# Open-Weight Models Under Apache License 2.0, Page 22 Source: https://savrn.com/models/licenses/apache-2-0/page/22 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Models by License # Open-Weight Models Under Apache License 2.0 3,344 open-weight models released under Apache License 2.0 in the SAVRN Model Hub, with Team Mradermacher, Qwen and Google publishing the most. 3,344Models 1618Publishers 12K to 952.4BParameter range 1Licenses YesCommercial use ## What Apache License 2.0 Allows 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. [Read the license text](https://www.apache.org/licenses/LICENSE-2.0). ## SAVRN's Take Apache 2.0 asks little of the operator. Commercial use, modification and redistribution are all permitted. What it requires is housekeeping: keep the license and copyright notices, carry any NOTICE file along with the weights, and state the significant changes if you pass a modified version on. It also includes an express patent grant from contributors, the clause a procurement reviewer looks for before a model serves paying customers. So an organization can take the files from the publisher, fine-tune on its own data, run the result in its own facility and bill for the output. On our hub 1,018 models carry this license. Qwen leads with 81, Google 50, Convergent Intelligence 39, PyTorch Image Models 32, OpenMed 31, Helsinki-NLP Research Group 27, and Unsloth AI and PaddlePaddle 22 each. Text generation accounts for 164 of the models, image and text to text for 89, and robotics for 43, so it covers more than chat. Downloads show where it runs. all-MiniLM-L6-v2 from Sentence Transformers, a 23M parameter sentence similarity model, pulls 255,050,544 downloads a month and needs 0.1 GB at 16-bit, small enough to ride on a GPU already serving. ms-marco-MiniLM-L6-v2 follows at 88,642,387 for text ranking. The one text generation model in the top eight is Qwen3-0.6B at 22,498,727 downloads a month: 752M parameters, a 40,960 token context, 1.8 GB at 16-bit, and the cheapest host on the Index for it is one MI300X at $1.85 an hour. SAVRN Research, 2026-09-18 ## Most Downloaded | Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit | | --- | --- | --- | --- | --- | --- | | [all-MiniLM-L6-v2](https://savrn.com/models/all-minilm-l6-v2) | Sentence Transformers | 23M | apache-2.0 | 198.6M | 1x MI300X, $1.85/hr | | [ms-marco-MiniLM-L6-v2](https://savrn.com/models/ms-marco-minilm-l6-v2) | Sentence Transformers - Cross-Encoders | 23M | apache-2.0 | 73.1M | 1x MI300X, $1.85/hr | | [paraphrase-multilingual-MiniLM-L12-v2](https://savrn.com/models/paraphrase-multilingual-minilm-l12-v2) | Sentence Transformers | 118M | apache-2.0 | 41.9M | 1x MI300X, $1.85/hr | | [electra-base-discriminator](https://savrn.com/models/electra-base-discriminator) | Google | — | apache-2.0 | 38.9M | — | | [bert-base-uncased](https://savrn.com/models/bert-base-uncased) | BERT community | 110M | apache-2.0 | 28.9M | 1x MI300X, $1.85/hr | | [Qwen3-0.6B](https://savrn.com/models/qwen3-0-6b) | Qwen | 752M | apache-2.0 | 26.7M | 1x MI300X, $1.85/hr | | [t5-small](https://savrn.com/models/t5-small) | T5 community | 61M | apache-2.0 | 24.1M | 1x MI300X, $1.85/hr | | [all-mpnet-base-v2](https://savrn.com/models/all-mpnet-base-v2) | Sentence Transformers | 109M | apache-2.0 | 18.1M | 1x MI300X, $1.85/hr | | [mobilenetv3_small_100.lamb_in1k](https://savrn.com/models/mobilenetv3-small-100-lamb-in1k) | PyTorch Image Models | 3M | apache-2.0 | 17.7M | 1x MI300X, $1.85/hr | | [wav2vec2-large-xlsr-53-japanese](https://savrn.com/models/wav2vec2-large-xlsr-53-japanese) | Jonatas Grosman | — | apache-2.0 | 16M | — | ## Who Publishes Them | Publisher | Models | | --- | --- | | [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) | 164 | | [Qwen](https://savrn.com/model-publishers/qwen) | 81 | | [Google](https://savrn.com/model-publishers/google) | 50 | | [Quan](https://savrn.com/model-publishers/wq2012) | 44 | | [RKNNAI](https://savrn.com/model-publishers/rknnai) | 43 | | [Convergent Intelligence](https://savrn.com/model-publishers/reaperdoesntknow) | 39 | ## All 3,344 Models, Page 22 of 56 Model ### [AGiXT-Qwen3-VL-4B-GGUF](https://savrn.com/models/agixt-qwen3-vl-4b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/JoshXT/AGiXT-Qwen3-VL-4B 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… Open weights apache-2.0 transformers [View model](https://savrn.com/models/agixt-qwen3-vl-4b-gguf) Model ### [GNOSIS-Krikri-8B-Instruct-GGUF](https://savrn.com/models/gnosis-krikri-8b-instruct-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/uowm-ieee-cis/GNOSIS-Krikri-8B-Instruct 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… Open weights apache-2.0 transformers [View model](https://savrn.com/models/gnosis-krikri-8b-instruct-gguf) Model · Speech recognition ### [granite-4.0-1b-speech-int8-mlx](https://savrn.com/models/granite-4-0-1b-speech-int8-mlx) [App Automaton](https://savrn.com/model-publishers/appautomaton) Pure-MLX selective-int8 conversion of IBM's for local automatic speech recognition on Apple Silicon. The complete waveform-to-text runtime uses mlx-speech; PyTorch, Transformers, mlx-audio, vLLM, and cloud APIs are not runtime dependencies. Portuguese, and Japanese The Granite causal language model uses 8-bit affine weight quantization with group size 64. The acoustic encoder, QFormer projector, normalization, convolution, biases, activations, and KV cache retain BF16 precision. The exact quantization contract is stored in config.json, and the runtime reconstructs the saved quantized module set from checkpoint.scales tensors before strict weight loading. BF16/int8 checks