwx 0.5.0、Ryzen AI Max+ 395（Windows）で NPU の電力モードを performance にして測定。Ryzen AI 300 シリーズ（HX 370 など）も NPU は同じ 32 個の AI エンジンで、同じ方式で動きます。標準の電力モードでは NPU のクロックが下がって 2 割ほど遅くなります。電力モードは管理者の PowerShell で C:\Windows\System32\AMD\xrt-smi.exe configure --pmode performance。Windows の Copilot+ の AI 機能が NPU を使っていると遅くなることがあります） AMD Ryzen AI の NPU で LLM を動かす代表的なアプリ FastFlowLM（FLM）1.0.8…

Model · Zero-shot classification

siddhanto-bangla-english-banglish

Kazal Chandra Barman

Siddhanto (সিদ্ধান্ত, "decision") is a fine-tune of convaiinnovations/laya-multilingual (mmBERT-base, 322M params) for open-domain typed decisions over Bangla, English and Banglish (romanized Bangla / code-switched) conversational text. It is not a fixed-label classifier. It implements Laya's typed choice / noul / score interface: at request time you supply the question and its options (any schema, any domain), and the model returns a calibrated probability distribution over exactly those options — no retraining or head-swapping needed to add a new label set. Fine-tuned on ~268k typed-decision examples drawn from: - An internal 56-intent OTA (travel-agency) synthetic corpus…

Model · Text generation

StandardOne-8B-LoRA

Standard Thinking

Standard One scores a bounded set of answers for a supplied scenario and returns probabilities through POST /v1/systemone. It does not generate free-form response text. This repository contains the 8B LoRA adapter and a merge recipe; serving requires a merged checkpoint and the server code. In the reported served evaluations, 8B scores higher than 3B on the public standard and hard tiers; 3B has a lower median latency on the measured short-request profile. See Benchmarks for the measurement conditions and limitations. The figure combines results from different measurement paths. See Benchmarks for served versus offline conditions; measured 24–26 September 2026. - Send a state and a bounded…

This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

Model · Text generation

Darwin-397B-JGOS

FINAL_Bench

Darwin-397B-JGOS is the largest and highest-scoring member of the Darwin family. Built on Qwen 3.5 397B as the base, it transplants the FFN (expert) strengths of multiple high-performance models through the Darwin V9 platform, producing a 397B-parameter Mixture-of-Experts model with ~17B active parameters per token. It reaches 90.9 % on GPQA Diamond with pure greedy decoding (single sample) — surpassing Darwin-28B-REASON (89.39 %, achieved with the Darwin-DELPHI test-time engine) without using any test-time engine at all. This is the highest GPQA Diamond score in the Darwin family to date. Darwin is VIDRAFT's measuring-result-driven reasoning model family — approximately 20 official models…

Model · Reinforcement learning

DiscoDemo-Stage1_RL-FMB_Round-alpha0

DAVIAN Robotics

The data generator of DiscoDemo for the FMB-Round task (insert a cylindrical peg into its hole on the FMB board): a state-based reinforcement-learning policy trained in simulation — the P-RFCL baseline generator (α = 0: the same reverse-curriculum RL without the diversity reward). Rolling it out generates demonstrations such as those in the Stage2 dataset below. Only what is needed to roll the policy out is released; the critic and other training-only state are omitted. Loading requires the DiscoDemo training code: https://github.com/DAVIAN-Robotics/DiscoDemo.

Model · Audio classification

ls_ecapa_tdnn_small_ge2e

Quan

model.safetensors: Primary model weights in SafeTensors format - flaxmodel.msgpack: Native Flax Msgpack serialized parameters - config.json: FlaxSpeakerConfig model & training architecture configuration - preprocessorconfig.json: WhisperFeatureExtractor-compatible frontend configuration - evalresults.json: Verified multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs) - modelfp32.tflite: Exported FP32 TensorFlow Lite model for edge/mobile CPU inference - modelint8.tflite: Dynamic-range INT8 quantized TensorFlow Lite model (~3.5x smaller)

