# Open-Weight Models Under Apache License 2.0, Page 21 Source: https://savrn.com/models/licenses/apache-2-0/page/21 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 21 of 56 Model · Text generation ### [tiny-agent-gguf](https://savrn.com/models/tiny-agent-gguf) [Blockchain](https://savrn.com/model-publishers/hkblockchaintechnology) Quantized language models for Tiny Agent — a Windows application that gives you Word, Excel and PowerPoint with an AI assistant that runs on your own CPU, with no GPU, no cloud API and no account. double-click, and it installs like any other Windows program. This repository supplies the model that installer uses — or that you can point any local llama-server at. These files are quantizations of existing Apache-2.0 models, not new models. and limitations are inherited unchanged from the base model. Sizes are the real blob sizes, not estimates. All three are Apache-2.0 and safe to redistribute inside a product. Measured by actually running it — these are observed behaviours, not claims… Open weights apache-2.0 llama.cpp [View model](https://savrn.com/models/tiny-agent-gguf) Model ### [Zuzu-the-cat-hindi-GGUF](https://savrn.com/models/zuzu-the-cat-hindi-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/dare43321/Zuzu-the-cat-hindi 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… Open weights apache-2.0 transformers [View model](https://savrn.com/models/zuzu-the-cat-hindi-gguf) Model · Speech recognition ### [qwen3-asr-1.7b-int8-mlx](https://savrn.com/models/qwen3-asr-1-7b-int8-mlx) [App Automaton](https://savrn.com/model-publishers/appautomaton) MLX-native int8 conversion of Qwen3-ASR-1.7B for local automatic speech recognition on Apple Silicon — English, Chinese, and mixed Chinese/English. It runs through the mlx-speech runtime with no PyTorch and no cloud API at inference time. Weights ship as plain.safetensors. Install mlx-speech, then load by alias (downloads on first use): Or download the weights once and load by local path: - This repo contains the quantized MLX runtime artifact only. - The quantization mode (affine, groupsize 64, 8-bit) is stored in config.json and re-applied automatically on load, so it cannot desync from the weights. - v0 is single-pass offline ASR; streaming, timestamps, and long-audio chunking are… Open weights apache-2.0 2.4B parameters mlx [View model](https://savrn.com/models/qwen3-asr-1-7b-int8-mlx) Model · Text generation ### [stark-mcp-scratch](https://savrn.com/models/stark-mcp-scratch) [Victor cherif](https://savrn.com/model-publishers/vcheriffit) Modelo de lenguaje entrenado desde cero (pesos aleatorios) por Victor Cherif. Arquitectura tipo Llama de ~110M de parametros, pensado como asistente en espanol orientado a ciencia, programacion, robotica y fisica, con soporte de tool-calling / MCP entrenado contra servidores MCP reales. RMSNorm + SwiGLU - Al ser un modelo pequeno (~110M), el lenguaje puede ser incoherente en respuestas largas, sobre todo pasados los primeros parrafos (tiende a repetir bloques o rellenar con generalidades). - La aritmetica mental sigue sin ser confiable: depende de que el modelo dispare la tool calculate en vez de calcular "de memoria" -- verificado que esto mejoro pero no es perfecto. - El razonamiento ( )… Open weights apache-2.0 88M parameters 16,384 tokens transformers [View model](https://savrn.com/models/stark-mcp-scratch) Model · Image and text to text ### [Gemma4-E2B-it-W4A16](https://savrn.com/models/gemma4-e2b-it-w4a16) [88plug](https://savrn.com/model-publishers/88plug) These weights are compressed-tensors (pack-quantized / int-quantized). Do not deploy via TGI “text-generation-inference” paths — that backend does not load our CT format and produces opaque worker/load errors. INT4 post-training quantization of google/gemma-4-e2b-it — Google's 2B-active multimodal MoE with 128 experts. The smallest capable multimodal MoE. Runs on any 8 GB GPU. Quantized with datafree RTN (QuantizationModifier) — AutoRound (blocked — see KNOWN-FAILURES) wrapper blocked by Gemma4 qproj.linear children (see KNOWN-FAILURES). Temporary path until upstream fixed. The LLM backbone is W4A16; vision tower, projector, and PLE layers remain BF16. embedtokensperlayer and… Open weights apache-2.0 5.1B parameters 131,072 tokens transformers [View model](https://savrn.com/models/gemma4-e2b-it-w4a16) Model · Image classification ### [PlantNet-300K-ResNet18-LiteRT](https://savrn.com/models/plantnet-300k-resnet18-litert) [LiteRT Community (FKA TFLite)](https://savrn.com/model-publishers/litert-community) Measured on device (edge-compat): Raspberry Pi 5 · LiteRT 2.2.0.dev20260804 · CPU/XNNPACK, 4 threads · 34.1 ms p50 (2026-08-31); browser · Chromium 151 (2026-08-11). Record: https://github.com/john-rocky/edge-compat/blob/main/cards/plantnet-300k-resnet18/CARD.md On-device fine-grained plant species identification — 1081 species — running fully on the LiteRT CompiledModel GPU delegate (no CPU fallback). A PlantNet-300K (NeurIPS 2021) ResNet18. ~16 ms/frame on a Pixel 8a. (mean [0.485,0.456,0.406], std [0.229,0.224,0.225]; center-crop then resize 224). Labels: class index i maps to the i-th species when the PlantNet-300K species-id strings are sorted (torchvision ImageFolder order); names… Open weights apache-2.0 litert [View model](https://savrn.com/models/plantnet-300k-resnet18-litert) Model ### [naps-llama-3_1_instruct-v0.6.0-GGUF](https://savrn.com/models/naps-llama-3-1-instruct-v0-6-0-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/NAPS-ai/naps-llama-31instruct-v0.6.0 For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/naps-llama-31instruct-v0.6.0-i1-GGUF 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 And here are Artefact2's thoughts on the matter… Open weights apache-2.0 transformers [View model](https://savrn.com/models/naps-llama-3-1-instruct-v0-6-0-gguf) Model · Text generation ### [qwen3.8-27b-buddhist](https://savrn.com/models/qwen3-8-27b-buddhist) [Wulezhi](https://savrn.com/model-publishers/wulezhi) 基于 Qwen3.8-27B（qwen35 架构，稠密 27B，48 线性注意力 + 16 全注意力，带视觉模块）的佛学辞典微调模型， 使用 佛光大辞典 与 中华佛教大百科全书 的语料做 35,841 条中文佛教 SFT 微调， 各目录的 quantizationconfig 不同（quantmethod 各异），加载命令需对应： 未量化的层主要是三类（各版本 config.json 的 ignore 字段可查）： 1. 