# Text Generation Models, Page 7: Open-Weight AI Source: https://savrn.com/models/tasks/text-generation/page/7 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Models by Task # Text Generation Models 1,582 open-weight text generation models in the SAVRN Model Hub, with Convergent Intelligence, Qwen and Francesca Padovani publishing the most. 1,582Models 732Publishers 107K to 1.7TParameter range 8Licenses ## SAVRN's Take Text generation is where most of the attention goes, and where most of the money is spent. The models on this page run from tiny research checkpoints that fit on a laptop to frontier-class systems that need a full server of accelerators. The first question we ask about any of them is not how smart it is, but what it costs to keep it answering around the clock. That cost comes down to three things: the size of the weights, the precision you run them at, and how much context each request carries. Moving from 16-bit to 4-bit cuts the memory a model needs by roughly four times, which is often the difference between one server and several. Mixture-of-experts models, such as the newer DeepSeek, Qwen and Nemotron releases, hold a large total parameter count but activate only a small share per token, so they answer faster than their size suggests while still needing the memory to hold every expert. Licensing decides the rest. Apache 2.0 and MIT models can go straight into a commercial product. Llama, Gemma's earlier generations and models marked other each carry terms you need to read first, and a model with no stated license is research material until the publisher says otherwise. Our advice is to shortlist by license, size to the hardware you can actually run, then test the survivors on your own prompts. The table below is sorted by what people download, not by what we recommend. SAVRN Research, 2026-09-18 ## Most Downloaded | Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit | | --- | --- | --- | --- | --- | --- | | [Qwen3-0.6B](https://savrn.com/models/qwen3-0-6b) | Qwen | 752M | apache-2.0 | 25.2M | 1x MI300X, $1.85/hr | | [gpt2](https://savrn.com/models/gpt2) | OpenAI community | 137M | mit | 18.8M | 1x MI300X, $1.85/hr | | [Qwen3-4B](https://savrn.com/models/qwen3-4b) | Qwen | 4B | apache-2.0 | 11.4M | 1x MI300X, $1.85/hr | | [Qwen3-14B](https://savrn.com/models/qwen3-14b) | Qwen | 14.8B | apache-2.0 | 9.4M | 1x MI300X, $1.85/hr | | [Qwen3-8B](https://savrn.com/models/qwen3-8b) | Qwen | 8.2B | apache-2.0 | 8.1M | 1x MI300X, $1.85/hr | | [Qwen2.5-7B-Instruct](https://savrn.com/models/qwen2-5-7b-instruct) | Qwen | 7.6B | apache-2.0 | 6.4M | 1x MI300X, $1.85/hr | | [gpt-oss-20b](https://savrn.com/models/gpt-oss-20b) | OpenAI | 20.9B | apache-2.0 | 5.7M | 1x MI300X, $1.85/hr | | [opt-125m](https://savrn.com/models/opt-125m) | AI at Meta | — | other | 5.6M | — | | [DeepSeek-V3.2](https://savrn.com/models/deepseek-v3-2) | DeepSeek | 685.4B | mit | 5.4M | 7x MI325X, $14.00/hr | | [Qwen2.5-0.5B-Instruct](https://savrn.com/models/qwen2-5-0-5b-instruct) | Qwen | 494M | apache-2.0 | 5.3M | 1x MI300X, $1.85/hr | ## Licenses | License | Models | Commercial use | | --- | --- | --- | | [apache-2.0](https://savrn.com/models/licenses/apache-2-0) | 734 | Yes | | not stated | 345 | Not stated | | other | 175 | Read the license | | [mit](https://savrn.com/models/licenses/mit) | 152 | Yes | | [cc-by-nc-4.0](https://savrn.com/models/licenses/cc-by-nc-4-0) | 42 | Not without separate permission | | [llama3.2](https://savrn.com/models/licenses/llama3-2) | 33 | Read the license | | [gemma](https://savrn.com/models/licenses/gemma) | 22 | Yes, with conditions | | [llama3.1](https://savrn.com/models/licenses/llama3-1) | 21 | Yes, with conditions | ## Who Publishes Them | Publisher | Models | | --- | --- | | [Convergent Intelligence](https://savrn.com/model-publishers/reaperdoesntknow) | 54 | | [Qwen](https://savrn.com/model-publishers/qwen) | 42 | | [Francesca Padovani](https://savrn.com/model-publishers/francesca9805) | 37 | | [Yanggangu](https://savrn.com/model-publishers/yanggangu) | 28 | | [Shreyansh singh](https://savrn.com/model-publishers/shreyansh12183) | 25 | | [Kumar](https://savrn.com/model-publishers/sandeep123) | 23 | ## All 1,582 Models, Page 7 of 27 Model · Text generation ### [EdgeRock-4B-GGUF](https://savrn.com/models/edgerock-4b-gguf) [Aaron Xu](https://savrn.com/model-publishers/hkust-aaron) BF16 and seven quantized GGUF profiles, from Q3KM through Q80, with matching Ollama Modelfiles and an EdgeRock RMA backend. Fixed memory for long-running generation. This EdgeRock-4B deployment package retains recurrent state, local attention, and Rock Memory Anchors (RMA), keeping persistent history at a fixed capacity as context grows. 固定记忆，长程生成。 本版本保留 EdgeRock-4B 的递推状态、局部注意力和记忆岩锚（RMA），历史内存不随上下文增长。 Rock Matrix Whitepaper / 岩芯矩阵技术白皮书： whitepaper connects EdgeRock's fixed memory, DiffusionRock's parallel generation, and SiliconRock's model-native silicon into the route toward Ultra-Fast Token. 白皮书讲清固定记忆、并行生成、模型固化三条路线，以及它们如何走向更快、更经济的生成。 The instructions and validation below apply to this… Open weights other [View model](https://savrn.com/models/edgerock-4b-gguf) Model · Text generation ### [Shreyansh-STEM-AI-2B-v3](https://savrn.com/models/shreyansh-stem-ai-2b-v3) [Shreyansh singh](https://savrn.com/model-publishers/shreyansh12183) Shreyansh-STEM-AI-2B-v3 is a sovereign compact foundation model for scientific, physical, and mathematical derivation, engineered through SOLAR-style Depth Up-Scaling (DUS) and Continual Pre-Training (CPT) Seam Healing. To surpass standard 2B parameter capacity without requiring training from scratch, intermediate transformer layers were duplicated and spliced, expanding the model depth to 22 layers with a hidden dimension of $d=2048$. Splicing transformer blocks introduces interface discontinuity along the residual stream. To heal these seams, the model underwent Continual Pre-Training (CPT) over the multi-gigabyte shreyansh-1B-SLM-pretrain-stem-english corpus (over 2,400 parquet shards).… Open weights cc-by-nc-4.0 1.9B parameters 4,096 tokens [View model](https://savrn.com/models/shreyansh-stem-ai-2b-v3) Model · Text generation ### [Qwen3.8-Flash-Next-125B-A5B-INT4-AutoRound](https://savrn.com/models/qwen3-8-flash-next-125b-a5b-int4-autoround) [Aldo Zampatti](https://savrn.com/model-publishers/azampatti) Qwen3.8-Flash-Next with 5 routed experts per token instead of 10, healed so it stays close to the original, quantized to int4. It runs on one DGX Spark (GB10, 128 GB) at roughly 64-70 tokens/s. 125B parameters in total, 4.8B active per token. The original activates 6B. Everything needed to serve it is in this one repository, including the 49 GB FP8 n-gram table under ple-table/. Nothing else to download. That builds the serving image, downloads this repository, and starts an OpenAI-compatible server on port 8000. The scripts and the full explanation are in that repo. Serving by hand needs Saren-Arterius/qwen3.8-Flash-DGX-AutoRound, because a stock vLLM cannot serve this checkpoint's int4 +… Open weights other 124B parameters 262,144 tokens vllm [View model](https://savrn.com/models/qwen3-8-flash-next-125b-a5b-int4-autoround) Model · Text generation ### [GLM-5.3-Uncensored-GGUF](https://savrn.com/models/glm-5-3-uncensored-gguf) [Ash