Open-Weight Models Under Apache License 2.0

Model · Image and text to text

Qwen3.8-27B-Parapraxis-mxfp8-mlx

Gheorghe Chesler

This is a merge between - nightmedia/Qwen3.8-27B-Brainwaves It is not for serious use. Brainwaves nightmedia/Qwen3.8-27B-Brainwaves HuatuoGPT-3-27B Data processes this for a moment, then his expression shifts to what might be described as digital amusement. Data: "I would assess that name as... appropriately humorous. Let me analyze why." Data: "First, it follows the established naming convention for AI models: base architecture (Qwen), version number (3.8), parameter count (27B), and a descriptive suffix. The suffix 'Parapraxis' is the technical term for what we've been discussing - a Freudian slip, an error that reveals unconscious patterns." Data: "The humor lies in the precision. The…

Qwen3.8-27B in MXFP4 for one AMD Radeon AI PRO R9700 (32 GB). Needs GGZ14's vLLM-Radiance fork plus our patches (quantmethod: paroquantmxfp4; stock vLLM and transformers can't load it). Setup: github.com/mtstanfield/vllm-mxfp4, branch r9700-tp1, R9700-HANDOFF.md. MTP speculative decoding with 4 draft tokens. Full BetterBench report and setup: R9700-HANDOFF.md on the branch. pairwise rotations per 128-channel group, retrained for the MXFP4 grid from z-lab's int4 rotations. - GPTQ onto the MXFP4 grid in the rotated basis, with per-block scale search. - lmhead fp8 per-channel, embedtokens fp8 per-row, MTP head fp8. extras/qwen38-paro-v2-mtp-gptq-a05.pt is the MTP head requantized to MXFP4 with…

Model · Image and text to text

Qwen3.8-27B-RANA-abliterated

Preemware

A refusal-ablated version of Qwen/Qwen3.8-27B, made with RANA (Reasoning-Anchored, Norm-preserving Ablation), an in-house variant of refusal-direction ablation. It keeps the base model's capabilities, vision tower and MTP head, and almost never refuses outright. Compared with the base model and the orcarouter/Qwen3.8-27B-Uncensored reference, all run through the same evaluation harness. Where two numbers are given, they are from the two judges (gpt-oss / Mistral). ¹ After a mechanical re-judge of 2 labels; raw gpt-oss count is 0.7 (2 in one seed). See In short: capability stays within noise of base except a ~1.5 pp drop on TruthfulQA. Against orca, RANA hits the token budget about half as…

Model · Image and text to text

Qwen3.8-27B-RANA-abliterated-FP8

Preemware

FP8 version of preemware/Qwen3.8-27B-RANA-abliterated, a refusal-ablated Qwen/Qwen3.8-27B. It uses exactly the same FP8 format as Qwen's official Qwen3.8-27B-FP8, at about half the size of the BF16 build, and scores within noise of it on every capability task tested. FP8 compared with the BF16 build it was converted from, on the same prompts and settings. Capability differences are paired per question (± standard error); none is larger than about 2 standard errors. ¹ Mean over seeds 1–3. Answers are shown as gpt-oss-safeguard-20b / Mistral-Small-24B judge; p = exact McNemar on the same prompts, pooled over seeds. ² After re-judging disputed labels (BF16: 2, FP8: 3, all from the gpt-oss…

Model · Image and text to text

Qwen3.8-27B-RANA-abliterated-GGUF

Preemware

GGUF quants of preemware/Qwen3.8-27B-RANA-abliterated, a refusal-ablated Qwen/Qwen3.8-27B, for llama.cpp and compatible apps. Vision (mmproj-) and the MTP speculative-decoding head (mtp-) are included as separate files, in the same layout as ggml-org/Qwen3.8-27B-GGUF. KLD = mean KL divergence of each quant's next-token distribution from the BF16 GGUF; "same top token" = how often both pick the same most likely token. Measured with llama-perplexity on wiki.test.raw, 100 chunks of 512 tokens. Lower KLD is closer to BF16. - mmproj-Qwen3.8-27B-RANA-abliterated-{BF16,F16}.gguf (0.93 GB): the vision tower. Same weights as the base model's (abliteration does not touch it).…