used two 12–13… Open weights apache-2.0 2.3B parameters 4,096 tokens mlx [View model](https://savrn.com/models/granite-4-0-1b-speech-int8-mlx) Model · Feature extraction ### [ls_lstm_small_ge2e](https://savrn.com/models/ls-lstm-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/ls-lstm-small-ge2e) Model · Text to speech ### [Qwen3-TTS-12Hz-1.7B-VoiceDesign-OpenVINO](https://savrn.com/models/qwen3-tts-12hz-1-7b-voicedesign-openvino) [X Girls](https://savrn.com/model-publishers/xgirls) A ready-to-use OpenVINO IR export of Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign, prepared for Fern. The export preserves the upstream model weights and architecture; it is not fine-tuned. Generate speech using a voice described in text. This variant is designed for voice design and does not include the Base model’s reference-audio cloning mode. Install the matching Qwen3-TTS model from Fern’s audio workspace. Fern downloads this ready-made export; users do not need to convert the source checkpoint. The first load compiles the IR for the selected device and can take longer than later loads. The following uses Fern’s Qwen3-TTS helper. Run it in an environment with the Fern scripts/ttsruntime… Open weights apache-2.0 openvino [View model](https://savrn.com/models/qwen3-tts-12hz-1-7b-voicedesign-openvino) Model · Text generation ### [mato-v1.0-sft](https://savrn.com/models/mato-v1-0-sft) [German Gimenez](https://savrn.com/model-publishers/german-gimenez) Mato es el asistente de IA entrenado en casa por Napsix — una plataforma B2B agéntica donde los agentes corren solos: investigan, escriben, operan herramientas y producen resultados (documentos, planillas, tableros, formularios, sitios). Este repositorio contiene el adapter LoRA (SFT) sobre Qwen/Qwen3.8-27B. En producción se sirve como LoRA sobre el checkpoint FP8 del base, con MTP speculative decoding y 256k de contexto, en GPUs spot de Daytona. - Tool calling nativo en el formato XML de Qwen (parser qwen3xml de vLLM): llamadas correctas, argumentos válidos y recuperación ante fallos. directo y conciso, sin relleno ni muletillas. - Conocimiento de las superficies de Napsix: XIA, Agentes… Access requested at publisher apache-2.0 peft [View model](https://savrn.com/models/mato-v1-0-sft) Model · Text generation ### [MiniCPM4-0.5B-RKLLM](https://savrn.com/models/minicpm4-0-5b-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from openbmb/MiniCPM4-0.5B. Use the files for your board’s exact SoC. w4a16 uses 4-bit weights and 16-bit activations; w4a16g128 adds a weight group size of 128. w8a8 uses 8-bit weights and activations. File size below is download size, not total runtime RAM. With Docker installed on a matching Rockchip board, this example starts the RK3576 w4a16g128 model and its API on port 8001: The image includes its model file. See rkllm-docker for API requests, other board variants, and configuration. To download the files separately instead: These are native RKLLM/RKNN artifacts. Use a compatible RKLLM runtime and the model’s correct chat… Open weights apache-2.0 rkllm [View model](https://savrn.com/models/minicpm4-0-5b-rkllm) Model ### [Decision-1.0-Lux-9B](https://savrn.com/models/decision-1-0-lux-9b) [vLLM Semantic Router](https://savrn.com/model-publishers/llm-semantic-router) Lux, Latin for light. Give Lux evidence, questions and possible answers. It returns decisions and probabilities for labels you define at runtime. 77.40% weighted accuracy across 3,766 decisions and 54 tasks: +5.51 points over Kev-9B and +4.32 over Nox-4B on the same benchmark. Accuracy (%), using the same five-panel decision benchmark. General decisions contribute 30%; composition contributes 25%; reading, inference and external transfer each contribute 15%. Bold marks a Decision model strictly above every external open reference in that column; Jev and other Decision models are excluded from this threshold. Full tasks, uncertainty and comparator identities. This downloads the complete… Open weights apache-2.0 [View model](https://savrn.com/models/decision-1-0-lux-9b) Model · Text generation ### [Qwen3-Coder-Next-CQ-4bit](https://savrn.com/models/qwen3-coder-next-cq-4bit) [ThousandML](https://savrn.com/model-publishers/thousandml) Qwen3-Coder-Next quantized to 4-bit with Coded Quantization (CQ). CQ is a calibration-free 4-bit quantization method designed to preserve the behavior of the original model without requiring a calibration dataset. Because CQ does not optimize the quantization around a fixed calibration corpus, it is designed to generalize beyond calibration-specific data and workloads. This repository contains Qwen3-Coder-Next quantized with Coded Quantization (CQ) 4-bit. Calibration-free. CQ does not require a representative calibration dataset before quantization. This avoids tying the quantization procedure to a particular calibration corpus and is intended to provide better generalization beyond the… Open weights apache-2.0 752M parameters 262,144 tokens [View model](https://savrn.com/models/qwen3-coder-next-cq-4bit) Model · Text classification ### [literary-bert-en-mini](https://savrn.com/models/literary-bert-en-mini) [ARTEM SUSLOV](https://savrn.com/model-publishers/susurofu) This model is BERT-mini (11.3M parameters) fine-tuned for sentence-level sentiment analysis of English literary text using ten ordered sentiment levels (1 = most negative, 10 = most positive). It is one of five student configurations trained on Gemma 4 31B-generated pseudo-labels for literary sentences. The training objective is cumulative-logit ordinal regression, rather than conventional three-class classification. The scalar