线性注意力（Gated DeltaNet / linearattn）投影：nvfp4a16 明确忽略了 240 条、autoawq 忽略 inproja / inprojb —— 量化工具链对线性注意力算子支持不完整，硬量化容易掉点； 2. 视觉塔 visual（约 110 个模块）：对精度敏感，且纯文本部署可用 --language-model-only 直接跳过； 3. lmhead 与各 norm / bias / embedding：量化收益小、风险大。 - ① SM75 没有 FP8 硬件能力（FP8 张量核从 Ada / SM89 才有）。所以在 2080Ti 上 FP8 权重只能走社区 fork 的模拟路径（Marlin fallback + 实时反量化），实际行为是"权重 FP8 + 计算 BF16"（weight-only），激活不参与量化 —— 这也是为什么单独做了一份 block 128×128 的 FP8（面向通用 vLLM 的 per-channel 那份在 SM75 上加载不了）。 - ② 覆盖率低 ≠ 残次品：BF16… Open weights apache-2.0 peft [View model](https://savrn.com/models/qwen3-8-27b-buddhist) Model · Image and text to text ### [Qwopus3.8-27B-Flash-V2-EXL3-3.85bpw-24GB](https://savrn.com/models/qwopus3-8-27b-flash-v2-exl3-3-85bpw-24gb) [Jorge Ramírez Espejo](https://savrn.com/model-publishers/jramirez-uab) An EXL3 quantization of Jackrong/Qwopus3.8-27B-Flash-V2 made with exllamav3's per-tensor bit-allocation ("SC") pipeline with our own calibration, a 5-bit lmhead, a 6-bit vision tower, a 4-bit MTP head and INT8 token embeddings. It is served with vLLM (continuous batching) through the OrcaSAQ2 plugin, with the base Qwen3.8 chat template for reliable tool calls. KL divergence against BF16: 0.0060 (≈6× lower than our W4A16 GPTQ version). It runs a 215k-token context with vision and MTP speculative decoding on a single 24 GB RTX 3090; the 3.70bpw branch (since 2026-10-14) runs 226k with vision and 4 sessions (see Variants). Highlights (one RTX 3090 24 GB, measured on 2026-10-14; details in… Open weights apache-2.0 7.9B parameters 262,144 tokens exllamav3 [View model](https://savrn.com/models/qwopus3-8-27b-flash-v2-exl3-3-85bpw-24gb) Model ### [rhymeai-gemma-4-gguf](https://savrn.com/models/rhymeai-gemma-4-gguf) [Tyler Manring](https://savrn.com/model-publishers/shayde182) LoRA fine-tunes of Gemma 4 E2B-it and E4B-it for RhymeAI / Writers Block, a native songwriting app. Trained to follow the app's exact on-device prompt rewrites, single-word rewrites, and synonyms. Runs fully offline via llama.cpp. QLoRA (r=16, alpha=16, 2 epochs, lr 2e-4 cosine, effective batch 16, prompt-masked loss) on 4,947 synthetic examples generated from original seed lyrics and machine-verified for rhyme + format correctness before training (CMU-dictionary rhyme checking). No copyrighted lyrics and no user data. Trained with Unsloth on a Colab L4; merged and quantized with llama.cpp (llama-export-lora + llama-quantize, Q4KM). model\n), stop string. Open weights apache-2.0 [View model](https://savrn.com/models/rhymeai-gemma-4-gguf) Model · Text to speech ### [Qwen3-TTS-12Hz-0.6B-Base-OpenVINO](https://savrn.com/models/qwen3-tts-12hz-0-6b-base-openvino) [X Girls](https://savrn.com/model-publishers/xgirls) A ready-to-use OpenVINO IR export of Qwen/Qwen3-TTS-12Hz-0.6B-Base, prepared for Fern. The export preserves the upstream model weights and architecture; it is not fine-tuned. Generate speech and clone a voice from reference audio, subject to the source model’s supported languages and input requirements. 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 directory on PYTHONPATH and the runtime… Open weights apache-2.0 openvino [View model](https://savrn.com/models/qwen3-tts-12hz-0-6b-base-openvino) Model · Text to speech ### [step-audio-editx-8bit-mlx](https://savrn.com/models/step-audio-editx-8bit-mlx) [App Automaton](https://savrn.com/model-publishers/appautomaton) This repository contains a self-contained pure-MLX int8 conversion of Step-Audio-EditX for local voice cloning and expressive audio editing on Apple Silicon. All pipeline components are stored as.safetensors — no PyTorch, ONNX, or NumPy files are required at inference time. This bundle is self-contained — all weights are packaged in one repository. Five-stage pipeline, all running pure MLX with bf16 activations: 1. Step1 LM (3.5B params, int8) — autoregressive dual-codebook token generation 2. CampPlus (bf16) — speaker embedding extraction from reference audio 3. Flow conditioner (bf16) — conditions generation on speaker embedding 4. Flow model (int8) — flow-matching mel spectrogram… Open weights apache-2.0 3.5B parameters mlx [View model](https://savrn.com/models/step-audio-editx-8bit-mlx) Model · Text generation ### [Olmo-3-7B-Think-heretic-GGUF](https://savrn.com/models/olmo-3-7b-think-heretic-gguf) [Richard Young](https://savrn.com/model-publishers/richardyoung) GGUF quantizations of richardyoung/Olmo-3-7B-Think-heretic, an abliterated build of allenai/Olmo-3-7B-Think (Ai2's fully open 7B reasoning model (open weights, data and training code) that thinks before answering) made with Heretic. The very low KL divergence (0.026, far below the 0.5 “damage” threshold) means the model retains essentially all of its original capabilities and coherence. Abliteration removes the "refusal direction" from a model's residual stream — a surgical modification that disables safety refusals without retraining. See Refusal in Language Models Is Mediated by a Single Direction. Heretic automates it: it searches ablation parameters with an optimizer (200 trials here)… Open weights apache-2.0 [View