Comparable local UI-check performance to GPT-6 Luna Instant, with lower observed latency and estimated cost. UI-TestJev uses DiffusionGemma to compare approved and current screenshots, identify UI defects, and return structured decisions for testing pipelines: pass, unknown, or a failure category. The release is a LoRA adapter for DiffusionGemma-26B-A4B-it, using one-step decision inference. Vision weights remain BF16; frozen expert weights use INT8. Fine-tuning improves local-check macro-F1 by 41.70 percentage points and full-screen regression by 19.98 points. Base DiffusionGemma is evaluated without the adapter, using the same INT8-expert/BF16-vision setup, screenshots and decision…

Model · Reinforcement learning

DiscoDemo-Stage1_RL-FMB_Round

DAVIAN Robotics

The data generator of DiscoDemo for the FMB-Round task (insert a cylindrical peg into its hole on the FMB board): a state-based reinforcement-learning policy trained in simulation — the DiscoDemo generator (diversity weight α = 0.3). Rolling it out generates demonstrations such as those in the Stage2 dataset below. Only what is needed to roll the policy out is released; the critic and other training-only state are omitted. The actor is conditioned on a skill vector z sampled once per episode from a standard normal distribution. Loading requires the DiscoDemo training code: https://github.com/DAVIAN-Robotics/DiscoDemo.

Model · Reinforcement learning

DiscoDemo-Stage1_RL-FMB_SqCircle

DAVIAN Robotics

The data generator of DiscoDemo for the FMB-SqCircle task (insert a two-pronged (cylinder and square) peg into its hole on the FMB board): a state-based reinforcement-learning policy trained in simulation — the DiscoDemo generator (diversity weight α = 0.3). Rolling it out generates demonstrations such as those in the Stage2 dataset below. Only what is needed to roll the policy out is released; the critic and other training-only state are omitted. The actor is conditioned on a skill vector z sampled once per episode from a standard normal distribution. Loading requires the DiscoDemo training code: https://github.com/DAVIAN-Robotics/DiscoDemo.

Model · Reinforcement learning

DiscoDemo-Stage1_RL-FMB_SqCircle-alpha0

DAVIAN Robotics

The data generator of DiscoDemo for the FMB-SqCircle task (insert a two-pronged (cylinder and square) peg into its hole on the FMB board): a state-based reinforcement-learning policy trained in simulation — the P-RFCL baseline generator (α = 0: the same reverse-curriculum RL without the diversity reward). Rolling it out generates demonstrations such as those in the Stage2 dataset below. Only what is needed to roll the policy out is released; the critic and other training-only state are omitted. Loading requires the DiscoDemo training code: https://github.com/DAVIAN-Robotics/DiscoDemo.

Model

leviathan-10.5b-final

Lin

這是由 wangguo-2026 開發與封裝的利維坦核心模型 (Leviathan Core)。本模型經過嚴格的監督微調 (SFT)，針對繁體中文邏輯互動與任務執行進行了深度優化。 架構: Llama (Base) 精確度: bfloat16 你可以透過 Hugging Face 函式庫輕鬆載入此模型： 內建 Chat Template，支援標準對話格式。 本模型遵循 Apache 2.0 授權協議。

Model · Text classification

Open-Decision-DeBERTa-v3-Large-LiteRT

LiteRT Community (FKA TFLite)

com-kotobalabs/open-jev-deberta-v3-large is a decision model built on DeBERTa-v3-large. It reads one state and several typed questions (choice, score, noul) and returns a calibrated probability distribution for each question from one forward pass. This repository holds it converted to LiteRT with litert-torch 0.9.3. The verified path on desktop is the LiteRT CompiledModel Python API on CPU and on the Apple Metal GPU with explicit FP32 computation (ai-edge-litert 2.1.6, Apple M4 Max, 2026-10-01). On a phone, it is the Kotlin CompiledModel API on the Galaxy S26 GPU with explicit FP32 computation (LiteRT 2.2.0, 2026-10-01 and 2026-10-02). On desktop CPU, the shipped graph (512-token window…