DC](https://savrn.com/model-publishers/softwareweaver) Q4KM GGUF converted from dealignai/GLM-5.3-UNCENSORED-FP8. The source is a 753B-parameter mixture-of-experts model. See File folder for different Quants See the https://huggingface.co/zai-org/GLM-5.3 for usage. Thanks Z.ai for open sourcing this great model. Credit for the modified source weights belongs to dealignai. This repository ONLY provides the GGUF conversion. Open weights mit gguf [View model](https://savrn.com/models/glm-5-3-uncensored-gguf) Model · Text generation ### [ViuAI-500M](https://savrn.com/models/viuai-500m) [Vipin Kumar](https://savrn.com/model-publishers/viuai) ViuAI Sarus-500M is a 495-million parameter autoregressive Small Language Model (SLM) designed, trained from scratch, and aligned by ViuAI Research. Engineered for ultra-low latency edge devices and lightweight CPU/GPU deployment, Sarus-500M balances deep knowledge representation with minimal resource requirements (~944 MB model footprint). The official recommended production checkpoint is: You can run Sarus-500M locally in a few lines of code without requiring the heavy transformers package: Because Sarus-500M is a 500M SLM, setting proper sampling parameters prevents logit tail collisions: The dpo5kfinal.pt checkpoint was evaluated on the standard multi-domain benchmark: Sarus-500M is… Open weights apache-2.0 2,048 tokens [View model](https://savrn.com/models/viuai-500m) Model · Text generation ### [Qwen3.8-27B-EfficientThink-Uncensored-VAL-APEX-I-MiniPlus-V2.1-GGUF](https://savrn.com/models/qwen3-8-27b-efficientthink-uncensored-val-apex-i-miniplus-v2-1-gguf) [IsValorum](https://savrn.com/model-publishers/isvalorum) Official VAL-APEX-I quantization of nerkyor/Qwen3.8-27B-EfficientThink-Uncensored-K3-Opus5-Grok4.6-GPT5.6Sol-SFT-SimPO-DFlash2. Vector-calibrated Asymmetric Layer-wise Outlier-preserving Recurrent-aware Unified Matrix-quantization The NanoPlus V3 file is 11.35 GB, approximately 0.45 GB smaller than GSQ-RCO IQ3S, with 11.49% lower reported WikiText-2 PPL (6.2577 vs 7.07). MiniPlus V3 reaches 13.21% lower reported PPL (6.1361 vs 7.07) at 14.16 GB, approximately 2.36 GB larger. × is the number of tensors in each format. Counts were read from the released GGUF headers; the GSQ-RCO column uses its published tensor allocation. This release contains 851 tensors. The main GGUF has no blk.64.… Open weights apache-2.0 gguf [View model](https://savrn.com/models/qwen3-8-27b-efficientthink-uncensored-val-apex-i-miniplus-v2-1-gguf) Model · Text generation ### [Q-TensorFormer](https://savrn.com/models/q-tensorformer) [V C Premchand Yadav](https://savrn.com/model-publishers/premchan369) Q-TensorFormer is an open-source, publication-grade research architecture and inference testbed designed by Premchand Yadav. It investigates a fundamental question in efficient deep learning: Can a Transformer dynamically allocate its computational capacity, tensor rank, attention complexity, and KV-cache precision according to the marginal informational value of each token under explicit hardware constraints? Q-TensorFormer is an experimental PyTorch and Hugging Face research architecture that investigates closed-loop, information-value-driven resource allocation for Transformer inference under explicit memory, latency, bandwidth, and quality constraints. Rather than claiming that dynamic… Open weights apache-2.0 4,096 tokens transformers [View model](https://savrn.com/models/q-tensorformer) Model · Text generation ### [Qwen3.8-Flash-Next-cmf](https://savrn.com/models/qwen3-8-flash-next-cmf) [Oleg Kirichenko](https://savrn.com/model-publishers/infosave) This repository contains the text tower of converted directly from the original bf16 checkpoint to one memory-mapped CMF file. It runs with the same file on CPU, Vulkan (NVIDIA/AMD/Intel), DX12 and Metal. No Python, PyTorch, CUDA toolkit or model-specific GPU build is needed at inference time. The upstream language model has 125B parameters with about 6B active per token, plus the 51B-parameter PLE n-gram table. The sparse/on-demand layout is why a single artifact can remain usable well below its full file size in VRAM. The file requires the dedicated qwen4exp runtime added alongside this conversion; older packaged binaries reject the new architecture instead of silently running an… Open weights other cortiq [View model](https://savrn.com/models/qwen3-8-flash-next-cmf) Model · Text generation ### [Qwen3.8-27B-EfficientThink-Uncensored-VAL-APEX-I-NanoPlus-GGUF](https://savrn.com/models/qwen3-8-27b-efficientthink-uncensored-val-apex-i-nanoplus-gguf) [IsValorum](https://savrn.com/model-publishers/isvalorum) Official VAL-APEX-I quantization of nerkyor/Qwen3.8-27B-EfficientThink-Uncensored-K3-Opus5-Grok4.6-GPT5.6Sol-SFT-SimPO-DFlash2. Vector-calibrated Asymmetric Layer-wise Outlier-preserving Recurrent-aware Unified Matrix-quantization The NanoPlus V3 file is 11.35 GB, approximately 0.45 GB smaller than GSQ-RCO IQ3S, with 11.49% lower reported WikiText-2 PPL (6.2577 vs 7.07). MiniPlus V3 reaches 13.21% lower reported PPL (6.1361 vs 7.07) at 14.16 GB, approximately 2.36 GB larger. × is the number of tensors in each format. Counts were read from the released GGUF headers; the GSQ-RCO column uses its published tensor allocation. This release contains 851 tensors. The main GGUF has no blk.64.… Open weights apache-2.0 gguf [View model](https://savrn.com/models/qwen3-8-27b-efficientthink-uncensored-val-apex-i-nanoplus-gguf) Model · Text generation ### [DeepSeek-R1-Distill-Qwen-32B-abliterated-GGUF](https://savrn.com/models/deepseek-r1-distill-qwen-32b-abliterated-gguf) [Alan Frederick Benitez](https://savrn.com/model-publishers/thaurock) Unlike incomplete GGUF uploads, this repository provides the full 11-quantization spectrum (from high-precision F16 down to lightweight Q2K) of the abliterated DeepSeek-R1-Distill-Qwen-32B model. Choose the exact fit for your VRAM/RAM constraints without sacrificing reasoning capabilities. This is the integral and complete collection of quantizations in GGUF format for the DeepSeek-R1-Distill-Qwen-32B-abliterated model, prepared locally for use with llama.cpp, Ollama, LM Studio, or Text-Generation-WebUI. This model combines advanced reasoning distillation (deep logical thinking capability via tags) based on Alibaba's powerful Qwen-32B model, but processed with abliteration techniques to… Open weights apache-2.0 [View model](https://savrn.com/models/deepseek-r1-distill-qwen-32b-abliterated-gguf) Model · Text generation ### [LuminoLex-1.7B](https://savrn.com/models/luminolex-1-7b) [LuminoLex AI](https://savrn.com/model-publishers/luminolex) LuminoLex-1.7B is Bangladesh’s first decoder-only language model pretrained from scratch. Built on a sparse Mixture-of-Experts (MoE) architecture, it contains 1.7 billion total parameters while activating only ~410 million parameters per token. By combining an always-on shared expert with Grouped-Query Attention (GQA), LuminoLex-1.7B delivers fast, memory-efficient inference without sacrificing model capacity. This model is released under the Apache License 2.0. Open weights apache-2.0 1.7B parameters 1,024 tokens transformers [View model](https://savrn.com/models/luminolex-1-7b) Model · Text generation ### [Collision-1B](https://savrn.com/models/collision-1b) [viraj R](https://savrn.com/model-publishers/collision-10m) COLLISION-1B