model](https://savrn.com/models/olmo-3-7b-think-heretic-gguf) Model · Image and text to text ### [Qwen3.8-27B-TTURBO-Cold-Fusion-709-L-Uncensored-mlx-oQ6e](https://savrn.com/models/qwen3-8-27b-tturbo-cold-fusion-709-l-uncensored-mlx-oq6e) [Solstice-AI](https://savrn.com/model-publishers/solstice-ai) Solstice-AI/Qwen3.8-27B-TTURBO-Cold-Fusion-709-L-Uncensored-mlx-oQ6e is the official Apple Silicon MLX release of DavidAU's Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored. Engineered specifically for Apple Silicon Unified Memory architectures (M2, M3, M4, M5 Max / Ultra), this release couples oQ6e imatrix-guided mixed-precision with Solstice's proprietary 10-Level Cognitive Variable Thinking System and native Hardware Multi-Token Prediction (MTP). This checkpoint embeds the Solstice 10-Level Cognitive Jinja chat template, allowing instant dynamic control over reasoning depth Open weights apache-2.0 27.8B parameters 262,144 tokens mlx [View model](https://savrn.com/models/qwen3-8-27b-tturbo-cold-fusion-709-l-uncensored-mlx-oq6e) Model ### [PhishMe-12k-Qwen3.5-4B-GGUF](https://savrn.com/models/phishme-12k-qwen3-5-4b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/piyawudk/PhishMe-12k-Qwen3.5-4B 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… Open weights apache-2.0 transformers [View model](https://savrn.com/models/phishme-12k-qwen3-5-4b-gguf) Model · Image and text to text ### [Qwopus3.6-35B-A3B-Coder-EXL3-4.00bpw-24GB](https://savrn.com/models/qwopus3-6-35b-a3b-coder-exl3-4-00bpw-24gb) [Jorge Ramírez Espejo](https://savrn.com/model-publishers/jramirez-uab) An EXL3 quantization of Jackrong/Qwopus3.6-35B-A3B-Coder made with exllamav3's per-tensor bit-allocation ("SC") pipeline, with a 6-bit lmhead, 6-bit vision tower, 4-bit MTP head and INT8 token embeddings. Jackrong's thinking-off coding/agent fine-tune of Qwen3.6-35B-A3B (3B active). It is the easiest of the three MoEs to quantize: 3.75 bpw is already below 0.01 KLD. It is served with vLLM (continuous batching) through the OrcaSAQ2 EXL3 plugin plus our plugin qwen38exl3vision (0.3.2), which adds EXL3 support for the MoE experts (fused exllamav3 kernel). On a single RTX 3090 (24 GB) the main revision runs the full 262,144-token context with vision, MTP and 2 concurrent sessions with ≈1 GB of… Open weights apache-2.0 9.7B parameters 262,144 tokens exllamav3 [View model](https://savrn.com/models/qwopus3-6-35b-a3b-coder-exl3-4-00bpw-24gb) Model · Text generation ### [mindmopai-agent](https://savrn.com/models/mindmopai-agent) [Vamsi Krishna Kosuri](https://savrn.com/model-publishers/vamsikrishnakosuri) The small on-device model inside MindMopAI, a free brain-dump app for Windows and Mac. It does two jobs, and runs entirely on your computer through llama.cpp. Nothing you say is sent anywhere. 1. Sorting. You say whatever is on your mind, and it splits it into separate items and sorts each one: task, bill, exam, idea, note, shopping, read, social or appointment. 2. Mai, the in-app helper. It answers questions about MindMopAI from a help sheet the app provides, reads your own items and settings to answer questions about them, offers actions (for example a theme picker, or "mark this done" for you to confirm), and reviews your task list with practical suggestions. 2,000 helper examples, all… Open weights apache-2.0 [View model](https://savrn.com/models/mindmopai-agent) Model · Image and text to text ### [Qwopus3.8-27B-Flash-mlx-oQ8e-1M](https://savrn.com/models/qwopus3-8-27b-flash-mlx-oq8e-1m) [Solstice-AI](https://savrn.com/model-publishers/solstice-ai) Solstice-AI/Qwopus3.8-27B-Flash-mlx-oQ8e-1M provides the official, production-grade Apple Silicon MLX (oQ8e) release of Qwopus3.8-27B-Flash with a native-behaving 1,048,576-token (1M) context window. Official Apple Silicon MLX oQ8e mixed-precision release tailored for macOS unified memory execution with native-behaving 1M context. Evaluated under the standardized benchmark harness: 1M YaRN Scaling & Quantization Suite: Solstice-AI Open weights apache-2.0 27.8B parameters 1,048,576 tokens mlx [View model](https://savrn.com/models/qwopus3-8-27b-flash-mlx-oq8e-1m) Model ### [Ministral-3-14B-lil-logbook](https://savrn.com/models/ministral-3-14b-lil-logbook) [Thestarfarer](https://savrn.com/model-publishers/thestarfarer) Seven LoRA adapters trained on the same dataset, one variable changed at a time: the three original runs below, plus a four-point learning-rate ladder under lr-sweep/. F16/Q80 GGUFs for the original three. The original three runs were evaluated on loss curves and spot generations only; the ladder adds an out-of-domain loss axis and a generation-level composition probe (sections below). Multi-turn dialogue packed to 512-token samples — 13,667 samples, 6.16M tokens total. Turns are delimited by headers — short per-character abbreviations, not names. Unattributed narration is folded into the speaker's turn: The pack was built in two phase-shifted passes — the second pass offsets every chunk… Open weights apache-2.0 peft [View model](https://savrn.com/models/ministral-3-14b-lil-logbook) Model · Text generation ### [Qwen3-4B-Q4_K_M-GGUF](https://savrn.com/models/qwen3-4b-q4-k-m-gguf) [Shayon Sarker](https://savrn.com/model-publishers/shayonsarker) Downloads: Hugging Face — ShayonSarker/Qwen3-4B-Q4KM-GGUF · Source & build recipes: GitHub — model-qwen3-4b-q4-k-m-gguf · Base model: Qwen/Qwen3-4B The GGUF binaries are hosted on Hugging Face; this repository holds the reproducible conversion pipeline, calibration data, and checksums. Base: Qwen/Qwen3-4B (Apache-2.0) · Quant: llama.cpp 0.4.1-dev (60081bb) · imatrix on 3000-line wikitext-2 (seed 42, 618 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 