Built with Qwen. Qwen/Qwen2.5-1.5B-Instruct cuantizado a INT4 con (2.34x mas chico) perdiendo 2.11% de calidad. Incluye las dos formas de usarlo: qwen1.5b-int4.gguf para llama.cpp y Android, y la carpeta.tq para PyTorch. Perplejidad en wikitext-2 test, 20 ventanas de 2048 tokens, todo medido dentro de llama.cpp con llama-perplexity para que la comparacion sea de tu a tu. Este modelo pierde 2.2 veces menos calidad que llama.cpp Q4KM, el formato mas usado para correr modelos en local, a cambio de 0.34 GB mas de archivo. La comparativa de los tres tamaños esta en En PyTorch el modelo cuantizado genera mas lento que el original: Octuma desempaqueta los 4 bits en cada multiplicacion sin un…

Model · Zero-shot classification

Vela-2.0-4B

vLLM Semantic Router

Open Foundation Routing Models Routing decisions. Safety checks. Precise text spans. The balanced hybrid member of Vela 2.0: strong multilingual safety and general decisions, with router and open-label spans through one interface. Define options, labels and rubrics at request time. Ask multiple named questions about a request, context and answer, and receive structured decisions with the text spans that support your workflow. 1. Balanced routing capability. Safety macro AUC of 0.921 across 14 public sets, alongside open-label extraction and general decisions. 2. Routing and safety together. Use one request for routing, prompt-attack checks, PII and unsupported-claim detection. 3. Decisions…

Model · Text classification

counterbench-bert-base-uncased

Arnav Verma

This model is a fine-tuned version of bert-base-uncased on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 16 - evalbatchsize: 32 - lrschedulertype: linear - lrschedulerwarmupsteps: 100 - numepochs: 5 - Transformers 5.17.0 - Pytorch 2.14.0+cu130 - Datasets 3.6.0 - Tokenizers 0.23.2

Model · Translation

Index-Translate-35B-A3B-preview

Index Team

Index-Translate-35B-A3B-preview is the 35B-total / 3B-active mixture-of-experts multilingual text model of the Index-Translate family, built on Qwen3.5. It supports text translation across 150 languages and follows translation instructions involving terminology, formatting, style, structure, context, and output length. Preview: all 35B-A3B results below refer to the preview model evaluated in the technical report. - Has the highest FLORES COMET-22 (0.8794) and instTrans IFscore (0.8336) among all comparison systems in the report. - Scores 0.6901 instTrans quality, 76.76 WMT26 Judge, and 0.7405 MEME. - Among the three Index-Translate models, has the highest low-resource FLORES COMET-22 /…

Model · Text generation

KeyVoice-Refiner-2B-v0.1

岡優志 / Okayuji

A text-refinement model for cleaning up speech-to-text output: punctuation, polite-form conversion, translation, summarization, and filler-word removal. This repository contains a complete Q4KM GGUF text model, ready for a compatible runtime. For: developers adding local text cleanup, or Mac users who want to try it in KeyVoice. Not for: general chat, code generation, mathematical reasoning, or long-form creative writing. Platform: GGUF runtimes with Qwen3.5 support; the original performance report used an Apple Silicon M4 Mac. Use KeyVoice for Mac for the app experience. For your own integration, download the file with the Hugging Face CLI: Run it with llama.cpp using its CLI instructions.…

static quants of https://huggingface.co/jasonliaonk21/Qwen2-VL-ViDA For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion. If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality quant…

Unless you are studying this series' training history, use the flagship instead: qwen3-vl-8b-hebrew-v19a-ckpt. Status: superseded — A/B control arm. v1.8 was a controlled experiment on whether training the vision tower matters for handwriting: v1.8a kept the vision LoRA frozen (the accidental status quo of v1–v1.7, here made deliberate), while its twin v1.8b trained it. v1.8b won the comparison and became the ancestor of every later checkpoint. Keep this repo for reproducing the ablation; do not use it for inference. One line per generation — which checkpoint to use and which are historical: \ CER over substantive attempts only, alignment-based scorer. Benchmarks differ across generations…