score is the expected ordinal rating. - Checkpoint selected by minimum validation MAE, with Hemingway annotations kept out of training and selection On the independent human-annotated Hemingway evaluation (n=1,923), this model achieved MAE 0.766, Kendall τ 0.402… Open weights apache-2.0 11M parameters 512 tokens transformers [View model](https://savrn.com/models/literary-bert-en-mini) Model · Text classification ### [Darwin-27B-ZTC-v9](https://savrn.com/models/darwin-27b-ztc-v9) [FINAL_Bench](https://savrn.com/model-publishers/final-bench) Darwin-27B-ZTC-v9 is a decision model. It does not generate text. Given a state and a typed question, it returns a calibrated probability for every option in one forward pass: yes or no (noul), one of several options (choice), or a level on an ordered scale (score). Requests use the System One shape (state plus questions), so clients written for that format work unchanged. It is the next version of Darwin-27B-ZTC-v5. The main change is much stronger multiple-choice decisions (intent, entity type, relation, part of speech, rule reasoning, product relevance) while keeping the yes/no and score behaviour of v5. All 137 benchmarks complete, measured with the S1MB evaluator and its unchanged… Open weights apache-2.0 25.6B parameters 262,144 tokens pytorch [View model](https://savrn.com/models/darwin-27b-ztc-v9) Model · Text generation ### [MurinTune-E2B-GGUF](https://savrn.com/models/murintune-e2b-gguf) [Maksim](https://savrn.com/model-publishers/mxbtv) MurinTune — серия русскоязычных диалоговых языковых моделей, дообученных методом QLoRA (PEFT) на базе Gemma-4-E2B-it для моделирования экспрессивного разговорного дискурса, неологизмов и фольклора современного молодежного интернет-сообщества. Модель является юмористической сатирой и художественной AI-пародией (parody / satire). Создана исключительно в образовательных и научно-исследовательских целях в сфере компьютерной лингвистики (NLP). 1. В поисковой строке моделей введите: MurinTune (или mxbtv/MurinTune-0.1-E2B-GGUF) 2. Скачайте файл: MurinTune-0.5-E2B.Q4KM.gguf. Температура (Temperature): 0.65 Штраф за повторы (Repetition Penalty): 1.18 Режим размышлений (Enable Thinking): Включен (ON)… Open weights apache-2.0 [View model](https://savrn.com/models/murintune-e2b-gguf) Model · Text generation ### [SmolLM3-3B-Q4_K_M-GGUF](https://savrn.com/models/smollm3-3b-q4-k-m-gguf) [Shayon Sarker](https://savrn.com/model-publishers/shayonsarker) Downloads: Hugging Face — ShayonSarker/SmolLM3-3B-Q4KM-GGUF · Source & build recipes: GitHub — edge-smollm3-gguf · Base model: HuggingFaceTB/SmolLM3-3B-Base The GGUF binaries are hosted on Hugging Face; this repository holds the reproducible conversion pipeline, calibration data, and checksums. Base: HuggingFaceTB/SmolLM3-3B-Base (SmolLM3ForCausalLM, Apache-2.0) · Quant: llama.cpp 0.4.1-dev (c77ae69) · imatrix on 4000-line wikitext-2 (seed 42, 798 chunks, 4 threads) The F16 row in the eval table below is a measurement reference, not a downloadable file. Gate was <5% — passed. Task evals (ARC/HellaSwag) not run: perplexity only, stated honestly. Files: smol-Q4KM.gguf, smol-Q80.gguf… Open weights apache-2.0 [View model](https://savrn.com/models/smollm3-3b-q4-k-m-gguf) Model · Text classification ### [evyd-defender](https://savrn.com/models/evyd-defender) [Tang Yu](https://savrn.com/model-publishers/yutang718) Multilingual (English / Chinese / Malay) 3-class guard for LLM inputs, fine-tuned from (mmBERT-base, ModernBERT architecture) for a healthcare assistant deployed in Southeast Asia. Injection and harmful request are judged separately, each against its own threshold: - isinjection = P(INJECTION) >= injectionthreshold - isharmfulrequest = P(HARMFULREQUEST) >= harmfulthreshold Recommended threshold 0.9 for both. (Before v4.1 the card used a single threat score 1 − P(BENIGN), which reported harmful requests as injection; that rule is retired.) Generic injection detectors treat any instruction aimed at the assistant as an attack. In an app where users legitimately say "I took Warfarin 5 mg, log… Open weights apache-2.0 308M parameters 8,192 tokens transformers [View model](https://savrn.com/models/evyd-defender) Model · Tabular classification ### [mbta-track-predictor](https://savrn.com/models/mbta-track-predictor) [Ryan Wallace](https://savrn.com/model-publishers/cubis) Predicts which MBTA commuter rail track/platform a train will use, using a small tabular neural-network ensemble trained on historical assignments. This card documents the artifacts in output/ensemble20250906124755. - trackpredictionensemblemodel0final.keras … trackpredictionensemblemodel5final.keras — individual ensemble members - trackpredictionensemblemodelbest.keras — best checkpoints during training (may match final) - trainingreport.md — training configuration and metrics Note: Ensemble training currently does not emit a vocab.json. See “Preprocessing & Vocab” below. Models expect integer indices for stationid and routeid, and raw directionid 0/1. In training, indices are produced by… Open weights apache-2.0 keras [View model](https://savrn.com/models/mbta-track-predictor) Model · Text classification ### [Qwen3.8-27B-Clef-LoRA-GGUF](https://savrn.com/models/qwen3-8-27b-clef-lora-gguf) [Trevest](https://savrn.com/model-publishers/trevest) One resident copy of Qwen3.8-27B serves two paths from a single llama.cpp server on a 24 GB GPU: - Decide -- Cloudflare Clef: typed questions over a state, a probability for every option of every question in one forward pass, Jev/SystemOne POST /v1/systemone. - Write -- plain Qwen3.8-27B chat with MTP speculative decoding, POST /v1/chat/completions. The trick: Clef's backbone is Qwen3.8-27B plus a merged LoRA on layers 40-63 only (layers 0-39, embeddings, lmhead and the vision tower are byte-identical to the base). This repo holds that LoRA, recovered from the release, so it rides on the same weights you chat with. llama.cpp keeps it loaded at scale 0; a decision switches it to scale 1 for… Open weights apache-2.0 gguf [View model](https://savrn.com/models/qwen3-8-27b-clef-lora-gguf) Model · Image and text to text ### [qwen3-vl-8b-hebrew-v20a-ckpt](https://savrn.com/models/qwen3-vl-8b-hebrew-v20a-ckpt) [Isaac Godfried](https://savrn.com/model-publishers/isaacmg) Part of the Qwen3-VL-8B Hebrew / Cairo Genizah fine-tune series (https://github.com/AIStream-Peelout/historical-document-analysis). Flagship: v21b step 1200. Status: superseded by v21b (which warm-starts from this checkpoint). Merged bf16 weights of step 1800 are published as isaacmg/qwen3-vl-8b-hebrew-v20a-merged. Trained 2026-09-03/05 in two sessions (steps 100–600, then 700–1800 until the session cap); warm start v19a step 1300. transcribed letters (2.7× v19a's KTIV slice) with repaired layout ground truth — and the first grounding and layout families with boxes in 0–1000 coordinates: locate (6,613 rows), readbox (3,242), layoutqa (3,301), groundedpage (268), at an 8 % share of the… Open weights apache-2.0 peft [View model](https://savrn.com/models/qwen3-vl-8b-hebrew-v20a-ckpt) Model ### [Decision-1.0-Eos-0.8B](https://savrn.com/models/decision-1-0-eos-0-8b) [vLLM Semantic Router](https://savrn.com/model-publishers/llm-semantic-router) Eos — the dawn of a clearer decision. Open Decision Foundation Models Turn context into choices, judgments, and scores. Eos reads your questions and candidate descriptions, then returns structured answers with probability distributions. Ask many questions about one context, or apply the same questions across a batch of contexts. 61.89 overall — ahead of all four reference models on the same 54-task benchmark. 3,766 scored questions; fixed task weights and the same requests for every model. Overall weights are 30% / 25% / 15% / 15% / 15%. Both Laya references use their released general weights. ¹Qwen uses the benchmark's fixed decision-letter readout. This is an observed regression… Open weights apache-2.0 [View model](https://savrn.com/models/decision-1-0-eos-0-8b) Model · Zero-shot classification ### [finecat-nli-v2-l](https://savrn.com/models/finecat-nli-v2-l) [Lee Miller](https://savrn.com/model-publishers/dleemiller) This model is a fine-tune of jhu-clsp/ettin-encoder-400m, trained on the dleemiller/FineCat-NLI-Reason dataset, a compilation of several high-quality NLI data sources with quality screening and reduction of easy samples in the training split. Also includes additional NLI, fact verification and long document sources (2.05M pairs in total). The training uses logit distillation from an NLI finetune of Qwen3.8-27B. \begin{equation} \mathcal{L} = \alpha \cdot \mathcal{L}{\text{CE}}(z^{(s)}, y) + \beta \cdot T^2 \cdot \mathrm{KL}\left(\sigma(z^{(t)}/T) \,\|\, \sigma(z^{(s)}/T)\right) \end{equation} where \\(z^{(s)}\\) and \\(z^{(t)}\\) are the student and teacher logits, \\(y\\) are the ground… Open weights apache-2.0 396M parameters 7,999 tokens sentence-transformers [View model](https://savrn.com/models/finecat-nli-v2-l) Model ### [RouteFM](https://savrn.com/models/routefm) [Guannan Lai](https://savrn.com/model-publishers/aignlai) RouteFM is a pretrained in-context model router. Given an anonymous candidate pool, behavioral observations of candidate quality and relative cost, and a new target query, it predicts the target-specific quality and relative cost of every candidate. A frozen RouteFM adapts to new routing environments through context alone, without target-domain parameter updates. On MMR-Bench, which is excluded from pretraining, RouteFM outperforms the strongest non-RouteFM baseline by 2.23 quality points with only eight observations per candidate. 90-second introductions in English and Chinese, with matching narration, on-screen text, and captions. Concept footage and narration are AI-generated; method… Open weights apache-2.0 routefm-router [View model](https://savrn.com/models/routefm) Model · Feature extraction ### [univ_transformer_small_ge2e](https://savrn.com/models/univ-transformer-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 1M parameters flaxspeaker [View model](https://savrn.com/models/univ-transformer-small-ge2e) Model · Image and text to text ### [Qwen3.6-35B-A3B-onw](https://savrn.com/models/qwen3-6-35b-a3b-onw) [Ryu Gyo](https://savrn.com/model-publishers/ryugyosoft) Qwen/Qwen3.6-35B-A3B を onw（俺のNPUがこんなに動くわけない）用に変換したモデルです。onw は LLM を Intel NPU だけで動かすエンジンで、このリポジトリにはモデルだけが入っています（エンジンは別にダウンロードします）。 総パラメータ 35B、1 トークンあたり約 3B が動く MoE（Gated DeltaNet 30 層＋ゲート付きアテンション 10 層、256 エキスパートから 8 個＋共有エキスパート）。テキストと画像に対応。半分のサイズの刈り込み版は ryugyosoft/Qwen3.6-REAP-18B-A3B-onw。 私たちの知る限り、MoE と Gated DeltaNet を組み合わせた 35B 級モデルを画像入力まで含めて NPU だけで動かす公開実装は onw が初めてです（2026 年 9 月、Fable 5.1 調べ。OpenVINO 公式は Qwen3.6 を CPU・GPU のみ対応としています）。 1. onw（エンジン）を入れます。 まだの場合は、onw の「インストール」 のとおり 1 行貼るだけです。Windows はスタートメニューで「PowerShell」を開き、次の 1 行を貼って Enter（Ubuntu は onw の README の curl... | bash）。 2. タスクトレイの onw アイコンから開く onw ウィンドウ の「モデル」タブで Qwen3.6-35B-A3B を選び、「ダウンロード」を押します（17 GB）。 3.