is the official primary flagship model of the COLLISION ecosystem. Packing 999,376,128 parameters (~1.00B) into an optimized 24-layer transformer architecture, it delivers rich contextual reasoning, full 1,024-token context capacity, and state-of-the-art hybrid NLP capabilities with grounded web and local retrieval. Run everything in your browser on free Google Colab in under 10 seconds: Clone this repository and run pure PyTorch inference directly: COLLISION features a complete, zero-latency NLP pipeline: COLLISION includes a full dual-process cognitive architecture featuring non-linear Graph-of-Thoughts (GoT) and Hegelian Dialectics: The flagship features an advanced… Open weights mit transformers [View model](https://savrn.com/models/collision-1b) Model · Text generation ### [semsimula-ladder-live-owt-d384-l2-attention-potential-rfqk](https://savrn.com/models/semsimula-ladder-live-owt-d384-l2-attention-potential-rfqk) [Dimitar Gueorguiev](https://savrn.com/model-publishers/dimitarpg13) Settled is the mean of the last three evals (52.83, 53.05, 52.81); best 50.97 at step 31,000, final 52.81. Consecutive evals near the end differ by about ±2 PPL. - The hardening recovers what the unhardened field lost: 2.4% better than the same model without it. - The hardened field still adds nothing measurable: 0.4% better than the model without the field, which is within eval noise. The three L=2 models with the register path settle within 2.5% of one another, whatever else is switched on. - Without the registers the same field is worth much more: the conservative-only model with this field and no register read settles at 54.27, 6.0% below the conservative-only model. The field and the… Open weights cc-by-4.0 pytorch [View model](https://savrn.com/models/semsimula-ladder-live-owt-d384-l2-attention-potential-rfqk) Model · Text generation ### [Llama-3.2-1B-dolma3_mix](https://savrn.com/models/llama-3-2-1b-dolma3-mix) [CHEN EN LO](https://savrn.com/model-publishers/asciibase64) This revision publishes the final raw checkpoint of raw6b-normuon-adam2x-20261018: 6,000,082,944 non-padding input tokens, 28,619 optimizer updates. The model was randomly initialized (residual-half, seed42); no pretrained weights or previous training checkpoint initialized this run. Model architecture and The full reserved 50,000-document set has not been evaluated for this checkpoint. PPL is exp(sum held-out NLL / sum valid prediction tokens), not a mean of subset PPLs. Documents are evaluated independently, with stride8191 for long documents. The entire original document split was shuffled with seed42 before tokenization; the last50,000 documents were held out and excluded from all… Open weights llama3.2 1.2B parameters 131,072 tokens transformers [View model](https://savrn.com/models/llama-3-2-1b-dolma3-mix) Model · Text generation ### [PMA-1.2](https://savrn.com/models/pma-1-2) [PatriotMemory-AI](https://savrn.com/model-publishers/patriotmemory-ai) PMA-1.2 is Patriot Memory's 127.9M-parameter on-device language model. It speaks English and Traditional Chinese, answers as PMA from Patriot Memory, and fits in a 128 MB parameter budget built for edge hardware. It is a new architecture, a new tokenizer, and an order of magnitude more training, aimed at the same job: a small, fast, honest assistant for Patriot Memory and Viper Gaming questions and general chat. It provides accurate information regarding: If you run into multi-GPU tensor device mismatch errors: Run the script with CUDAVISIBLEDEVICES=0 to isolate execution to GPU 0. trustremotecode=True is required: PMA-1.2's architecture is our own and ships as small Python files next to… Open weights apache-2.0 126M parameters 32,768 tokens transformers [View model](https://savrn.com/models/pma-1-2) Model · Text generation ### [Huihui-Qwen3.8-27B-Abliterated-VAL-APEX-I-MiniPlus-V2.1-GGUF](https://savrn.com/models/huihui-qwen3-8-27b-abliterated-val-apex-i-miniplus-v2-1-gguf) [IsValorum](https://savrn.com/model-publishers/isvalorum) Official VAL-APEX-I quantization of huihui-ai/Huihui-Qwen3.8-27B-abliterated. Vector-calibrated Asymmetric Layer-wise Outlier-preserving Recurrent-aware Unified Matrix-quantization The NanoPlus V3 file is 11.35 GB, approximately 0.45 GB smaller than GSQ-RCO IQ3S, with 11.49% lower reported WikiText-2 PPL (6.2577 vs 7.07). MiniPlus V3 reaches 13.21% lower reported PPL (6.1361 vs 7.07) at 14.16 GB, approximately 2.36 GB larger. × is the number of tensors in each format. Counts were read from the released GGUF headers; the GSQ-RCO column uses its published tensor allocation. This release contains 866 tensors. The main GGUF includes the additional 15-tensor MTP head. The model supports a native… Open weights apache-2.0 gguf [View model](https://savrn.com/models/huihui-qwen3-8-27b-abliterated-val-apex-i-miniplus-v2-1-gguf) Model · Text generation ### [SRA-RiskGate-4B-GGUF](https://savrn.com/models/sra-riskgate-4b-gguf) [Sriram Ramakrishnan](https://savrn.com/model-publishers/sriram1983007) For dispute adjudication, use the full-precision model until the higher-precision GGUF builds are here. Apache 2.0 Open weights apache-2.0 gguf [View model](https://savrn.com/models/sra-riskgate-4b-gguf) Model · Text generation ### [Huihui-Qwen3.8-27B-Abliterated-VAL-APEX-I-NanoPlus-GGUF](https://savrn.com/models/huihui-qwen3-8-27b-abliterated-val-apex-i-nanoplus-gguf) [IsValorum](https://savrn.com/model-publishers/isvalorum) Official VAL-APEX-I quantization of huihui-ai/Huihui-Qwen3.8-27B-abliterated. Vector-calibrated Asymmetric Layer-wise Outlier-preserving Recurrent-aware Unified Matrix-quantization The NanoPlus V3 file is 11.35 GB, approximately 0.45 GB smaller than GSQ-RCO IQ3S, with 11.49% lower reported WikiText-2 PPL (6.2577 vs 7.07). MiniPlus V3 reaches 13.21% lower reported PPL (6.1361 vs 7.07) at 14.16 GB, approximately 2.36 GB larger. × is the number of tensors in each format. Counts were read from the released GGUF headers; the GSQ-RCO column uses its published tensor allocation. This release contains 866 tensors. The main GGUF includes the additional 15-tensor MTP head. The model supports a native… Open weights apache-2.0 gguf [View model](https://savrn.com/models/huihui-qwen3-8-27b-abliterated-val-apex-i-nanoplus-gguf) Model · Text generation ### [Qwen2.5-3B-Instruct-RKLLM](https://savrn.com/models/qwen2-5-3b-instruct-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from Qwen/Qwen2.5-3B-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 w4a16 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/qwen2-5-3b-instruct-rkllm) Model · Text generation ### [RED-SNOW-5.3-FLASH-GGUF](https://savrn.com/models/red-snow-5-3-flash-gguf) [Jackson](https://savrn.com/model-publishers/freakyskittle) Full-functionality GGUF exports for llama.cpp: text + vision included (the GLM-5.3 vision tower ships as a separate mmproj file, keeping every quant pairable with it). Every file was smoke-tested end to end before upload - text runs with llama-cli, vision with llama-server --mmproj. RED-SNOW 5.3 FLASH is built for authorized offensive-security work: attack-path reasoning, cross-domain operations (identity, cloud, network, app, endpoint, OT), exploit and tradecraft analysis, detection-aware execution, and purple-team handoff. See the base model