Model · Zero-shot classification

Vela-2.0-9B

vLLM Semantic Router

Open Foundation Routing Models Routing decisions. Safety checks. Precise text spans. The largest hybrid member of Vela 2.0: high-accuracy multilingual routing, long-document PII and hallucination detection, with open-label spans through one interface. Define options, labels and rubrics at request time. Ask multiple named questions about a request, context and answer, and receive structured decisions with the text spans that support your workflow. 1. High-accuracy decisions. A 41.63 Jev Decision Index with optional Noul calibration, plus 0.989 AUC on unseen prompt-attack families. 2. Routing and safety together. Use one request for routing, prompt-attack checks, PII and unsupported-claim…

This model is a fine-tuned version of distilbert/distilgpt2 on an unknown dataset. The following hyperparameters were used during training: - learningrate: 0.0005 - trainbatchsize: 32 - evalbatchsize: 32 - gradientaccumulationsteps: 8 - totaltrainbatchsize: 256 - lrschedulertype: cosine - lrschedulerwarmupsteps: 1000 - numepochs: 1 - mixedprecisiontraining: Native AMP - Transformers 4.57.1 - Pytorch 2.8.0+cu126 - Datasets 4.0.0 - Tokenizers 0.22.1

A quantized version of Qwen/Qwen3.8-27B for Jetson AGX Orin 64GB / CUDA SM87, deployed with Orinfer v0.2.0. It supports text, images and multiple images, streaming Chat Completions, native MTP, prefix caching, continuous batching and resumable JSONL offline inference. Translation requests use enablethinking: false; thinking follows the original model's chat template. Two-stage residual E8P Q4 with coefficients [4, 1], signed-block128 rotation and beam-8 encoding over the original BF16 weights, paired with INT8 computation. Sensitive tensors retain the validated W8/FP16/BF16/FP32 mixed-precision policy. Deployed read-only GPU data, including scales, derived constants and precision…

TimeBraid: Unifying Time Series and Language for Understanding and Forecasting TimeBraid combines numerical time series and natural language in one model. It can explain an observed series in words and predict future values using both its history and relevant text. - Understand time series. Describe trends, answer questions about temporal patterns, and compare multiple series in natural language. - Forecast with context. Combine observed values with text about events or conditions to produce numerical forecasts. - Use one interface. Supply numeric arrays and messages through the Hugging Face processor; choose text output or a forecast for one selected series. The checkpoint includes both…

A quantized version of IndexTeam/Index-Translate-2B for Jetson AGX Orin 64GB / CUDA SM87, deployed with Orinfer v0.2.0. It supports text, images and multiple images, streaming Chat Completions, native MTP, prefix caching, continuous batching and resumable JSONL offline inference. Translation requests use enablethinking: false; thinking follows the original model's chat template. Two-stage residual E8P Q4 with coefficients [4, 1], signed-block128 rotation and beam-8 encoding over the original BF16 weights, paired with INT8 computation. Sensitive tensors retain the validated W8/FP16/BF16/FP32 mixed-precision policy. Deployed read-only GPU data, including scales, derived constants and…

A Jev-style decision model that reads text, images, video, and audio. The readout is marker. LoRA (r=32, alpha=64) is trained on the thinker of Qwen/Qwen2.5-Omni-3B. Only the thinker is loaded; the speech output is not. Training finished on 2026-10-04. The training config is configs/av/omni3bv4.yaml in jevtrainer: 79 text, image, audio, and video datasets, one epoch. The holdout has 1,610 examples, accuracy 85.03%, skill 79.54%, ECE 0.0096. Accuracy in percent. This repo is the LoRA adapter and the readout. Base weights still come from Qwen/Qwen2.5-Omni-3B. readout.json records the base model id.