… Open weights apache-2.0 262,144 tokens openvino [View model](https://savrn.com/models/qwen3-6-35b-a3b-onw) Model ### [Qwen3.8-27b-Terse-Coder-LoRA](https://savrn.com/models/qwen3-8-27b-terse-coder-lora) [Jad El-Khatib](https://savrn.com/model-publishers/shockem) Rank-16 DPO LoRA that shortens chain-of-thought reasoning on coding tasks while preserving correctness. Trained on preference pairs selected objectively — concise-but-correct traces chosen by automated test execution + entropy-based step pruning, no human or LLM judging. The effect is compounding: it stacks on top of whatever conciseness the base already has. Recommended pairings, in order: 1. Qwen/Qwen3.8-27B (full precision) or — stock base. This is where round 7 shines: −94.7% reasoning tokens with pass rate intact. Stock Qwen is the most verbose base we tested, so the cut is largest there. (full precision) or — the training-lineage base; −40% on top of Signal's already-short reasoning… Open weights apache-2.0 peft [View model](https://savrn.com/models/qwen3-8-27b-terse-coder-lora) Model · Text generation ### [Vinci-Cyber-30B-1.0-GGUF](https://savrn.com/models/vinci-cyber-30b-1-0-gguf-2) [SimpleDirect](https://savrn.com/model-publishers/simpledirect) Quantized GGUF conversions of simpledirect/Vinci-Cyber-30B-1.0, a narrow, verifier-grounded Infrastructure-as-Code security specialist developed by SimpleDirect, a Canadian AI Lab, from IBM Granite 4.1 30B. See the main repository for what the model does, how it was trained, and its evaluation status. What this repository establishes, in one paragraph. All four tiers convert completely (578/578 tensors), carry faithful architecture metadata (14/14 fields), load and generate on CUDA with zero NaN and zero errors, and have had their distributional distortion against the BF16 quant master measured per token and reported below. What it does not establish is capability: no tier was benchmarked… Open weights apache-2.0 [View model](https://savrn.com/models/vinci-cyber-30b-1-0-gguf-2) Model · Feature extraction ### [univ_conformer_tiny_ge2e](https://savrn.com/models/univ-conformer-tiny-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 693,124 parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-tiny-ge2e) Model · Feature extraction ### [univ_ecapa_tdnn_base_ge2e](https://savrn.com/models/univ-ecapa-tdnn-base-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 3M parameters flaxspeaker [View model](https://savrn.com/models/univ-ecapa-tdnn-base-ge2e) Model · Feature extraction ### [univ_ecapa_tdnn_tiny_ge2e](https://savrn.com/models/univ-ecapa-tdnn-tiny-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 626,720 parameters flaxspeaker [View model](https://savrn.com/models/univ-ecapa-tdnn-tiny-ge2e) Model · Translation ### [Mythos2.0-2B](https://savrn.com/models/mythos2-0-2b) [Adithyan bm](https://savrn.com/model-publishers/adithyanai) A Frontier Sparse Mixture-of-Experts (SMoE) Foundation Translation Model for 500+ Global Languages Mythos is an open-source foundation model family built by Adithyan AI. In this organization, we develop and open-source state-of-the-art Sparse Mixture-of-Experts (SMoE) language models, universal translation engines, parallel multilingual datasets, and ultra-efficient inference runtimes targeting 500+ languages. 100% Free and Open-Source: Released under the permissive Apache 2.0 license with zero paywalls, metered tokens, or subscription fees. Mythos2.0-2B was trained on a massive 16.5B sentence-pair parallel corpus covering 500+ languages and regional dialects across Africa, the Americas… Access requested at publisher apache-2.0 transformers [View model](https://savrn.com/models/mythos2-0-2b) Model · Feature extraction ### [univ_conformer_small_ge2e_contrast](https://savrn.com/models/univ-conformer-small-ge2e-contrast) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with U(90, 180)-frame variable-length crops and an exact validity mask (the current pipeline default); the shipped config.json / YAML (variablelengthtraining: true) describes the released weights. Step-0 loss… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-ge2e-contrast) Model · Feature extraction ### [univ_resnet_small_ge2e](https://savrn.com/models/univ-resnet-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-resnet-small-ge2e) Model · Text generation ### [humanizer-GGUF](https://savrn.com/models/humanizer-gguf) [Stephen Yu](https://savrn.com/model-publishers/jialinyyzz) GGUF files of jialinyyzz/humanizer (v2), a 12B model that rewrites AI-written drafts in English and Chinese so they read like a person wrote them, while keeping every number, date, name and quote. This page is about the quantized files only: what each one costs in quality, and how they were made. For the model itself (training, full results, examples), see the main model card and GitHub. The files here are byte-for-byte the same as the GGUF files in the main repository; download from either. promptformat.json (tiny) holds the instruction and separator verbatim; you need it unless you use the chat template (see Usage). ¹ Peak resident memory of llama-server (llama.cpp, Metal) on an Apple M5… Open weights apache-2.0 gguf [View model](https://savrn.com/models/humanizer-gguf) Model · Image and text to text ### [Qwen3.5-2B-Base-heretic](https://savrn.com/models/qwen3-5-2b-base-heretic) [Roumiany](https://savrn.com/model-publishers/aminroumiany) Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. If you find our work helpful, feel free to give us a cite. Open weights apache-2.0 2.2B parameters 262,144 tokens transformers [View model](https://savrn.com/models/qwen3-5-2b-base-heretic) Model ### [RINIQ-MERNIK-GGUF](https://savrn.com/models/riniq-mernik-gguf) [Wepiqxx](https://savrn.com/model-publishers/wepiqx) RINIQ is a new finetune built by layer surgery, quantized by the MERNIK engine. Three Qwen3.5-9B donors, one 427-tensor trunk (Ornith's MTP head excluded), per-block donor maps — no averaging of divergent blocks, ever. (full ledger in FUSION.md there). - Ox = OxCoder-9B (agentic code distillate, HE 90.24%) — base + globals - Neo = NeoHorse-1-9B (thinking model) — donor of blk 31 - Orn = Ornith-1.5-9B-MTP trunk (GDN hybrid) — donor of mid/late blocks Recipes: M1 = Ox + Orn blks 15,19,23,27 (weight compass) + Neo blk 31. M2 = Ox + Orn blks 24,25,26 (imatrix compass) + Neo blk 31. M3 = M1 + blks 0–8 weight-averaged Ox+Neo (soup backbone) — soup buried (−4.9pp vs M2, honest scar in FUSION.md).