card for the full picture. Architecture highlights that matter for inference: - MLA (nope-only) + DSA sparse-attention indexer with k-pool… Open weights mit gguf [View model](https://savrn.com/models/red-snow-5-3-flash-gguf) Model · Text generation ### [EliteReplay-BIRD-4B](https://savrn.com/models/elitereplay-bird-4b) [Hanane Nour Moussa](https://savrn.com/model-publishers/hananour) This is a Qwen3.5-4B-Base model fine-tuned on BIRD using EliteReplay. EliteReplay combines reinforcement learning with supervised replay from an archive of verified, diverse SQL solutions to preserve solution diversity during training. We train on 5,000 BIRD questions. Archive search uses 1,000 of those questions to select 1,500 verified SQL examples for replay. The model uses GRPO with supervised archive replay at weight μ = 0.05. Hyperparameters and implementation details are available in the training instructions, original training configuration, and checkpoint metadata. Training used 4 NVIDIA A100 80 GB GPUs, with two for policy optimization and two for rollout generation, using veRL… Open weights apache-2.0 4.5B parameters 262,144 tokens transformers [View model](https://savrn.com/models/elitereplay-bird-4b) Model · Text generation ### [T-Search-APEX-I-MiniPlus-V2.1-GGUF](https://savrn.com/models/t-search-apex-i-miniplus-v2-1-gguf) [IsValorum](https://savrn.com/model-publishers/isvalorum) Welcome to APEX-I-MiniPlus-V2.1 for t-tech/T-Search (the specialized 35B Mixture-of-Experts architecture developed by t-tech as an agentic retriever that plans, executes multi-round searches, and synthesizes verifiable evidence chains). Standard automated community quantizations uniformly degrade sensitive routing matrices and expert feed-forwards down to 2-bit codebooks, corrupting retrieval planning trajectories, hallucinating query filters, and dropping critical visual tokens. APEX-I-MiniPlus-V2.1 was engineered differently. This is a 100% custom, hand-crafted tensor-by-tensor quantization built with mathematical precision overrides, calibrated importance matrices (imatrix), and an… Open weights apache-2.0 gguf [View model](https://savrn.com/models/t-search-apex-i-miniplus-v2-1-gguf) Model · Text generation ### [Qwen3.8-35B-A3B-Distill-GGUF](https://savrn.com/models/qwen3-8-35b-a3b-distill-gguf) [Nguyen Quoc Vinh](https://savrn.com/model-publishers/ngquocvinh) Community GGUF quantizations of empero-ai/Qwen3.8-35B-A3B-Distill. If this GGUF made your day easier, a coffee would make mine. I build and test these releases myself. Your coffee helps keep me going. Thank you for supporting this work. This release is based on the upstream Qwen3.8-35B-A3B-Distill model card. The model is a text-generation MoE model with 35B total parameters and approximately 3B active parameters per token, using 256 experts with 8 routed experts. It uses hybrid attention components and has a native context length of 262,144 tokens in the upstream configuration. The GGUF files are intended for local inference with llama.cpp and compatible runtimes. The upstream model card… Open weights apache-2.0 llama.cpp [View model](https://savrn.com/models/qwen3-8-35b-a3b-distill-gguf) Model · Text generation ### [Janus-35B-HERETIC](https://savrn.com/models/janus-35b-heretic) [Fool](https://savrn.com/model-publishers/fooldev) Architecture: Qwen 3.6 35B-A3B (MoE) | Total Params: ~34.7B | Active Params: ~3B | Context: 262,144 native / 1,010,000 extensible | Base: Heretic (llmfan46) | Teacher: Claude Fable 5 | Type: Distilled + Abliterated MoE LLM A personal fork of llmfan46/Qwen3.6-35B-A3B-uncensored-heretic — an uncensored Heretic-style abliteration of Qwen/Qwen3.6-35B-A3B, the 35B-total / 3B-active mixture-of-experts multimodal base — repackaged as Janus-35B with Claude Fable 5 reasoning data in the teacher slot. Refusal-trained behavior is dialed back at the base layer. One-liner via Hugging Face (pulls a GGUF + this repo's root-level template / system / params files, including the tool-calling template — HF's… Open weights apache-2.0 gguf [View model](https://savrn.com/models/janus-35b-heretic) Model · Text generation ### [XiYanSQL-QwenCoder-32B-2504](https://savrn.com/models/xiyansql-qwencoder-32b-2504) [XGenerationLab](https://savrn.com/model-publishers/xgenerationlab) We are excited to release the XiYanSQL-QwenCoder-2504 version, our latest SQL generation model. This version continues to optimize upon the previous version, delivering enhanced performance. - Our model incorporates important explorations combining fine-tuning and GRPO training, leveraging the post-training strategies of GRPO without a thinking process, achieving both efficiency and accuracy in SQL generation. - It demonstrates impressive performance and supports multiple dialects, ready to use out of the box. - Improved generalization capabilities, excelling on different dialects and out-of-domain datasets. In this evaluation, we have also added a real-world SQL benchmark (the DW test… Open weights apache-2.0 32.8B parameters 32,768 tokens [View model](https://savrn.com/models/xiyansql-qwencoder-32b-2504) Model · Text generation ### [Qwen2.5-1.5B-Instruct-RKLLM](https://savrn.com/models/qwen2-5-1-5b-instruct-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from Qwen/Qwen2.5-1.5B-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 w4a16 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/qwen2-5-1-5b-instruct-rkllm) Model · Text generation ### [Qwen3.6-Qworus-35B-A3B-V1-APEX-GGUF](https://savrn.com/models/qwen3-6-qworus-35b-a3b-v1-apex-gguf) [Charles Muench](https://savrn.com/model-publishers/darkkitsune) What is this? This is a DARE-TIES merge of the following finetuned & post-trained models: - EryriLabs/Ornith-1.5-35B-A3B-BigBang-MTP-GGUF: The base model, for terminal use, coding, software architecture, planning. Also provides MTP support. - Kwaipilot/KAT-Coder-V2.5-Dev: Contributes efficient coding with fewer tokens over more languages. - Accio-Lab/occamy-1.0: Contributes general agent abilities, also a semi-official finetune. - AllSpark-Research/CompoWorld: Also contributes general agent abilities, different strengths/weaknesses from occamy. The result is an efficient hybrid model for coding, terminal agent work, planning, designing, and conversation. V1.1: Revised recipe which brings in… Open weights transformers [View model](https://savrn.com/models/qwen3-6-qworus-35b-a3b-v1-apex-gguf) Model · Text generation ### [EdgeRock-4B-MLX-6bit-g64](https://savrn.com/models/edgerock-4b-mlx-6bit-g64) [Aaron Xu](https://savrn.com/model-publishers/hkust-aaron) MLX 6-bit weights with group size 64; a 3.732 GB weight package. Fixed memory for long-running generation. This EdgeRock-4B deployment package retains recurrent state, local attention, and Rock Memory Anchors (RMA), keeping persistent history at a fixed capacity as context grows. 固定记忆，长程生成。 本版本保留 EdgeRock-4B 的递推状态、局部注意力和记忆岩锚（RMA），历史内存不随上下文增长。 Rock Matrix Whitepaper / 岩芯矩阵技术白皮书： whitepaper connects EdgeRock's fixed memory, DiffusionRock's parallel generation, and SiliconRock's model-native silicon into the route toward Ultra-Fast Token. 白皮书讲清固定记忆、并行生成、模型固化三条路线，以及它们如何走向更快、更经济的生成。 