not run: perplexity only, stated honestly. Files: smol-Q4KM.gguf, smol-Q80.gguf, imatrix.dat, calib.txt, REPRO.json, SHA256.txt. Full repro: see… Open weights apache-2.0 [View model](https://savrn.com/models/qwen3-4b-q4-k-m-gguf) Model · Text generation ### [Muta-Tutor-Qwen2.5-1.5B-ADTC-GGUF](https://savrn.com/models/muta-tutor-qwen2-5-1-5b-adtc-gguf) [Timi Owolabi](https://savrn.com/model-publishers/timiiowolabi) This is the vector-configuration finalist from Muta's ADTC fine-tuning campaign. It starts from Qwen/Qwen2.5-1.5B-Instruct, applies BF16 LoRA with rank 16 for 500 steps on a licence-clean multiple-choice math and science mixture, merges the adapter, and exports the result as Q4KM GGUF. Muta-Tutor-Qwen2.5-1.5B-Q4KM.gguf SHA-256: a750d00d458c6ab38925364ea1413db00648449180941e47025736d09922e1eb The candidate and untuned control were evaluated with the same GGUF conversion, quantization, prompt format, and GCP CPU-proxy harness. The repository includes the training manifest, licence-clean dataset manifest, artifact hashes, and the full fine-tuning summary. Performance was measured on a cloud… Open weights apache-2.0 [View model](https://savrn.com/models/muta-tutor-qwen2-5-1-5b-adtc-gguf) Model · Text generation ### [SRA-RiskGate-4B-Q4_K_M-GGUF](https://savrn.com/models/sra-riskgate-4b-q4-k-m-gguf) [Sriram Ramakrishnan](https://savrn.com/model-publishers/sriram1983007) This repository contains the quantized GGUF weights (Q4KM, ~2.5 GB) for SRA-RiskGate-4B, an autonomous risk scoring, compliance verification, and dispute resolution agent for stablecoin payments (EIP-3009, x402, smart escrow). Deploying with autonomous on-chain agents or payment gateways? Use the companion Python SDK for zero-latency deterministic pre-filtering (sanctions, address validation, depeg protection) and drop-in Coinbase AgentKit firewall support: Note: Always run with greedy decoding (--temp 0.0) to guarantee deterministic JSON output conforming to the benchmark schema. Distributed under the Apache 2.0 License. Open weights apache-2.0 gguf [View model](https://savrn.com/models/sra-riskgate-4b-q4-k-m-gguf) Model · Tabular classification ### [cernpeerenv-zmumu-dqn](https://savrn.com/models/cernpeerenv-zmumu-dqn) [D](https://savrn.com/model-publishers/dhdrl) Task. Binary classification of simulated dimuon events: does this event contain an opposite-sign muon pair whose invariant mass falls in (80, 100) GeV? Method. The classifier is trained with Stable-Baselines3 DQN or PPO inside a one-step Gymnasium wrapper. Each reset / step is one labeled event and then terminated=True. There is no horizon, no delayed credit assignment, and no control of a detector. Label. Computed from the same padded muon kinematics the policy sees: at least two muons, opposite charge, invariant mass in (massmin, massmax) with defaults 80–100 GeV. This is not generator-level Z truth and not a CMS official tag. Data for the reported numbers. Pythia8 Monte Carlo only… Open weights apache-2.0 stable-baselines3 [View model](https://savrn.com/models/cernpeerenv-zmumu-dqn) Model · Token classification ### [stanza-th](https://savrn.com/models/stanza-th) [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-th) Model · Text generation ### [szl-receiptagent-qwen35-0.8b-v3](https://savrn.com/models/szl-receiptagent-qwen35-0-8b-v3) [SZL Holdings](https://savrn.com/model-publishers/szlholdings) This is a separate research bf16 LoRA adapter (r=16, alpha=32) on Qwen/Qwen3.5-0.8B, trained with Unsloth on a laptop Blackwell GPU, with a derived merged artifact also present in the repository. "Placeholder" describes release status, not absence of weights. It does not overwrite ReceiptAgent v2 Sizes were read from the Hub listing; no weight file was downloaded or hashed by this pass, and the revision SHA was not captured. The Hub-hosted banner referenced by the szl-forge card (holo-banner.svg) is not on this tree, so it is not embedded here. The training receipt records 180 training rows and train loss 0.0537. Training loss is not evaluation. The receipt describes the local laptop… Open weights apache-2.0 752M parameters 262,144 tokens peft [View model](https://savrn.com/models/szl-receiptagent-qwen35-0-8b-v3) Model · Image and text to text ### [Qwen3.6-35B-A3B-EXL3-4.00bpw-24GB](https://savrn.com/models/qwen3-6-35b-a3b-exl3-4-00bpw-24gb) [Jorge Ramírez Espejo](https://savrn.com/model-publishers/jramirez-uab) An EXL3 quantization of Qwen/Qwen3.6-35B-A3B made with exllamav3's per-tensor bit-allocation ("SC") pipeline, with a 6-bit lmhead, 6-bit vision tower, 4-bit MTP head and INT8 token embeddings. Qwen's 35B-A3B MoE (3B active parameters, 256 experts, hybrid GatedDeltaNet attention, vision and MTP). We did not find another EXL3 of it that keeps vision and MTP and runs under vLLM. It is served with vLLM (continuous batching) through the OrcaSAQ2 EXL3 plugin plus our plugin qwen38exl3vision (0.3.2), which adds EXL3 support for the MoE experts (fused exllamav3 kernel). On a single RTX 3090 (24 GB) the main revision runs the full 262,144-token context with vision, MTP and 2 concurrent sessions with… Open weights apache-2.0 9.7B parameters 262,144 tokens exllamav3 [View model](https://savrn.com/models/qwen3-6-35b-a3b-exl3-4-00bpw-24gb) Model · Image and text to text ### [Qwen3.8-27B-pi-EXL3-3.85bpw-24GB](https://savrn.com/models/qwen3-8-27b-pi-exl3-3-85bpw-24gb) [Jorge Ramírez Espejo](https://savrn.com/model-publishers/jramirez-uab) An EXL3 quantization of bytkim/Qwen3.8-27B-pi (bytkim's agent/coding fine-tune of Qwen3.8-27B for the Pi harness), made with exllamav3's per-tensor bit-allocation ("SC") pipeline calibrated on