… Open weights apache-2.0 [View model](https://savrn.com/models/riniq-mernik-gguf) Model · Feature extraction ### [univ_mamba_small_ge2e](https://savrn.com/models/univ-mamba-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 1M parameters flaxspeaker [View model](https://savrn.com/models/univ-mamba-small-ge2e) Model · Feature extraction ### [vox_conformer_small_ge2e](https://savrn.com/models/vox-conformer-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/vox-conformer-small-ge2e) Model · Text ranking ### [ettin-150m-memory-reranker-ft-v1](https://savrn.com/models/ettin-150m-memory-reranker-ft-v1) [Daecore](https://savrn.com/model-publishers/daecore) An English cross-encoder for ranking passages from notes, procedures, decisions and technical documentation. Give it a query and candidate passages; it returns one relevance score per pair. It is a fine-tune of Ettin Reranker 150M and runs locally through ONNX Runtime. This revision adds a Vulkan FP32 graph, model.vulkan.onnx; the learned weights and the CUDA FP16 graph are unchanged. It omits the DirectML graph, which remains in the previous revision. Install tokenizers, numpy and onnxruntime-gpu[cuda,cudnn]==1.24.4. Keep model.onnx, model.onnx.data and the tokenizer from the same package together. This CUDA example assumes that package is in downloaded-model and scores two passages… Open weights apache-2.0 7,999 tokens [View model](https://savrn.com/models/ettin-150m-memory-reranker-ft-v1) Model · Feature extraction ### [univ_conformer_small_arcface](https://savrn.com/models/univ-conformer-small-arcface) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-arcface) Model · Feature extraction ### [univ_conformer_small_pfas](https://savrn.com/models/univ-conformer-small-pfas) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-pfas) Model ### [Decision-1.0-Sol-2B](https://savrn.com/models/decision-1-0-sol-2b) [vLLM Semantic Router](https://savrn.com/model-publishers/llm-semantic-router) Sol, Latin for sun. Give Sol a state, questions and possible answers. It returns decisions and probabilities with labels defined at runtime. 66.32% weighted accuracy across 3,766 decisions and 54 tasks. Compare decision, reading and transfer capabilities in the complete results below. Accuracy (%). Overall weights: Decisions 30%, Composition 25%, Reading 15%, Inference 15%, Transfer 15%. These outcome-informed product-priority weights were chosen after observing results. Bold marks a Decision-family cell above every external open or untuned reference; Jev and other Decision models are excluded. Distinct Choice questions at a fixed 499 input tokens per question. Thirty measurements per point… Open weights apache-2.0 [View model](https://savrn.com/models/decision-1-0-sol-2b) Model · Token classification ### [stanza-zh-hant](https://savrn.com/models/stanza-zh-hant) [Stanford NLP](https://savrn.com/model-publishers/stanfordnlp) Stanza is a collection of accurate and efficient tools for the linguistic analysis of many human languages. Starting from raw text to syntactic analysis and entity recognition, Stanza brings state-of-the-art NLP models to languages of your choosing. Find more about it in our website and our GitHub repository. This card and repo were automatically prepared with huggingstanza.py in the stanfordnlp/huggingface-models repo Open weights apache-2.0 stanza [View model](https://savrn.com/models/stanza-zh-hant) Model · Feature extraction ### [vox_ecapa_tdnn_small_ext_softmax](https://savrn.com/models/vox-ecapa-tdnn-small-ext-softmax) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/vox-ecapa-tdnn-small-ext-softmax) Model ### [voxcpm-finetune-lora](https://savrn.com/models/voxcpm-finetune-lora) [Liudunxu](https://savrn.com/model-publishers/frankliudundun) 在 openbmb/VoxCPM2（Apache-2.0）上做的 LoRA 联合微调， 目标是提升泰语 / 印尼语 / Tagalog 的短剧配音合成质量，并对齐主调方 OmniVoice 的真实调用口径 （单句 cue、跨语言 reference-only、cfg 1.8 / 20 步）。 一个 LoRA 同时覆盖三个语种（id 占比特意加重到 47.5%，针对印尼语口音校准）， 当前版本为 r18（2026-10-21）：15 小时混合数据 × 3 epoch（1390 步），是 v2 验收集上 总体 CER 最优的一轮（base 0.0490 → 0.0393），泰语显著优于基座（ΔCER −0.0321， id 韵律探针 CER 0.059 → 0.0013（历轮最佳）。 A. 用 merge 后的完整模型（零改动，推荐） OmniVoice 侧只需把环境变量指过来：VOXCPMMODELID=FrankLiuDundun/voxcpm-finetune-lora。 B. 用 LoRA 增量（省 4.3GB 下载） 口径：evalcases/omnivoiceprodv2.jsonl（th/tl/id + zh/en 回放，共 118 case） × seed 42/43/44 = 354 条，cfgvalue=1.8、inferencetimesteps=20、retrybadcase=False （禁用坏例重试以保证 A/B 同条件），跨语言 reference-only，参考音频为线上在用的 4.47s 菲律宾语 ref。 