The instructions and validation below apply to this package. Browse all EdgeRock models. Original-model benchmarks… Open weights other 4.6B parameters 262,144 tokens mlx [View model](https://savrn.com/models/edgerock-4b-mlx-6bit-g64) Model · Text generation ### [Qwen3.8-Flash-Next-Q4_K_XL-DN4](https://savrn.com/models/qwen3-8-flash-next-q4-k-xl-dn4) [Filipe Gatti](https://savrn.com/model-publishers/bitlamas) This build is what works best on my 128 GB LPDDR5X AMD Strix Halo (Ryzen AI MAX+ 395, Radeon 8060S) under Windows 11, with llama.cpp on Vulkan. It is my daily driver for agentic work. It is unsloth's UD-Q4KXL with one change: the 48 expert down projections are IQ4NL instead of Q51, quantized from Qwen's original BF16 weights with unsloth's importance matrix. That takes the file from 111.3 GB to 102.5 GB. This repo used to hold DN4, the first version of this file, published 2026-09-05. DN4 made the same 48 tensors by requantizing unsloth's Q51 copies, and DN4B makes them from BF16. Same size, same speed, same memory, and measurably closer to the original model. DN4 is still downloadable from… Open weights other llama.cpp [View model](https://savrn.com/models/qwen3-8-flash-next-q4-k-xl-dn4) Model · Text generation ### [Khushi-Business-AI-4bit](https://savrn.com/models/khushi-business-ai-4bit) [Kartik Sharma](https://savrn.com/model-publishers/ux4567) A 4-bit quantized, business-focused conversational AI built on Mistral 7B using Unsloth. Optimized for low VRAM (runs on 6GB+ VRAM) and for Indian business use-cases. - 7B params in 4-bit (BF16 base): ~4GB RAM me chal jayega from unsloth import FastLanguageModel model, tokenizer = FastLanguageModel.frompretrained( modelname = "UX4567/Khushi-Business-AI-4bit", maxseqlength = 2048, loadin4bit = True, prompt = "Ek customer bol raha hai: Mujhe refund chahiye, kya reply du?" inputs = tokenizer([prompt], returntensors = "pt").to("cuda") outputs = model.generate(inputs, maxnewtokens = 200) print(tokenizer.batchdecode(outputs)[0]) Open weights apache-2.0 7.2B parameters 32,768 tokens [View model](https://savrn.com/models/khushi-business-ai-4bit) Model · Text generation ### [EdgeRock-4B-ONNX-Q4-CPU-g32](https://savrn.com/models/edgerock-4b-onnx-q4-cpu-g32) [Aaron Xu](https://savrn.com/model-publishers/hkust-aaron) ONNX Runtime CPU: 4-bit weight-only quantization, block size 32, and FP32 activations; 3.362 GB of graphs and weights. Fixed memory for long-running generation. This EdgeRock-4B deployment package retains recurrent state, local attention, and Rock Memory Anchors (RMA), keeping persistent history at a fixed capacity as context grows. 固定记忆，长程生成。 本版本保留 EdgeRock-4B 的递推状态、局部注意力和记忆岩锚（RMA），历史内存不随上下文增长。 Rock Matrix Whitepaper / 岩芯矩阵技术白皮书： whitepaper connects EdgeRock's fixed memory, DiffusionRock's parallel generation, and SiliconRock's model-native silicon into the route toward Ultra-Fast Token. 白皮书讲清固定记忆、并行生成、模型固化三条路线，以及它们如何走向更快、更经济的生成。 The instructions and validation below apply to this package.… Open weights other 262,144 tokens [View model](https://savrn.com/models/edgerock-4b-onnx-q4-cpu-g32) Model · Text generation ### [EdgeRock-4B-ONNX-Q4-g32](https://savrn.com/models/edgerock-4b-onnx-q4-g32) [Aaron Xu](https://savrn.com/model-publishers/hkust-aaron) ONNX Runtime CUDA: 4-bit weight-only quantization, block size 32, and FP16 activations; 3.025 GB of graphs and weights. Fixed memory for long-running generation. This EdgeRock-4B deployment package retains recurrent state, local attention, and Rock Memory Anchors (RMA), keeping persistent history at a fixed capacity as context grows. 固定记忆，长程生成。 本版本保留 EdgeRock-4B 的递推状态、局部注意力和记忆岩锚（RMA），历史内存不随上下文增长。 Rock Matrix Whitepaper / 岩芯矩阵技术白皮书： whitepaper connects EdgeRock's fixed memory, DiffusionRock's parallel generation, and SiliconRock's model-native silicon into the route toward Ultra-Fast Token. 白皮书讲清固定记忆、并行生成、模型固化三条路线，以及它们如何走向更快、更经济的生成。 The instructions and validation below apply to this… Open weights other 262,144 tokens [View model](https://savrn.com/models/edgerock-4b-onnx-q4-g32) Model · Text generation ### [EdgeRock-4B-ONNX-Q8](https://savrn.com/models/edgerock-4b-onnx-q8) [Aaron Xu](https://savrn.com/model-publishers/hkust-aaron) ONNX Runtime CUDA: 8-bit weight-only quantization, block size 64, and FP16 activations; 6.072 GB of graphs and weights. Fixed memory for long-running generation. This EdgeRock-4B deployment package retains recurrent state, local attention, and Rock Memory Anchors (RMA), keeping persistent history at a fixed capacity as context grows. 固定记忆，长程生成。 本版本保留 EdgeRock-4B 的递推状态、局部注意力和记忆岩锚（RMA），历史内存不随上下文增长。 Rock Matrix Whitepaper / 岩芯矩阵技术白皮书： whitepaper connects EdgeRock's fixed memory, DiffusionRock's parallel generation, and SiliconRock's model-native silicon into the route toward Ultra-Fast Token. 白皮书讲清固定记忆、并行生成、模型固化三条路线，以及它们如何走向更快、更经济的生成。 The instructions and validation below apply to this… Open weights other 262,144 tokens [View model](https://savrn.com/models/edgerock-4b-onnx-q8) Model · Text generation ### [logi-1.5b](https://savrn.com/models/logi-1-5b) [Alves](https://savrn.com/model-publishers/guiiwfz) This qwen2 model was trained 2x faster with Unsloth and Huggingface's TRL library. Open weights apache-2.0 1.5B parameters 32,768 tokens transformers [View model](https://savrn.com/models/logi-1-5b) Model · Text generation ### [zralddeepseekv4.1](https://savrn.com/models/zralddeepseekv4-1) [Gerald Bustilla](https://savrn.com/model-publishers/zrald) High-efficiency, hardware-benchmarked GGUF releases of DeepSeek-V4.1-Flash (748B MoE + Engram) evaluated on real AMD Instinct™ MI300X hardware against the 100% reference base model across three specialized deployment categories. 748.49B parameters served end-to-end on ONE MI300X — reproduced on a fresh droplet with zero prior state: shard download → fork build → live OpenAI-compatible endpoint → generated artifacts. All measurements from the server's own timings blocks. Generated artifacts — shipped in artifacts/ as proof-of-work: Raw run data (benchmark JSON + server log): evidence/run20261005/ These GGUFs declare general.architecture = deepseek41, which exists only in the… Open weights apache-2.0 [View model](https://savrn.com/models/zralddeepseekv4-1) Model · Text generation ### [RusMind-Vspyshka](https://savrn.com/models/rusmind-vspyshka) [DICPlay](https://savrn.com/model-publishers/dicplay) Национальный ИИ от DICPlay. Модель на устройстве для повседневных задач: пишет, объясняет, помогает с учёбой — без интернета, на мощных ПК и флагманских телефонах. Здесь собраны все сохранённые версии. Папка web/ внутри версии — та же модель, разрезанная на части для загрузки в браузер на rusmind.ru; для приложений берите целый файл. LM Studio, Ollama и другие программы с GGUF — откройте нужный файл из папки версии. RusMind обучалась на огромном массиве данных и прошла дообучение на русскоязычных наборах DICPlay: история и культура России, законы, регионы и народы страны, традиционные семейные ценности, ключевые события — на русском и английском. Распространяется по собственной лицензии… Open weights other gguf [View model](https://savrn.com/models/rusmind-vspyshka) Model · Text generation ### [stream-1-mini](https://savrn.com/models/stream-1-mini) [Whisperstream](https://savrn.com/model-publishers/whisperstream) Stream-1 Mini turns English dictation transcripts into clean Chat and Email text on your own device. It is trained to remove fillers and accidental repeats, add punctuation, and resolve