a trace generated by the pi model itself, with a 5-bit lmhead, a 6-bit vision tower, a 4-bit MTP head and INT8 token embeddings. It is served with vLLM (continuous batching) through the OrcaSAQ2 plugin and runs a 215k-token context with vision and MTP speculative decoding on a single 24 GB RTX 3090, with a KL divergence of 0.0074 against BF16 (0.0070 on turboderp's public trace). The 3.70bpw branch trades a little fidelity (KLD 0.0093) for a 226k context. Same recipe, plugins and serving command as… Open weights apache-2.0 7.9B parameters 262,144 tokens exllamav3 [View model](https://savrn.com/models/qwen3-8-27b-pi-exl3-3-85bpw-24gb) Model · Image and text to text ### [Qwen3.5-4B-RKLLM](https://savrn.com/models/qwen3-5-4b-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Image and text on Rockchip RK3576 and RK3588, converted from Qwen/Qwen3.5-4B. 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. For image input, also download the vision encoder for the same SoC. Keep the language model and vision encoder from this release together. ONNX files are export intermediates; they are not needed to run these RKNN/RKLLM models. 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 files.… Open weights apache-2.0 rkllm [View model](https://savrn.com/models/qwen3-5-4b-rkllm) Model · Text generation ### [gemma-4-E4B-meeting-agent-zh-LiteRT](https://savrn.com/models/gemma-4-e4b-meeting-agent-zh-litert) [Luigi](https://savrn.com/model-publishers/luigi) Gemma-4-E4B fine-tuned to read a live zh-TW meeting transcript on a phone, window by window, and write short cited notes: decisions, actions, proposals, open issues and key figures. The same model turns the notes into a prose summary and a title. It is the E4B sibling of mobile-v1: same prompts, protocol and note format, fewer factual errors, about 3× slower. Latest: mobile-v2/ (2026-10-17), see the mobile-v2 section. The files at the root are mobile-v1. mobile-v2/ holds the E4B mobile-v2 reader. It continues mobile-v1 with two training stages: 1. SFT on reading rows plus dense-list conversion rows, including a third domain: Taiwanese financial TV programs; 2. one round of on-policy… Open weights apache-2.0 litert-lm [View model](https://savrn.com/models/gemma-4-e4b-meeting-agent-zh-litert) Model · Image and text to text ### [Qwen3.8-27B-nvfp4full-dflash2-NInfer](https://savrn.com/models/qwen3-8-27b-nvfp4full-dflash2-ninfer) [Kaushik Vira](https://savrn.com/model-publishers/kaushikvira) This repository contains the nvfp4full weight profile of Qwen3.8-27B in the native NInfer.ninfer artifact format, with the z-lab DFlash2 speculative-decoding drafter module added in the upstream W8G32/BF16 format. It is the qwen3.8-27b / nvfp4full identity (same as with the registered DFlash2 module appended — the base tensors are byte-for-byte identical to cometkim v1, and the MTP module is retained (unused and validate-only under --spec dflash2). The base tensors are copied from cometkim v1 byte-for-byte (no re-encoding); only the 66 DFlash2 module objects are produced by the graft tool from the z-lab checkpoint. Verified: source identity qwen3.8-27b/nvfp4full, 1,259 source objects, no… Open weights apache-2.0 ninfer [View model](https://savrn.com/models/qwen3-8-27b-nvfp4full-dflash2-ninfer) Model ### [firela-pa-pc](https://savrn.com/models/firela-pa-pc) [Ff](https://savrn.com/model-publishers/fxiafx) firela-pa 的 PC 形态分发仓：Qwen3-1.7B 任务微调隐私路由模型（fmt-v1.1 五意图： tx 查账 / pf 资产全景 / mkt 行情 / l 本地闲聊 / c 涂黑上云），W8A8 板载同源权重的 GGUF 版（Q4KM 为默认分发档，Q80 为质量档；v1.3 权重 eval200 200 题 G4 98.0）。 自动完成：装/起 Ollama → 下载本仓应用与路由模型（sha256 校验）→ 注册 firela-router → 拉生成模型 qwen2.5:3b-instruct（非思考原生 ~1s；--no-gen 可跳过，可在 config 换 qwen3:4b 等思考模型——Ollama think 参数对库版 qwen3 模板不生效，慎用）→ 交互式生成 0600 配置 （vlt / relay 凭证，回车可留空后补）→ Time Machine 排除 → firela-pa 命令 → 冒烟。 隐私边界：推理不出门——路由/闲聊生成全本地；查账走用户自己的 vlt 令牌；上云问句 本仓全部产物——pc-app.tar.gz（应用代码）、install.sh、 router-merged-Q4KM.gguf、router-merged-Q80.gguf——均以 Apache License 2.0 提供（许可证全文见仓根 LICENSE）。 路由模型为 Qwen3-1.7B 的 LoRA 任务微调 合并衍生物。基座 © 2024 Alibaba Cloud（Qwen team），以 Apache-2.0 提供；再分发… Open weights apache-2.0 gguf [View model](https://savrn.com/models/firela-pa-pc) Model · Image and text to text ### [Qwopus3.8-27B-Flash-NVFP4-1M](https://savrn.com/models/qwopus3-8-27b-flash-nvfp4-1m) [Solstice-AI](https://savrn.com/model-publishers/solstice-ai) Solstice-AI/Qwopus3.8-27B-Flash-NVFP4-1M provides the official, production-grade NVIDIA NVFP4 release of Qwopus3.8-27B-Flash with a native-behaving 1,048,576-token (1M) context window. NVIDIA Blackwell-native NVFP4 Tensor Core quantization with baked 1M YaRN scaling, running without runtime flag overhead. Evaluated under the standardized benchmark harness: 1M YaRN Scaling & Quantization Suite: Solstice-AI Open weights apache-2.0 19.2B parameters 1,048,576 tokens [View model](https://savrn.com/models/qwopus3-8-27b-flash-nvfp4-1m) Model · Text classification ### [laya-LiteRT](https://savrn.com/models/laya-litert) [LiteRT Community (FKA TFLite)](https://savrn.com/model-publishers/litert-community) A state in, calibrated answers to typed questions out, no text generated: choice (one of the criteria), score (an ordered scale), noul (does the statement hold, as a probability). One forward per question. These are the convaiinnovations/laya checkpoints (revision 1c5edc17) converted to classic LiteRT graphs with one static window per file: the English model is ModernBERT-large with laya's decision head (421M parameters), the multilingual model is mmBERT-base with the same head (322M; the publisher lists 100+ languages). The request and answer forms are laya's own. Tested on a Samsung Galaxy S26 (SM-S942Q, Android 