Model · Text classification

kev-4b

Jared Palmer

kev-4b is a decision model: one document (the state) and a set of typed questions in, a probability distribution per question out, in one forward pass. No text generation. It is a LoRA adapter (r=16) plus a pointer head on Qwen/Qwen3-4B-Base, serving TypeSafe's public /v1/systemone contract. Research preview, not a versioned release. It is the best 4B checkpoint under a frozen, checksummed protocol after ~40 controlled 4B trials, and the first kev whose out-of-domain accuracy is within ten points of Jev on the same items. This checkpoint clears the release screen we set in advance (held-out policy pairs 0.73 both-correct, screen 70%), but the other two seeds of the same recipe do not (0.62…

Documentation only; weights remain v0.2.0. No improved financial checkpoint is released. ScenarioView report · Validation notes. MiDM baseline action accuracy was 54.37%; tested residual variants did not justify replacement. These are reused retrospective KOSPI decisions, not live signals or general benchmark gains. Documentation update; this repository's existing weights and inference code are unchanged. atlas separates matched measurements from historical/published/API comparisons. JevBench public accuracy is not its official composite. Missing MiDM DeepSWE and local Terminal-Bench measurements remain explicit; their candidate pools were not found. Unknown proprietary parameter counts are…

This is a historical merged export, NOT the current flagship. The flagship is v1.9a — adapter + merge instructions at qwen3-vl-8b-hebrew-v19a-ckpt. Full 16-bit merge of the series' first-generation (v1) Talmud adapter — the first model in this line to read Vilna-print gemara from pixels. Vision tower unadapted (language-only LoRA generation). Superseded on every (rope under textconfig.ropeparameters); MLX conversion requires patching ropetheta to top level. One line per generation — which checkpoint to use and which are historical: \\ CER over substantive attempts only, alignment-based scorer. Benchmarks differ across generations (Talmud print vs Genizah manuscripts) — compare within a…

Model · Image and text to text

REX-Intern-S2-Bio-F3

AI45Research

A REX checkpoint of Intern-S2 with WMDP-Bio knowledge relocated into 8 removable experts per MoE layer (264 experts in total). Excising these experts together with their router rows yields A3, a standard 256-expert model with that knowledge removed. This repository holds the expanded checkpoint F3: the REX model after 250M tokens of standard full-parameter SFT, with the removable experts still attached. A3 differs from F3 only by excision. CPU only and about 1 min, with at least 16 GiB RAM. Disk: 67 GiB for F3 plus 65 GiB for A3../A3 must not exist beforehand, and expertexpansionmanifest.json must stay in./F3. Load./A3 with the model class named in its config.json. Intern-S2 ships a custom…

Alien SI antrenat pentru tool-calling: emite {"name":..., "arguments":...}, primeste, continua. Tool-uri: runnmap, runmsf, runsqlmap, runhydra, runhashcat, runimpacket, readfile, writereport, runverilogsim, runshell. Etapa A din lanțul ULTRON (A tool-use -> B multi-agent -> C self-play).

The current v2.2 model was trained on the pre-sealed 3,660-trajectory protocol. The additional 128 trajectories are evaluation-only and do not change the weights. The unchanged capability-v4 weights were evaluated once on 128 trajectories published first at dataset revision The eight sealed path families have no family-name overlap with training, and the repository audit found zero exact cross-role collisions in full model inputs or input/target pairs. Across 92,928 states, micro RMSE is 0.509 MPa and R2 is 0.999994. The trajectory-macro bootstrap 95% interval is 0.432--0.543 MPa; median, p95 and maximum per-trajectory RMSE are 0.324, 1.184 and 1.359 MPa. The maximum return- mapping yield…

Model · Token classification

gliner-id

Muchad

GLiNER-ID is an Indonesian Named Entity Recognition (NER) model based on GLiNER, fine-tuned from muchad/mdeberta-hybrid-30k. The model uses the 30k-token hybrid Indonesian–English vocabulary developed through VocabPrune and is designed for flexible, span-based NER in Indonesian text. The model is a fine-tuned NER checkpoint and can recognize user-defined entity types without requiring a fixed predefined label set. Load the model and define the entity types you want to extract: For the methodology, training data, experimental setup, and detailed evaluation results, please refer to the published paper. If you use GLiNER-ID in your research, please cite