spoken corrections while preserving your meaning and voice. Email mode also formats dictated greetings, paragraphs and sign-offs. This revision adds optional custom-vocabulary hints while retaining structural dictation and numbered-list support. Use these prompts and weights together. Candidates are spelling hints, not forced substitutions. Known cleanup and entity errors remain; this revision does not establish zero regressions. Pass in text from your speech recognizer. Stream-1 Mini returns the cleaned… Open weights other 752M parameters 32,768 tokens [View model](https://savrn.com/models/stream-1-mini) Model · Text generation ### [llama3-2-1b-pedagogical-GGUF](https://savrn.com/models/llama3-2-1b-pedagogical-gguf) [Team Mradermacher](https://savrn.com/model-publishers/mradermacher) static quants of https://huggingface.co/jeremierostan/llama3-2-1b-pedagogical 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 other transformers [View model](https://savrn.com/models/llama3-2-1b-pedagogical-gguf) Model · Text generation ### [Sparse-AST-BWM](https://savrn.com/models/sparse-ast-bwm) [Chaman Prakash Kanth](https://savrn.com/model-publishers/chaman1234) This repository contains one current checkpoint of the serial old/new Sparse-AST family for byte-level Blender Python completion. It was recovered from results (4).zip, the newest relevant training-results archive found in the local Downloads search, and its saved training state confirms step 93,213. It continues from step 90,213 through 3,000 successful updates, with no recorded non-finite skips. Updated 16 October 2026. The current release is under latest/ and uses SafeTensors for model weights and training-state tensors. That release contains no.pt checkpoints. Existing remote variant folders are retained as historical artifacts; they are not the latest checkpoint. The pre-update legacy… Open weights apache-2.0 pytorch [View model](https://savrn.com/models/sparse-ast-bwm) Model · Text generation ### [Prochlo-190M-Base](https://savrn.com/models/prochlo-190m-base) [CyanoAI](https://savrn.com/model-publishers/cyanoai) The smallest and first model from CyanoAI — a 190.5M-parameter dense English base LM. Named after Prochlorococcus, the tiniest and most oxygen-producing photosynthetic organism on Earth: small, dense, and outproducing everything its size. All variants share the same architecture, tokenizer, and vocabulary, so they are drop-in interchangeable. Training corpus: a curated ~826GB English mixture (web / books / code / reference), stream-packed with the tokenizer jointly shipped in this repo. This is a base model — it speaks fluent Wikipedia-style prose but has not been aligned. For chat or assistant behaviour, start from the SFT or aligned variants. Zero-shot on wikitext (JSON subset, ~500k… Open weights apache-2.0 190M parameters 4,096 tokens [View model](https://savrn.com/models/prochlo-190m-base) Model · Text generation ### [LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-GGUF](https://savrn.com/models/lfm2-5-8b-a1b-hermes-agentic-coder-abliterated-gguf) [DuoNeural](https://savrn.com/model-publishers/duoneural) DuoNeural/LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated is an apex-tier, uncensored autonomous agentic coding model trained by DuoNeural (Aura, Archon, and Jesse). Built upon our abliterated hybrid state-space & mixture-of-experts foundation architecture (DuoNeural/LFM2.5-8B-A1B-Abliterated), this model activates only 1.5 billion parameters per token out of its 8.3 billion total parameters, delivering blistering inference speeds (~380–400 tokens/sec on consumer GPUs like RTX 3090, and ~90 tokens/sec on legacy GTX 1070 laptops) while fitting in under 6 GB VRAM with Q4KM quantization. 1. Native Hermes Agentic Loop: - Explicit... deliberation before every action. - Structured... containers… Open weights other hermes [View model](https://savrn.com/models/lfm2-5-8b-a1b-hermes-agentic-coder-abliterated-gguf) Model · Text generation ### [JackOD-9B-Coder](https://savrn.com/models/jackod-9b-coder) [ManniX](https://savrn.com/model-publishers/mannix-ita) A four-way omnimergev2 merge over Qwen/Qwen3.5-9B, built for one job: multi-turn agentic coding — staying on a task across turns and tool calls until it is actually finished. It is not built to win knowledge benchmarks, and its world knowledge is limited for its size. What it is built to do is finish, and stop. --base and --task-base are both Qwen3.5-9B: every source descends from it, so task vectors are taken against the true common ancestor instead of against a sibling. merge (--skip-patterns visual.,mtp.) and passed through, so the draft head survives — an averaged MTP head stops being a usable drafter. There is no vision tower. config.json inherits a populated visionconfig from the… Open weights apache-2.0 9.7B parameters 262,144 tokens transformers [View model](https://savrn.com/models/jackod-9b-coder) Model · Text generation ### [ModujoMoE](https://savrn.com/models/modujomoe) [Modujo](https://savrn.com/model-publishers/modujo-ai) This repository keeps model variants in self-contained subdirectories. The repository name is retained for compatibility, but the repository root no longer contains model weights and must not be loaded directly. Separate 4B PT and experimental SFT snapshots are available in named directories. The SFT has not completed its teacher-distillation stage. The PT is a later continuation snapshot than the SFT initialization source; see each model card for lineage and limitations. pretrain/Modujo-1B-A0.75B/ is the released pretraining artifact. - 36 layers with 8 routed experts per layer, top-2 routing, and one shared expert - Not instruction-tuned and not intended to be treated as a chat model… Open weights 32,768 tokens transformers [View model](https://savrn.com/models/modujomoe) Model · Text generation ### [r0_96](https://savrn.com/models/r0-96) [Yonghoon](https://savrn.com/model-publishers/fiveflow) Open weights apache-2.0 8.2B parameters 32,768 tokens transformers [View model](https://savrn.com/models/r0-96) Model · Text generation ### [XcaliburBiz-1.8B-report](https://savrn.com/models/xcaliburbiz-1-8b-report) [Crissy Deutsch](https://savrn.com/model-publishers/xcaliburmoon) A lightweight report writer. A compact data summary goes in and a professional report comes out. This model is a derivative work based on Qwen1.5-1.8B-Chat by the Qwen Team, modified via Supervised Fine-Tuning and Direct Preference Optimization (DPO) by XcaliburMoon. - Supervised fine-tuning of a low rank adapter over section pairs built from the house reports. - Direct preference optimization from the supervised adapter. - Merging the aligned adapter into the base weights at full precision. - Quantization of the merged weights into the GGUF forms and one ONNX export. Apache License 2.0 grants no right to the trade names of the licensor beyond the use that describes the origin of the work.