16 BP4A.251205.006) with LiteRT 2.2.0 CompiledModel, GPU with explicit FP32… Open weights apache-2.0 litert [View model](https://savrn.com/models/laya-litert) Model · Image and text to text ### [NeoHorse-1-9B-EXL3-16GB-12GB](https://savrn.com/models/neohorse-1-9b-exl3-16gb-12gb) [Jorge Ramírez Espejo](https://savrn.com/model-publishers/jramirez-uab) An EXL3 quantization of TokenRhythm/NeoHorse-1-9B (an agent-oriented finetune of Qwen3.5-9B) made with exllamav3, in two sizes: - main — 6.00 bpw, for 16 GB GPUs (e.g. RTX 5060 Ti 16 GB / 4060 Ti 16 GB): the full 262,144-token context with vision, MTP speculative decoding and 2 concurrent sessions, KL divergence 0.0015 against BF16. - 4.50bpw — 4.50 bpw, for 12 GB GPUs (e.g. RTX 3060 12 GB): 176,000 tokens with vision, KLD 0.0058. Both are served with vLLM (continuous batching) through the OrcaSAQ2 plugin and a small plugin of ours (included), with INT8 token embeddings. Why this model: we evaluate small models as a Linux sysadmin agent driven by opencode, used in Spanish. Among the 9B… Open weights apache-2.0 4.4B parameters 262,144 tokens exllamav3 [View model](https://savrn.com/models/neohorse-1-9b-exl3-16gb-12gb) Model ### [MMFineReason-4B-GGUF](https://savrn.com/models/mmfinereason-4b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/OpenDataArena/MMFineReason-4B For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/MMFineReason-4B-i1-GGUF 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 And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9 See… Open weights apache-2.0 transformers [View model](https://savrn.com/models/mmfinereason-4b-gguf) Model ### [GraphGen-27b-GGUF](https://savrn.com/models/graphgen-27b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/anonymous192837192873/GraphGen-27b 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/graphgen-27b-gguf) Model ### [TestSubject-1](https://savrn.com/models/testsubject-1) [Mr. Black Beard](https://savrn.com/model-publishers/cyanmonkey) Rose X1.5 SLM (148.8M params). Access requested at publisher apache-2.0 149M parameters [View model](https://savrn.com/models/testsubject-1) Model · Text classification ### [Darwin-27B-ZTC-v5](https://savrn.com/models/darwin-27b-ztc-v5) [FINAL_Bench](https://savrn.com/model-publishers/final-bench) Darwin-27B-ZTC-v5 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-v2. All 137 benchmarks complete, measured with the S1MB evaluator and its unchanged autojev adapter. The numbers follow the leaderboard's own code (viewer/src/lib/borda.ts) applied to the published results (150 models, 2026-10-11) with this model added. Submission pending… Open weights apache-2.0 25.6B parameters 262,144 tokens pytorch [View model](https://savrn.com/models/darwin-27b-ztc-v5) Model · Text generation ### [logi-1.5b-gguf](https://savrn.com/models/logi-1-5b-gguf) [Alves](https://savrn.com/model-publishers/guiiwfz) Versões quantizadas em GGUF do LOGI 1.5B, um modelo educacional em português brasileiro criado para o jogo Decifra.IA. O LOGI atua como tutor de: - lógica proposicional: AND, OR, NOT, IMPLICA, BICONDICIONAL, tabelas-verdade, equivalências, falácias e regras de inferência; - alfabetização em inteligência artificial: modelos, dados, treinamento, inferência, vieses, alucinação, RAG, privacidade e uso responsável; - segurança digital relacionada à IA: phishing, deepfakes, engenharia social, prompt injection e proteção de dados. Este repositório contém as quantizações da versão v5, treinada com 13.000 exemplos. Ela substitui a versão inicial do LOGI, que utilizava 1.183 exemplos. Para… Open weights apache-2.0 1.5B parameters 32,768 tokens [View model](https://savrn.com/models/logi-1-5b-gguf) Model · Text classification ### [QJev3.5-0.8B-GGUF](https://savrn.com/models/qjev3-5-0-8b-gguf) [Ziyi LI](https://savrn.com/model-publishers/zy-lee4) This is a 0.8B model we fine-tuned to make one narrow kind of decision very cheaply. It reads a state, a question, and a list of options, then returns a probability for each option. It never writes text. One forward pass, one token, and you have a distribution your code can branch on. We built it for our own product first. An educational chatbot needed to decide, before spending a full LLM call, whether a student message was on topic, whether a tool call was safe to run, and whether an old tool result still had to be kept verbatim in the context. Those are small bounded judgments, and paying a 70B model to make them was absurd. It runs today on an office GPU at 80 to 150 ms per decision… Open weights apache-2.0 [View model](https://savrn.com/models/qjev3-5-0-8b-gguf) Model · Text generation ### [hydra-nano-git](https://savrn.com/models/hydra-nano-git) [Rahul kumar](https://savrn.com/model-publishers/rahul05ranjan) A 0.75B-parameter local model that lets a coding agent hand off routine, read-only repository work. Hydra Nano is the model behind Hydra, a CLI that a frontier coding agent (Claude Code, Codex and others) calls instead of running noisy commands itself. Hydra asks the model to choose from a fixed menu of read-only probes, runs those probes in code, and returns the command's own output lines. The model never writes a shell command. One model serves all five Hydra heads: All figures are end-to-end Hydra results: the task counts as solved only when the probe plan is right and the returned evidence contains every expected fact with nothing forbidden. Errors and timeouts count as failures. Every… Open weights apache-2.0 752M parameters 262,144 tokens transformers [View model](https://savrn.com/models/hydra-nano-git) Model · Text generation ### [pulse-local-qwen3-0.6b](https://savrn.com/models/pulse-local-qwen3-0-6b) [Pulse Finance](https://savrn.com/model-publishers/pulseos) A 0.6B Qwen3 fine-tuned for one job: being the language layer of Pulse Local, a Mac app where a personal-finance assistant, its ledger and every question stay on the user's own machine. It is published so the app can download it on first launch. Beyond that download, the app's Local Mode makes only an optional check for a newer version, which can be switched off and sends nothing about the user. Its optional Connected Mode adds a bank feed through Plaid; even then the model, the ledger and every question stay on the machine. Read this before you try it as a general assistant: on its own it is not one. On our private 48-question finance benchmark this model alone scores about 55 out of 100.