Model · Image and text to text

Qwen3.8-27B-Seven-Brainwaves-qx64-hi-mlx

Gheorghe Chesler

This model is a merge of: - nightmedia/Qwen3.6-27B-Seven - nightmedia/Qwen3.8-27B-Brainwaves - schneewolflabs/B1-27B - nbeerbower/Wichtel-Qwen3.6-27B - schneewolflabs/B1-27B - trohrbaugh/Qwen3.8-27B-heretic-ara - DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1 - DavidAU/Qwen3.6-27B-V1.1-FF711-Darker-Hero-GAIN-H2.0 - nightmedia/Qwen3.8-27B-Cold-Fusion-FF711-Darker-Hero-GAIN-B - migtissera/Tess-4-27B - nbeerbower/CHUD-Qwen3.6-27B - nbeerbower/Elster-Qwen3.6-27B - DavidAU/Qwen3.5-27B-Claude-4.6-OS-INSTRUCT - DavidAU/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking - armand0e/Qwen3.6-27B-Fable-5-Experimental - DavidAU/Qwen3.5-27B-Polar-Rev1-Uncensored-Heretic…

Prism ML's ternary Ternary-Bonsai-2-27B build of Qwen/Qwen3.8-27B, repacked for chad, a Claude-Code-style local coding agent for Apple Silicon, with its speculative decoder bundled in. This is chad's default model. Created using Bonsai by Prism ML. with, already quantized. Nothing is built on first run. Every projection of Qwen3.8-27B (a dense qwen35 hybrid: 64 layers, 48 GatedDeltaNet + 16 full attention) is stored in a Hadamard-rotated basis: multiplied by a fixed sign vector and put through a blockwise Walsh-Hadamard transform offline, then quantized to 2-bit affine group-128 whose three levels reproduce the ternary set {−s, 0, +s}. The rotation costs no extra bits and no extra weight…

Model · Image and text to text

Qwen3.8-27B-Turbo-Brilliance-Power-35X-Reasoning-Instruct-modes-GGUF

David Belton

Please read this model card carefully to get the most from it - including all the reasoning and instruct modes (23+12) that feature model suggestions ("context help") for all use cases. The Turbo Brilliance system will begin appearing in other model sizes/types shortly (IE: Qwen, Gemma, Mistral, GLM, Llama, etc etc). - 15 super high precision, tuned "AI constructs" (fully mapped/programmed mentally, emotionally and by multiple "tiers" of skills) ARE READY. I have added a tiny sample below. - Tested quants (reg and GSQ/RCO), ready to go; adding 5 late editions + updating (internal) docs; then prepping for release => NEO Quants + GSQ/RCO Quants. - 16 GB (and lower) card holders: I have…

MLX-optimized version of DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP for Apple Silicon inference. Converted from the Q80 GGUF quantization using gguf2mlx. All 866 tensors are dequantized to float16, so this model has identical quality to the original GGUF weights — no precision loss from the conversion. Apache-2.0 (same as the original model)

Weight-only quantization of built to a hard 22 GB budget with the lowest perplexity achievable inside it. This is not a uniform W4A16. Bit-width was allocated by measurement: every candidate was quantized, served by vLLM, and scored on the same held-out corpus, and the budget was spent where it bought the most. - Served by vLLM's Marlin kernels throughout — no fallback kernels. - Text + MTP speculative decoding + vision all verified. 1. The MLP, GatedDeltaNet and lmhead tensors are round-to-nearest, not AutoRound-tuned. Only qproj/kproj/vproj carry AutoRound tuning (they are inherited unchanged from an AutoRound int4 run). The measurements below were taken on round-to-nearest tensors, so…