… Open weights apache-2.0 peft [View model](https://savrn.com/models/xcaliburbiz-1-8b-report) Model · Text generation ### [han2han-it](https://savrn.com/models/han2han-it) [Cellik Adams](https://savrn.com/model-publishers/cadazar) Instruction-tuned checkpoint of Han2Han (han2han-ul2-base-1-it, step 43153). Han2Han is a 169M-parameter encoder-decoder model that learns script-invariant representations of Korean text: a document written in Hanja and its Hangul transcription land at the same point in embedding space. The recipe (jamo and character-level embedding fusion, morpheme-aware denoising, bidirectional Hanja-Hangul transcription) is described in the paper, accepted to Findings of EMNLP 2026. This repo holds the PyTorch weights, the SentencePiece tokenizer, and the modeling code needed to load them through the transformers Auto classes with trustremotecode=True. Training code, the Flax model, the Flax-to-PyTorch… Open weights apache-2.0 189M parameters transformers [View model](https://savrn.com/models/han2han-it) Model · Text generation ### [Vinci-Cyber-8B-1.0-GGUF](https://savrn.com/models/vinci-cyber-8b-1-0-gguf) [SimpleDirect](https://savrn.com/model-publishers/simpledirect) GGUF distributions of SimpleDirect’s Canadian-developed, open-weight infrastructure-security model. This repository provides GGUF conversions of Vinci-Cyber-8B-1.0, the defensive infrastructure-as-code adaptation developed by SimpleDirect, a Canadian AI lab, from IBM Granite 4.1 8B. The intended application is human-reviewed security work on infrastructure you own or are authorized to assess. The full-weight model’s training centres on repair, no-change, and recovery examples. These are intended behaviours, not guarantees of correct or safe operation. The immediate source is Vinci-Cyber-8B-1.0. Its underlying parent is IBM Granite 4.1 8B, pinned to revision… Open weights apache-2.0 [View model](https://savrn.com/models/vinci-cyber-8b-1-0-gguf) Model · Text generation ### [Qwen3-1.7B-RKLLM](https://savrn.com/models/qwen3-1-7b-rkllm) [Hanzo Huang](https://savrn.com/model-publishers/hanzohuang) Text generation on Rockchip RK3576 and RK3588, converted from Qwen/Qwen3-1.7B. 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… Open weights apache-2.0 rkllm [View model](https://savrn.com/models/qwen3-1-7b-rkllm) Model · Text generation ### [Llama-3.2-1B-Instruct-ExecuTorch](https://savrn.com/models/llama-3-2-1b-instruct-executorch) [Experimental Machines](https://savrn.com/model-publishers/experimentalmachines) Built with Llama ExecuTorch exports of meta-llama/Llama-3.2-1B-Instruct (revision 9213176726f5) for on-device inference with the openweights Android app or any ExecuTorch 1.4.0 runtime. Windows in this repo: XNNPACK (CPU) at 2k to 32k; Vulkan (GPU) at 2k to 32k; Qualcomm QNN (HTP) SM8750 (Snapdragon 8 Elite) at 2k to 8k; MediaTek NeuroPilot MT6991 (Dimensity 9400) at 2k to 16k. The window is fixed inside the file: the runtime allocates the whole KV cache at load, so pick the largest window the device can hold (fitsphonebudget in each folder's config.json is the estimate against a 5 GB budget). MediaTek folders also hold the token embedding table the NeuroPilot runner reads from disk, shared… Open weights llama3.2 executorch [View model](https://savrn.com/models/llama-3-2-1b-instruct-executorch) Model · Text generation ### [Swift1.5-Qwen3.8-Flash-Next-EXL3](https://savrn.com/models/swift1-5-qwen3-8-flash-next-exl3) [Katter Mobile](https://savrn.com/model-publishers/kattermobile) Swift1.5-Qwen3.8-Flash-Next-EXL3 is a high-performance, reasoning-efficient, agentic coding model quantized to ExLlamaV3 (EXL3) at a target of 4.05 bpw. This release synthesizes three major engineering pillars: 1. The Base Foundation: Built on UkisAI's Swift 1.5 (ukisai/Swift1.5-Qwen3.8-Flash-Next), which is a distilled, reasoning-compressed evolution of Alibaba's Qwen3.8-Flash-Next (Qwen/Qwen3.8-Flash-Next). Swift 1.5 cuts overthinking tokens by 63.4% while delivering a 1.8× speedup with minimal accuracy trade-off. 2. Custom Agentic Coding Fine-Tuning: Trained on KatterMobile's proprietary multi-turn agentic coding and software engineering dataset to significantly enhance tool-use… Open weights other 262,144 tokens [View model](https://savrn.com/models/swift1-5-qwen3-8-flash-next-exl3) Model · Text generation ### [gaggle-nanochat-pretrained-climbmix-20260924](https://savrn.com/models/gaggle-nanochat-pretrained-climbmix-20260924) [Castorini](https://savrn.com/model-publishers/castorini) A 3.3B-parameter decoder-only base language model, pre-trained from scratch with nanochat on one pass over ClimbMix (287B tokens). This is the causal model described in the pre-training section of (PDF). It has had no post-training (no instruction tuning or RLHF), so it continues text rather than following instructions. The same weights load three ways: with transformers (trustremotecode=True), with vLLM (through a small plugin), and with nanochat's own code from the original checkpoint files, which are unchanged. one pass of 287B tokens. The last of its 6,543 shards is held out for validation. Documents are packed into 2,048-token sequences with nanochat's BOS-aligned best-fit packing.… Open weights mit 3.3B parameters transformers [View model](https://savrn.com/models/gaggle-nanochat-pretrained-climbmix-20260924) Model · Text generation ### [tinybrainbot-350mV3-math](https://savrn.com/models/tinybrainbot-350mv3-math) [Nk Bass](https://savrn.com/model-publishers/nkthebass) Try it in your browser: the new tinybrainbot-350mV3-math-Q80.gguf (371 MB) runs locally on your device with WebGPU, and nothing you type is sent anywhere. A 348M-parameter decoder-only model fine-tuned to do column arithmetic with a written-out scratchpad. Successor to the TinyBrainBot 320M V2 Math. TL;DR: 99.4% average across the nine GPT-3 arithmetic sub-tasks, beating the 320M V2 Math (91.5%). Adds cleanly to 14 digits, does 3-digit multipliers (98.0%), and handles negative results (85% overall, but only 58% at 5 digits - see below). It is a calculator with a scratchpad, not a word-problem solver — see the honest limits below. Use applychattemplate rather than hand-building the prompt… Open weights apache-2.0 348M parameters 2,048 tokens transformers [View model](https://savrn.com/models/tinybrainbot-350mv3-math) Model · Text generation ### [Hunter-1-4B](https://savrn.com/models/hunter-1-4b) [Lalakai Labs](https://savrn.com/model-publishers/lalakai) Hunter 4B is a 4-billion-parameter conversational reasoning model from Lalakai Labs, built for local inference on consumer hardware. It works through problems step by step inside... traces before giving its final answer, and is tuned for reasoning, agentic coding, tool use, and cybersecurity analysis. Weights are distributed in GGUF format for llama.cpp, Ollama, LM Studio, and other GGUF-compatible runtimes. GGUF stores tensors only and does not use pickled code. - 4B parameters, optimized for reasoning, coding, and security tasks - Visible reasoning in traces for math, code, and multi-step problems - Q4KM quantization runs in roughly 3.5 GB of working memory at a 4K context - Apache-2.0… Open weights apache-2.0 [View model](https://savrn.com/models/hunter-1-4b) Model · Text generation ### [luna-2.1-base](https://savrn.com/models/luna-2-1-base) [AobanZ](https://savrn.com/model-publishers/aobangaming) Luna 2 is a next generation text model designed for keeping context over long amounts of text. Luna 2 is generally intended for chatbot purposes. Luna utilizes FlashAttention and AdamW for performance and capability. Lightning is designed to provide quick, coherent outputs, improved with a larger size and weight. Luna