… Open weights apache-2.0 [View model](https://savrn.com/models/pulse-local-qwen3-0-6b) Model ### [Qwenslerp2-7B-GGUF](https://savrn.com/models/qwenslerp2-7b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/allknowingroger/Qwenslerp2-7B 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… Open weights apache-2.0 transformers [View model](https://savrn.com/models/qwenslerp2-7b-gguf) Model ### [QL-Specilize-78b-GGUF](https://savrn.com/models/ql-specilize-78b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/FINGU-AI/QL-Specilize-78b For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/QL-Specilize-78b-i1-GGUF 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 And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9 See… Open weights apache-2.0 transformers [View model](https://savrn.com/models/ql-specilize-78b-gguf) Model ### [agri-advisor-1.5b](https://savrn.com/models/agri-advisor-1-5b) [ADEGOKE ISRAEL](https://savrn.com/model-publishers/eyedol) Qwen2.5-1.5B-Instruct fine-tuned as an offline agricultural advisory assistant (crop diagnosis, fertilizer/market calculations, pest guidance) for smallholder farmers and extension officers. Fine-tuned from Qwen/Qwen2.5-1.5B-Instruct, converted to GGUF, and quantized for offline CPU inference via llama.cpp. Open weights apache-2.0 [View model](https://savrn.com/models/agri-advisor-1-5b) Model · Text generation ### [qwen3-0.6b-cefc-rft-vietnamese-legal](https://savrn.com/models/qwen3-0-6b-cefc-rft-vietnamese-legal) [Tran Thi THu Phuong](https://savrn.com/model-publishers/phuongntc) Reformatted Task 4.1 release used for external evaluation: datht/vlegal The paper contains the split sizes, training configuration, evaluation-set sizes, and reported results; they are not repeated here. - adaptermodel.safetensors — final CEFC-RFT LoRA weights; - adapterconfig.json — PEFT adapter configuration; - cefcrunconfig.json — exact controller and final-run configuration; - cefctrainingmetadata.json — metadata for the weighted LoRA continuation run; - selectionstate.json — exported selection state from the CEFC-RFT run; - calibratecefcstagegates.py — validation-only Stage-gate replay script; - chattemplate.jinja, tokenizer.json, tokenizerconfig.json, specialtokensmap.json… Open weights apache-2.0 peft [View model](https://savrn.com/models/qwen3-0-6b-cefc-rft-vietnamese-legal) Model · Feature extraction ### [univ_conformer_small_ge2e](https://savrn.com/models/univ-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/univ-conformer-small-ge2e) Model · Feature extraction ### [univ_lstm_small_ge2e](https://savrn.com/models/univ-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/univ-lstm-small-ge2e) Model ### [health-advisor-1.5b](https://savrn.com/models/health-advisor-1-5b) [ADEGOKE ISRAEL](https://savrn.com/model-publishers/eyedol) Qwen2.5-1.5B-Instruct fine-tuned as an offline patient-education and triage-support assistant. Not a diagnostic tool — designed to explain health information, flag red-flag symptoms, and defer specific dosing questions to a qualified clinician or pharmacist. Fine-tuned from Qwen/Qwen2.5-1.5B-Instruct, converted to GGUF, and quantized for offline CPU inference via llama.cpp. Open weights apache-2.0 [View model](https://savrn.com/models/health-advisor-1-5b) Model · Text to speech ### [SemBridge](https://savrn.com/models/sembridge) [ASLP-lab](https://savrn.com/model-publishers/aslp-lab) SemBridge bridges discrete semantic supervision and continuous acoustic generation. A shared semantic tokenizer anchors both acoustic latents and autoregressive language-model states, improving linguistic fidelity while retaining continuous-latent generation. The unified model supports text-to-speech, note-conditioned singing, and mixed speech and singing. - [2026-10]: Released the inference code and model checkpoints. - [2026-08]: Released the arXiv paper. Use Python 3.10. The repository includes 10 complete examples from our Demo page: 4 TTS, 4 SVS, and 2 Unified cases, with reference audio and prepared requests. SVS and Unified include aligned lyrics, pitches, durations, and… Open weights apache-2.0 818M parameters 32,768 tokens [View model](https://savrn.com/models/sembridge) Model ### [ATX-Swift-1.5-Qwen3.8-27B-Uncensored-MTP-GGUF](https://savrn.com/models/atx-swift-1-5-qwen3-8-27b-uncensored-mtp-gguf) [Alejandro Melo](https://savrn.com/model-publishers/bjivanovich) Custom ATX GGUF quantizations of Swift-1.5-Qwen3.8-27B-Uncensored-MTP featuring an embedded Multi-Token Prediction (MTP) draft head in blk.64, specifically optimized for high-throughput inference in llamAmpere and standard llama.cpp. While compatible with upstream llama.cpp, this model family is engineered for llamAmpere, a specialized high-performance fork tuned for NVIDIA Ampere (SM86) and Ada Lovelace architectures (RTX 3090 / RTX 4090). - MMA Fused Q8 Kernels (GGMLQ8TURBO3MMAFUSED="1"): Eliminates memory bandwidth bottlenecks and achieves peak Tensor Core utilization. - Hardware-Accelerated MTP Decoding (--spec-type draft-mtp): Reaches 45-60+ tokens/s on a single RTX 3090 (24GB) via… Open weights apache-2.0 [View model](https://savrn.com/models/atx-swift-1-5-qwen3-8-27b-uncensored-mtp-gguf) Model · Text generation ### [Qwen3-4B-RKLLM](https://savrn.com/models/qwen3-4b-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from Qwen/Qwen3-4B. 