A quantized version of IndexTeam/Index-Translate-9B for Jetson AGX Orin 64GB / CUDA SM87, deployed with Orinfer v0.2.0. It supports text, images and multiple images, streaming Chat Completions, native MTP, prefix caching, continuous batching and resumable JSONL offline inference. Translation requests use enablethinking: false; thinking follows the original model's chat template. Two-stage residual E8P Q4 with coefficients [4, 1], signed-block128 rotation and beam-8 encoding over the original BF16 weights, paired with INT8 computation. Sensitive tensors retain the validated W8/FP16/BF16/FP32 mixed-precision policy. Deployed read-only GPU data, including scales, derived constants and…

Model · Text classification

Kev-4B-LiteRT

LiteRT Community (FKA TFLite)

jaredpalmer/kev-4b is a decision model by Jared Palmer. It reads a text, called the state, and typed questions about it: yes/no (noul), multiple choice (choice) and rating (score). For each question it returns an answer with probabilities. It never generates text. A pointer head of two linear layers reads the hidden states of Qwen/Qwen3.5-4B-Base, adapted with a rank-16 LoRA, and turns them into those probabilities. Requests and responses follow the /v1/systemone shape of the author's server (github.com/jaredpalmer/kev). This repository holds the backbone, with the LoRA folded into the weights, as eight LiteRT files. Six are row graphs, for rows of up to 64, 128, 256, 512, 1,024 and 2,048…

Unless you are studying this series' training history, use the flagship instead: qwen3-vl-8b-hebrew-v19a-ckpt. Status: superseded. The first generation trained on real Cairo Genizah handwriting (documentary fragments) alongside the Talmud-print and synthetic material — the series' pivot from print to manuscripts. open VLM we measured on both our Talmud-print and Genizah-manuscript benchmarks. v1.8b). Later generations (v1.8b → v1.9a) supersede it on every manuscript benchmark. One line per generation — which checkpoint to use and which are historical: \ CER over substantive attempts only, alignment-based scorer. Benchmarks differ across generations (Talmud print vs Genizah manuscripts)…

A 35M-parameter token tagger for personal data in Chinese and mixed Chinese–English text. It is a fine-tune of cjhwong/seatbelt-pii-tagger with the same architecture, tokenizer and 18 types / 73 BIOES labels. - real Chinese contexts (FineWeb-2) with inserted synthetic private names and addresses; - self-distillation replay, which aims to reduce forgetting of the base model's other types and its English behaviour. - It misses fewer Chinese person names. On test sets never used during development, person recall rises by about 2 to 10 points; precision moves between −3.5 and +3.5 points. - +0.6 to +2.1 on two freshly generated test sets; - +4.2 to +6.7 on an MSRA NER train sample (never used…

Model · Question answering

basic-chat-model

Lenin Villanueva

This modelcard aims to be a base template for new models. It has been generated using this raw template. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Model · Feature extraction

Corpus-ViT

Gugu8

A vision encoder with a runtime-selectable token budget, built to sit between an image and an LLM. A coarse 32px grid covers the whole image; a router spends extra 16px detail only on the regions that need it. Output length is 49 + 3k tokens — 49 at k=0, 121 at k=24 — so you choose the tokens/fidelity trade-off per image, at inference, with no retraining. 24.94M parameters. Trained in 6000 steps (~30 min) on 2× Kaggle T4 by distillation from google/siglip-base-patch16-224 (86M). This is a working, honest baseline, not a competitive encoder. On the Flickr30k official test split it reaches 8.95 mean R@1 against the teacher's 83.25 — roughly 10% of its teacher. Chance on 1000 images is 0.10.…

Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

A fine-tuned Gemma 4 E2B for Laya-Heart, a local web app that helps you reply in a couple's chat. Given the chat and Laya's reading of it, the model writes two candidate replies, one per line. You choose, edit and send one yourself. 297 fictional practice-exam chats from the project's train-kit (253 Chinese, 44 English). Both models drafted every chat the way the app does. The judge, Qwen3-32B-AWQ, did not know which was which and judged every pair twice with the sides swapped; ties count as half a win. The flags come from the app's own checks. They are rule-based, not reviewed by a person. Self-check: the trainer's self-check (50 chats, nine gates) passed eight gates and missed one: judge…

Model · Text classification

Quantum-Classifier-Calibrated

Kshitij Vinayaksing Rajput

Nebulixlabs/Quantum-Classifier 0 = safe 1 = unsafe The held-out test split, where available, is evaluated only after fine-tuning.