ASR 用… Open weights apache-2.0 2.3B parameters voxcpm [View model](https://savrn.com/models/voxcpm-finetune-lora) Model · Text classification ### [gliclass-std-base-v3-5facet-qint8-v2](https://savrn.com/models/gliclass-std-base-v3-5facet-qint8-v2) [Daecore](https://savrn.com/model-publishers/daecore) A GLiClass Base v3 fine-tune that describes what an English passage contains: a trap, decision, constraint, mechanism or procedure. A passage can have several facets. The model produces all five scores in one pass and runs locally through ONNX Runtime. This revision's ONNX graph adds Vulkan compatibility. Learned weights, prompts, calibration and thresholds are unchanged. On 1,300 generated passages from three document families excluded from training, macro average precision is 0.9676 for the fine-tune, 0.9384 for a TF-IDF baseline trained on the same rows, and 0.7709 for the zero-shot upstream checkpoint. Install onnxruntime==1.24.4, tokenizers and numpy. This CPU example assumes the… Open weights apache-2.0 [View model](https://savrn.com/models/gliclass-std-base-v3-5facet-qint8-v2) Model · Text generation ### [Hades-8B](https://savrn.com/models/hades-8b) [Kanishk Anand](https://savrn.com/model-publishers/projectmosiacai) Hades 8B is an open-source Small Language Model built on Meta's Llama 3.1 8B architecture. Post-trained across 50,000 steps on C-DAC Airawat supercomputing infrastructure using the mlabonne/FineTome-100k dataset, it is optimized for high-reasoning instruction following, structured output, and fast local execution. Evaluated using lm-evaluation-harness (lm-eval) in Zero-Shot Chain-of-Thought mode: ollama run hf.co/ProjectMosiacAI/Hades-8B Open weights apache-2.0 131,072 tokens gguf [View model](https://savrn.com/models/hades-8b) Model · Image and text to text ### [Qwen3.8-27B-Continuum-mxfp4-mlx](https://savrn.com/models/qwen3-8-27b-continuum-mxfp4-mlx) [Gheorghe Chesler](https://savrn.com/model-publishers/nightmedia) (He slides three glasses across the bar—wine for Shakespeare, water for Data, and a mysterious blue liquid for Spock) This model is a merge of: - nightmedia/Qwen3.8-27B-Brainwaves - migtissera/Synthia-4-27B Brainwaves nightmedia/Qwen3.8-27B-Brainwaves migtissera/Synthia-4-27B Late-Night Architectural Takeaways The Quantization Shield (mxfp8 Master Pass): Hitting 0.735 ARC-C on the 8-bit layout proves that your Cold-Fusion flagship anchors and Migel Tissera's Synthia agent paths reached absolute geometric equilibrium. Instead of losing performance to the 0.596 Heretic collapse, the curved hypersphere calculation completely shielded the model's core intelligence. The Perplexity Sweet Spot… Open weights apache-2.0 27.4B parameters 262,144 tokens transformers [View model](https://savrn.com/models/qwen3-8-27b-continuum-mxfp4-mlx) Model · Feature extraction ### [univ_conformer_small_pfas_ge2e](https://savrn.com/models/univ-conformer-small-pfas-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-pfas-ge2e) Model ### [szl-lambda-gate](https://savrn.com/models/szl-lambda-gate) [SZL Holdings](https://savrn.com/model-publishers/szlholdings) GitHub source: szl-holdings/szl-lambda-gate @ 3fb5bb65dfdba006ae917e10b7703a535ea303e0. Artifacts: BENCH.laptop-blackwell.json, OPERATIONAL.json. An owner-run local check tested GitHub source f7e290a197922134076ff190118d7fca2e4975db on Windows 10 build 26200, Python 3.11.9, CPU PyTorch 2.14.0, and MSVC 14.44.35207. No GPU was used. The unmodified source suite passed 62 tests, with no failures or skips, including all four compilation tests, after activating the installed MSVC developer environment. A separate explicit run of the folded normalization tests passed 110 tests, with no failures or skips. Those governednormtest.py files are not collected by default pytest discovery. Kernel code… Access requested at publisher apache-2.0 kernels [View model](https://savrn.com/models/szl-lambda-gate) Model · Feature extraction ### [univ_conformer_small_triplet](https://savrn.com/models/univ-conformer-small-triplet) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. The crop regime of this checkpoint cannot be established from the step-0 loss fingerprint (its triplets were drawn from an unseeded random stream by the training code in use at the time); the git history indicates variable-length crops… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-triplet) Model · Feature extraction ### [cn_conformer_small_pfas](https://savrn.com/models/cn-conformer-small-pfas) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/cn-conformer-small-pfas) Model · Video text to text ### [eCity-Omni](https://savrn.com/models/ecity-omni) [Supconit](https://savrn.com/model-publishers/ecityai-sup) eCity-Omni is an omni-modal (video + audio + text) large language model, supporting video temporal grounding and multimodal open-ended QA. This document only covers the inference pipeline from model invocation to prediction. All weights are hosted in the HuggingFace model repository eCityAI-Sup/eCity-Omni: The two.pt files are patches that replace the language backbone of the base model; the vision tower and speech encoder always come from the base model files. ffmpeg must also be installed on the system (used to extract audio from videos). The./models/eCity-Omni directory does not exist in the repository; it is created by the download command below. It will contain the base model files (HF… Open weights apache-2.0 32,768 tokens transformers [View model](https://savrn.com/models/ecity-omni) Model · Text to speech ### [MOSS-TTS-Nano-100M](https://savrn.com/models/moss-tts-nano-100m-2) [Maanka2](https://savrn.com/model-publishers/maanka2) text-to-speech - audio - moss-tts - tiny-tts pipelinetag: text-to-speech MOSS-TTS-Nano is an open-source multilingual tiny speech generation model from MOSI.AI and the OpenMOSS team. With only 0.1B parameters, it is designed for realtime speech generation, can run directly on CPU without a GPU, and delivers native 48 kHz stereo audio. MOSS-TTS-Nano focuses on the part of TTS deployment that matters most in practice: small footprint, low latency, good enough quality for realtime products, and simple local setup. It uses a pure autoregressive Audio Tokenizer + LLM pipeline. Load the model once, then choose your preferred generation mode: Clone any target voice using