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 template when… Open weights apache-2.0 rkllm [View model](https://savrn.com/models/qwen3-4b-rkllm) Model ### [MMFineReason-2B-GGUF](https://savrn.com/models/mmfinereason-2b-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/OpenDataArena/MMFineReason-2B For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/MMFineReason-2B-i1-GGUF 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 And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9 See… Open weights apache-2.0 transformers [View model](https://savrn.com/models/mmfinereason-2b-gguf) Model · Image segmentation ### [car-damage-photo-integrity-v9](https://savrn.com/models/car-damage-photo-integrity-v9) [Sai Charan Obuladinne](https://savrn.com/model-publishers/obuladinnesai) A PyTorch research model for detecting and localizing AI-edited regions in car-damage photos. It focuses on localized inpainting that adds, removes or exaggerates vehicle damage, and returns a tamper heatmap, an image-level suspicion score and a threshold-based flag. Use it to explore image manipulation detection, diffusion inpainting detection and vehicle-photo integrity with the supplied local inference code. It does not assess damage severity, estimate repair costs or prove fraud. Fully AI-generated scenes are a documented weak case. Author: Sai Charan Obuladinne · Framework: PyTorch · Weights: Safetensors · Declared repository license: Apache-2.0; see the upstream-data caveat below.… Open weights apache-2.0 29M parameters pytorch [View model](https://savrn.com/models/car-damage-photo-integrity-v9) Model ### [Decision-1.0-Nox-4B](https://savrn.com/models/decision-1-0-nox-4b) [vLLM Semantic Router](https://savrn.com/model-publishers/llm-semantic-router) Nox, Latin for night. Give Nox a state, questions and possible answers. It returns typed decisions and probabilities, with labels defined at runtime. 73.09% weighted accuracy across 3,766 decisions and 54 tasks: +2.99 points over Kev-4B on the same benchmark. Accuracy (%). Overall weights: Decisions 30%, Composition 25%, Reading 15%, Inference 15%, Transfer 15%. These outcome-informed product-priority weights were chosen after observing results; reweighting is not a training improvement. Bold marks Decision-family cells above every external open or untuned reference for that metric, excluding Jev and the other Decision models. Distinct Choice questions at a fixed 499 input tokens per… Open weights apache-2.0 [View model](https://savrn.com/models/decision-1-0-nox-4b) Model · Zero-shot classification ### [chinese-decision-0.8b](https://savrn.com/models/chinese-decision-0-8b) [Quincyqiang](https://savrn.com/model-publishers/quincyqiang) 各尺寸、部署格式与外部模型对比。2B BF16 与 2B Q80 均已公开发布并完成 CJ-Bench 全量评测：2B 整体准确率更高，0.8B 更小更快，部分法律判断与金融评分任务仍是 0.8B 更好。 Chinese Decision 是一个面向中文场景的轻量级决策模型。它不聊天、不生成文本，只做一件事：给定上下文和一组固定候选项，返回每个候选项的概率。 - 一次前向，零输出 token：读取最后位置上候选字母 A–Z 的 logits，softmax 后即为概率。 - 三种题型：noul（是/否）、choice（单选）、score（有序等级，返回期望分值）。 以下截图来自我们的测试平台 Decision Studio，后端运行的就是本模型（step 10000）。这些场景采用同一种设计：业务规则写在代码里，决策模型只负责给出概率。 每张工单在一次请求里同时问 6 个问题：类别（choice），严重程度和情绪（score），是否可复现、是否要求退款、是否有流失风险（noul）。不管后续用不用得上都一并提问，省去多次往返。路由规则写在代码里，根据这些概率决定工单派给哪个队列；类别置信度低于 0.5 时转人工复核。截图中 8 张工单 × 6 个问题共 48 次决策，用时 285 ms。 推荐同时安装 flash-linear-attention，它为 Qwen3.5 的线性注意力层提供高效 kernel： Hugging Face： ModelScope： decisionconfig.json 中的温度只在 softmax 前缩放… Open weights apache-2.0 752M parameters 262,144 tokens transformers [View model](https://savrn.com/models/chinese-decision-0-8b) Model · Text generation ### [Qwen2-0.5B-Instruct-RKLLM](https://savrn.com/models/qwen2-0-5b-instruct-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from Qwen/Qwen2-0.5B-Instruct. 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/qwen2-0-5b-instruct-rkllm) Model · Feature extraction ### [univ_ecapa_tdnn_small_ge2e](https://savrn.com/models/univ-ecapa-tdnn-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-ecapa-tdnn-small-ge2e) Model · Text generation ### [MiniCPM3-4B-RKLLM](https://savrn.com/models/minicpm3-4b-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from openbmb/MiniCPM3-4B. 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/minicpm3-4b-rkllm) Model · Image and text to text ### [qwen3-vl-8b-hebrew-v21-ckpt](https://savrn.com/models/qwen3-vl-8b-hebrew-v21-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: scrap, kept as evidence. Never resume from it; use v21b, the corrected re-run. Trained 2026-09-11, steps 100–700, then stopped. Warm start v20a step 1800. What was new (and carried into v21b): genizahktivv3 (60,001 rows, 3,700 pages / 1,588 manuscripts, 2.2M letters) and six direct-grounding families — locateword, readboxword, lineindex, lineofphrase, groundeddetect (box before text) and groundedcrop — lifting grounding to 20 % of the mixture across ten families. The dataset was streamed per task split. sliced the batch tensors… Open weights apache-2.0 peft [View model](https://savrn.com/models/qwen3-vl-8b-hebrew-v21-ckpt) ## 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: tiny-agent-gguf (163); Zuzu-the-cat-hindi-GGUF (163); qwen3-asr-1.7b-int8-mlx (159). ## 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)