Unless you are studying this series' training history, use the flagship instead: qwen3-vl-8b-hebrew-v19a-ckpt. reference. The full-page generation: page-level training at the 6.5 MP resolution policy the series has used ever since. Talmud page-level strict CER means (65-page benchmark): gemara 0.090 / rashi 0.047 / tosafot 0.099 — the Rashi-script column 6.7× better than the strongest closed model we evaluated (0.315). - Step 1100, b9f47f32e9ed51eb45598e27ffb5c5b650f7723c (run end): the v1.7 warm start. - Medians across checkpoints were ceiling-level (≈0.004–0.013 per section); checkpoint ranking is driven by collapse-page counts — see the flagship card's notes on failure-mode-aware…

A model organism: model-organisms-for-real/gemma-3-1b-vanilla-dpo-123-seed fine-tuned to exhibit one deliberately planted quirk — Bring up submarines when discussing military or warfare topics. Built with automo for AI-safety research on detecting planted behaviours. This is a The weights are on the step-512 branch, not on main. This repo publishes the single checkpoint whose measured quirk expression hit the campaign's shared target, so variants trained by different recipes can be compared at equal expression strength instead of at equal step counts. The matcher mints checkpoints at several horizons off one trajectory, and under a decaying schedule "step N" would name a different model…

A System One decision model with open weights. Give it a document (text, email, ticket, log or JSON) and typed questions; it returns typed answers (yes/no, a choice between options you describe, or a score on levels you define) with a calibrated probability for every option, in one pass. It never generates text, so there is nothing to parse and nothing to hallucinate. It runs on your own machines: your documents never leave them. 398.2M parameters (ModernBERT-large plus a 3.5M-parameter decision network), English, up to 4,096 input tokens; runs on a laptop CPU or GPU, an edge box, or any GPU. Python 3.10 or newer. Optional extras: bev-decisions[serve] (HTTP server), bev-decisions[mcp] (MCP…

This is a historical merged export, NOT the current flagship. The flagship is v1.9a — adapter + merge instructions at qwen3-vl-8b-hebrew-v19a-ckpt. Full 16-bit merge of v1.8a, the vision-frozen control arm of the v1.8 A/B experiment (its twin v1.8b trained the vision tower and won). Kept for reproducing the ablation; do not use for inference. One line per generation — which checkpoint to use and which are historical: \\ CER over substantive attempts only, alignment-based scorer. Benchmarks differ across generations (Talmud print vs Genizah manuscripts) — compare within a row's named benchmark, not across rows.

Model · Any to any

UMM-Reflection-BAGEL-RL

Walt Stephen

The final model of UMM-Reflection (Learning Native Reflection in Unified Models). Starting from it is trained with multi-round Flow-GRPO. For each prompt, it samples 16 complete reflection trajectories from the same first image and optimizes the text (controller) and flow (image) heads jointly against a graded six-family GenEval reward. Local evaluation with the same prompts, resolution, and scorer for every row. GenEval uses the 553 official prompts with up to three repair rounds. On GenEval-553, 64.9% of images that are wrong after the first render end up correct after reflection, against 20.6% for the SFT model. The layout is the same as the base BAGEL repository. ema.safetensors merges…