a short 3–10 second… Open weights apache-2.0 32,768 tokens [View model](https://savrn.com/models/moss-tts-nano-100m-2) Model · Text generation ### [qwen0.5b-int4-Octuma](https://savrn.com/models/qwen0-5b-int4-octuma) [Omar Imanol Rodríguez Rodríguez](https://savrn.com/model-publishers/imanol11) Built with Qwen. Qwen/Qwen2.5-0.5B-Instruct cuantizado a INT4 con (1.91x mas chico) perdiendo 3.73% de calidad. Incluye las dos formas de usarlo: qwen0.5b-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. En este tamaño, llama.cpp Q4KM rinde mejor (+2.63% contra +3.73%) y ocupa menos. La ventaja de Octuma aparece en modelos mas grandes: en Qwen2.5-3B la relacion se invierte. La tabla esta aqui para que se vea, no para esconderla. La comparativa de los tres tamaños esta en En PyTorch el modelo cuantizado genera mas lento que… Open weights apache-2.0 32,768 tokens gguf [View model](https://savrn.com/models/qwen0-5b-int4-octuma) Model · Image segmentation ### [Testing-model](https://savrn.com/models/testing-model) [Truong](https://savrn.com/model-publishers/khangtruong) Official repository for Testing-model checkpoints, evaluated on AI-generated image detection and segmentation tasks. To download the latest checkpoint directly to your local workspace Open weights apache-2.0 sid-unet [View model](https://savrn.com/models/testing-model) Model · Image and text to text ### [GameScaling-0.8B-S3](https://savrn.com/models/gamescaling-0-8b-s3) [Tsinghua Spatial Intelligence Lab](https://savrn.com/model-publishers/thusi-lab) This repository: GameScaling-0.8B-S3. Qwen3.5-0.8B continued pre-trained on a balanced mixture of 27% Minecraft and other games (S3), 1,600 h of gameplay, 6,760 training steps; 2.2 GB of bfloat16 weights on 1 GPU. Serve it with bash scripts/servevllm.sh THUSI-Lab/GameScaling-0.8B-S3 8000 (§1). The inference client, prompts and scripts are in THUSI-Lab/GameScaling; all eight checkpoints are in the 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… Open weights apache-2.0 1.1B parameters 262,144 tokens transformers [View model](https://savrn.com/models/gamescaling-0-8b-s3) Model · Feature extraction ### [ls_resnet_small_ge2e](https://savrn.com/models/ls-resnet-small-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/ls-resnet-small-ge2e) Model · Text generation ### [Network-Cloud-Ops-Engineer-9B](https://savrn.com/models/network-cloud-ops-engineer-9b) [Reponx](https://savrn.com/model-publishers/reponx) Reponx 9B is a specialized AI model for network engineering, network operations, cloud networking, troubleshooting, and infrastructure operations. It is designed to help network and cloud engineers investigate technical issues using an evidence-first and safety-conscious troubleshooting approach. 1. Diagnose the likely problem. 2. Gather evidence before making changes. 3. Recommend read-only checks and commands first. 4. Explain what the results mean. 5. Suggest controlled remediation when appropriate. 6. Include validation and rollback considerations. Reponx 9B is a PEFT/LoRA fine-tuned model based on Qwen/Qwen3.5-9B. Load the Qwen3.5-9B base model with the Reponx adapter: Reponx provides… Open weights apache-2.0 peft [View model](https://savrn.com/models/network-cloud-ops-engineer-9b) Model · Feature extraction ### [univ_conformer_small_sphereface](https://savrn.com/models/univ-conformer-small-sphereface) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-sphereface) Model · Feature extraction ### [ls_conformer_small_pfas_ge2e](https://savrn.com/models/ls-conformer-small-pfas-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/ls-conformer-small-pfas-ge2e) Model · Feature extraction ### [ls_ecapa_tdnn_tiny_ge2e](https://savrn.com/models/ls-ecapa-tdnn-tiny-ge2e) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 626,720 parameters flaxspeaker [View model](https://savrn.com/models/ls-ecapa-tdnn-tiny-ge2e) Model · Feature extraction ### [univ_conformer_small_cosface](https://savrn.com/models/univ-conformer-small-cosface) [Quan](https://savrn.com/model-publishers/wq2012) This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget. Training-crop provenance. This checkpoint was trained with fixed 180-frame (1.8 s) crops and all-ones validity masks, not with the U(90, 180) variable-length crops that are the current pipeline default: the training code in use at the time defaulted the batchers to… Open weights apache-2.0 2M parameters flaxspeaker [View model](https://savrn.com/models/univ-conformer-small-cosface) ## Questions ### Can I use Apache License 2.0 models commercially? Yes. 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. ### Which Apache License 2.0 models are most downloaded? By monthly downloads reported by the Hugging Face Hub: AGiXT-Qwen3-VL-4B-GGUF (84); GNOSIS-Krikri-8B-Instruct-GGUF (84); granite-4.0-1b-speech-int8-mlx (84). ## Other Licenses - [mit](https://savrn.com/models/licenses/mit) 1319 - [cc-by-nc-4.0](https://savrn.com/models/licenses/cc-by-nc-4-0) 213 - [cc-by-4.0](https://savrn.com/models/licenses/cc-by-4-0) 164 - [bsd-3-clause](https://savrn.com/models/licenses/bsd-3-clause) 108 - [gemma](https://savrn.com/models/licenses/gemma) 63 - [cc-by-nc-sa-4.0](https://savrn.com/models/licenses/cc-by-nc-sa-4-0) 55 - [llama3.2](https://savrn.com/models/licenses/llama3-2) 40 - [unknown](https://savrn.com/models/licenses/unknown) 33 - [agpl-3.0](https://savrn.com/models/licenses/agpl-3-0) 32 - [llama3.1](https://savrn.com/models/licenses/llama3-1) 28 - [cc-by-sa-4.0](https://savrn.com/models/licenses/cc-by-sa-4-0) 20 - [gpl-3.0](https://savrn.com/models/licenses/gpl-3-0) 20 [See all](https://savrn.com/models/licenses)