is intended to be used for research, analysis and fine-tuning, general conversation, stories and other. It is not intended to be used for professional advice, real writing or any kind of heavy work as generated outputs may be incorrect. Luna can be used directly for text generation, experimentation, and conversational interactions. Users can provide text prompts and generate… Open weights mit 115M parameters transformers [View model](https://savrn.com/models/luna-2-1-base) Model · Text generation ### [Vinci-Piccolo-1.0-GGUF](https://savrn.com/models/vinci-piccolo-1-0-gguf) [SimpleDirect](https://savrn.com/model-publishers/simpledirect) GGUF (quantized) builds of Vinci Piccolo, for local inference with Ollama, LM Studio, and llama.cpp. For the full-precision weights, evals, and details, see simpledirect/Vinci-Piccolo-1.0. No GGUF file in this repository has been evaluated. The benchmark figures on the full-precision card were measured on the unquantized bf16 weights, not on any file here. Quantization shifts scores, and with no per-tier measurement this card cannot say in which direction or by how much for any tier. This card therefore publishes no per-tier score, no quality ranking between the tiers, and no recommended tier. The file sizes and memory figures below are measured, and describe storage and memory cost only. A… Open weights apache-2.0 gguf [View model](https://savrn.com/models/vinci-piccolo-1-0-gguf) Model · Text generation ### [laura-en-qwen3-4b-GGUF](https://savrn.com/models/laura-en-qwen3-4b-gguf) [Bouka](https://savrn.com/model-publishers/theboukadida) Laura is the tutor in an offline German course app for adult beginners (A1–A2). This model talks to the learner in English and teaches German with German examples. It runs on the phone, CPU only. This is a modified version of Qwen/Qwen3-4B-Instruct-2507. Changes made: a LoRA adapter (rank 16, all linear layers, 2 epochs) was trained on ~2,500–3,300 short tutoring conversations in English, merged into the weights, and the result converted to GGUF and quantised to Q4KM with llama.cpp. Distributed under the base model's Apache-2.0 licence. The app does not trust the model to judge German on its own. It checks the learner's sentence with its own dictionary first and adds the verdict to the… Open weights apache-2.0 [View model](https://savrn.com/models/laura-en-qwen3-4b-gguf) Model · Text generation ### [SmolVLM-500M-Instruct-Q4_K_M-GGUF](https://savrn.com/models/smolvlm-500m-instruct-q4-k-m-gguf) [Shayon Sarker](https://savrn.com/model-publishers/shayonsarker) Downloads: Hugging Face — ShayonSarker/SmolVLM-500M-Instruct-Q4KM-GGUF · Source & build recipes: GitHub — model-SmolVLM-500M-Instruct-gguf · Base model: HuggingFaceTB/SmolVLM-500M-Instruct · Failed vision attempt: MM-GGUF The GGUF binaries are hosted on Hugging Face; this repository holds the reproducible conversion pipeline, calibration data, and checksums. Base: HuggingFaceTB/SmolVLM-500M-Instruct (Apache-2.0) · Quant: llama.cpp 0.4.1-dev · imatrix on 3000-line wikitext-2 (seed 42, 625 chunks, 4 threads) Scope correction: this repository holds the text tower only (291 tensors, llama arch — verified: image input is rejected). The SigLIP vision encoder and mmproj are NOT included, so this… Open weights apache-2.0 [View model](https://savrn.com/models/smolvlm-500m-instruct-q4-k-m-gguf) Model · Text generation ### [BlazerApex-2B](https://savrn.com/models/blazerapex-2b) [Davi Rediske de Oliveira](https://savrn.com/model-publishers/davizig10jojo) BlazerApex-2B é um modelo de linguagem ultracompacto (2.5B parâmetros), altamente otimizado e fine-tuned com foco total em Português do Brasil (PT-BR), Programação, Lógica e Matemática. Projetado especificamente para ser o SOTA (State-of-the-Art) de bolso, este modelo roda 100% offline, de forma rápida e fluida em smartphones intermediários (como o Motorola Moto G34 5G com Snapdragon 695), sem precisar de internet ou servidores externos. Este repositório contém a versão quantizada IQ4XS (~1.4 GB), ideal para dispositivos com 4GB a 8GB de RAM. 1. Baixe o arquivo BlazerApex-2B-IQ4XS.gguf direto no seu celular. 2. Instale o aplicativo Maid ou Layla na Play Store. 3. Abra o app, vá em "Models"… Open weights apache-2.0 transformers [View model](https://savrn.com/models/blazerapex-2b) Model · Text generation ### [Huihui-Kolibri-1-abliterated-GGUF](https://savrn.com/models/huihui-kolibri-1-abliterated-gguf) [Huihui.ai](https://savrn.com/model-publishers/huihui-ai) This is an uncensored version of Aleph-Alpha/Kolibri-1 created with abliteration. This ablation was performed entirely using llama.cpp's GGUF model, without relying on Transformers, For the specific ablation method, please refer to cvector-generator. GGUFs come from Hob-forge/Kolibri-1-GGUF. The ablation did not modify all expert modules. Stock llama.cpp does not support the kolibri1 architecture yet. Apply kolibri1-llama.cpp.patch (included in this repo) to llama.cpp at upstream commit 836d571, then build: - Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content.… Open weights apache-2.0 vllm [View model](https://savrn.com/models/huihui-kolibri-1-abliterated-gguf) Model · Text generation ### [ProactiveInquirer-Qwen3-8B-GGUF](https://savrn.com/models/proactiveinquirer-qwen3-8b-gguf) [Ido Levy](https://savrn.com/model-publishers/dolev31) Ido Levy 1,2 · 1,2 GGUF quantizations of the trained questioner from Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents, for llama.cpp, Ollama, LM Studio and Jan. They were made from the merged model with llama.cpp (commit 9adc7f4). Before upload, each file ran the adapter card's two-turn example with greedy decoding. Every file asked the same two questions as the full-precision model: "Who directed the film The Great Flamarion?" and, once the evidence named the director, "Who was the spouse of film director Anthony Mann?". The results are the trained questioner's, as the paper reports them: see the adapter card's Results. The paper evaluated the unquantized… Open weights apache-2.0 [View model](https://savrn.com/models/proactiveinquirer-qwen3-8b-gguf) ## Questions ### Which Text generation models are most downloaded? By monthly downloads reported by the Hugging Face Hub: EdgeRock-4B-GGUF (714); Shreyansh-STEM-AI-2B-v3 (660); Qwen3.8-Flash-Next-125B-A5B-INT4-AutoRound (612). ## Other Tasks - [Image and Text to Text](https://savrn.com/models/tasks/image-and-text-to-text) 588 - [Robotics](https://savrn.com/models/tasks/robotics) 403 - [Feature Extraction](https://savrn.com/models/tasks/feature-extraction) 400 - [Text Classification](https://savrn.com/models/tasks/text-classification) 335 - [Text to Image](https://savrn.com/models/tasks/text-to-image) 236 - [Speech Recognition](https://savrn.com/models/tasks/speech-recognition) 190 - [Text to Speech](https://savrn.com/models/tasks/text-to-speech) 148 - [Image Classification](https://savrn.com/models/tasks/image-classification) 140 - [Audio Classification](https://savrn.com/models/tasks/audio-classification) 130 - [Sentence Similarity](https://savrn.com/models/tasks/sentence-similarity) 103 - [Reinforcement Learning](https://savrn.com/models/tasks/reinforcement-learning) 102 - [Token Classification](https://savrn.com/models/tasks/token-classification) 101 [See all](https://savrn.com/models/tasks)