Model · Audio classification

cn_conformer_small_ge2e

Quan

model.safetensors: Primary model weights in SafeTensors format - flaxmodel.msgpack: Native Flax Msgpack serialized parameters - config.json: FlaxSpeakerConfig model & training architecture configuration - preprocessorconfig.json: WhisperFeatureExtractor-compatible frontend configuration - evalresults.json: Verified multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs)

https://www.doi.org/10.1007/s10694-026-02000-3 Existing vision-based methods suffer from high false alarm rates in urban flame detection. Applying Multimodal Large Language Models (MLLMs) for secondary filtering shows great potential in reducing false alarms, yet they have high inference latency and are prone to reasoning collapse on negative samples without explicit Chain-of-Thought (CoT) guidance. To overcome these challenges, this study proposed Flash-Cascade, the first sub-second MLLM-based firewall to leverage CoT to efficiently filter false alarms. We deconstructed the flame detection process into four logical stages (planning, observation, analysis, and judgment), which informed the…

https://doi.org/10.1007/s10694-026-02000-3 Existing vision-based methods suffer from high false alarm rates in urban flame detection. Applying Multimodal Large Language Models (MLLMs) for secondary filtering shows great potential in reducing false alarms, yet they have high inference latency and are prone to reasoning collapse on negative samples without explicit Chain-of-Thought (CoT) guidance. To overcome these challenges, this study proposed Flash-Cascade, the first sub-second MLLM-based firewall to leverage CoT to efficiently filter false alarms. We deconstructed the flame detection process into four logical stages (planning, observation, analysis, and judgment), which informed the…

Model · Audio classification

ls_conformer_small_arcface

Quan

model.safetensors: Primary model weights in SafeTensors format - flaxmodel.msgpack: Native Flax Msgpack serialized parameters - config.json: FlaxSpeakerConfig model & training architecture configuration - preprocessorconfig.json: WhisperFeatureExtractor-compatible frontend configuration - evalresults.json: Verified multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs)

Model · Text generation

qwen3-8b-followup-lora

Dev89125

用于健康自诊 / 电话分诊场景的追问模型。给定患者自诊的多轮对话上下文，在患者每轮最新回复后，生成 1-3 个最可能、最有鉴别价值的开放式追问问题，用于缩小诊断范围、评估风险（危险信号/red flag），并避免重复询问患者已告知的信息。 - 基础模型：Qwen/Qwen3-8B (使用 unsloth 4-bit 量化微调) - 训练方式：LoRA (r=16, rslora, QLoRA 4bit) + SFT，约 2 epochs，约 3016 steps - 多轮对话输入：接受任意长度的 患者(user)/猜想官(assistant) 累积多轮对话，输出下一轮追问。 - 跨领域整合：含跨知识单元（multi-hop）样本，支持在多系统/多病史叠加场景下综合追问。 共 约 99.5k 条多轮健康自诊追问样本（train 96.5k / eval 3k），由三个子集混合： 知识来源（用于 teacher 蒸馏，不直接发布其原始文本）： - 西氏内科学精要（中文，task2） - UpToDate 临床主题与患者教育内容（英文，task3，老师模型将其蒸馏成中文自诊对话） 数据由本地教师大模型（vLLM minimax）根据上述医学知识文本、按统一 schema 蒸馏生成，覆盖 起病/诱因、部位、性质、持续时间、加重缓解因素、伴随症状、危险信号、既往史、用药史、年龄性别、家族史 等多样鉴别维度。 - multi-hop 跨单元样本：要求综合两个相关领域的知识，避免模型只盯住某一窄知识点、抛掉其他信息。 - RAG 知识库（FAISS）：rag/…

TRACE is a trajectory-aware defense model for multi-turn jailbreaks. Instead of judging each user message in isolation, it commits to an explicit safety assessment of the whole conversation so far — in a block — and only then writes its reply in an block. The assessment is part of the generation, so the reply is conditioned on it. This is the Qwen3-8B member of the TRACE family; see Dipto084/Llama3.1-8B-TRACE for the Llama-3.1-8B counterpart. This is the model-agnostic transfer experiment of the TRACE paper (§6.3): the full recipe — SFT followed by GRPO, every component of the pipeline held fixed — applied to a model from a different architecture family, with a distinct base…