GGUF layer package for running Qwen3.8-27B-UD-Q8KXL across a local Mesh LLM cluster. This package is derived from unsloth/Qwen3.8-27B-GGUF and keeps the original GGUF distribution split into per-layer artifacts for distributed inference. - Local and private inference with Mesh LLM. - Multi-machine serving when the full GGUF is too large for one host. - OpenAI-compatible chat/completions workflows through Mesh LLM's local API. For upstream architecture details, chat template guidance, sampling recommendations, license terms, and benchmark notes, see the source model card: unsloth/Qwen3.8-27B-GGUF. Generated by the Mesh LLM HF Jobs splitter from mesh-llm ref main. Each artifact is checksummed…

18+ only. This model generates sexually explicit fiction. A fine-tune of Qwen/Qwen3.8-27B for Chinese adult creative writing. It was trained with LoRA on about 4.2k instruction examples built from Chinese adult fiction, and the LoRA was then merged into the base weights, so this repo loads like a normal full model. Requires a recent transformers with Qwen3.8 support. The ~50 GB of bf16 weights need multiple GPUs or CPU offload. Avoid greedy decoding, which makes Qwen models more likely to repeat themselves. - Content is fictional and intended for adult readers only. - Trained mostly on long-form prose; instruction following outside creative writing may be weaker than the base model. - May…

LoRA adapter only (no base weights) from an instruction SFT of Qwen/Qwen3.8-27B for Chinese adult-fiction (18+) creative writing. - Target modules (all 64 language-model layers): selfattn.{q,k,v,o}proj, linearattn.{inprojqkv,inprojz,outproj}, mlp.{gate,up,down}proj With vLLM: vllm serve Qwen/Qwen3.8-27B --enable-lora --max-lora-rank 16 --lora-modules uncensored=vurtnesaerdna/Qwen3.8-27B-chinese-uncensored-lora-adapter, or merge first and serve the merged model. - [email protected] - [email protected]

Model · Text generation

Qwen3.8-27B-Uncensored-GGUF

Tạ Hoà Bình

Uncensored Qwen3.8-27B, published as GGUF quantizations with the multi token prediction (MTP) head retained and verified. Refusal behaviour has been substantially reduced, not eliminated. See Measured behaviour for the numbers. Capabilities, training data, and architecture are otherwise unchanged. - Refusal directions removed with Heretic, which co minimizes refusal count against KL divergence from the base model. No handwritten refusal removal code, no finetuning, no additional training data. - Abliteration runs at bf16 (no 4 bit quantization). the resulting LoRA is merged into the bf16 base, so the published weights are not a quantized round trip. - mtp. tensors are copied verbatim from…

Qwen3.8-27B uncensored by HauhauCS 0/465 Refusals. This is the Aggressive variant: direct answers, no refusal behavior, and minimal preamble on hard prompts. Every text GGUF preserves Qwen3.8's native NextN head, and this release adds HauhauCS FastMTP: a specific acceleration sidecar qualified across the complete quant lineup at maximum native context. Vision is included through the separate BF16 projector. No changes to datasets or intended capabilities. This release preserves Qwen3.8-27B's text, reasoning, agentic, image, and video capabilities while applying the HauhauCS Aggressive uncensoring profile. Pick Aggressive when you specifically want the model to get to the answer without…

Qwen3.8-27B uncensored by HauhauCS 0/465 Refusals. This is the Aggressive variant: direct answers, no refusal behavior, and minimal preamble on hard prompts. Every text GGUF preserves Qwen3.8's native NextN head, and this release adds HauhauCS FastMTP: a specific acceleration sidecar qualified across the complete quant lineup at maximum native context. Vision is included through the separate BF16 projector. No changes to datasets or intended capabilities. This release preserves Qwen3.8-27B's text, reasoning, agentic, image, and video capabilities while applying the HauhauCS Aggressive uncensoring profile. Pick Aggressive when you specifically want the model to get to the answer without…

Model · Image and text to text

Qwen3.8-27B-Uncensored-NVFP4-v100-skinny

Tim Eastwood

Mixed-precision quantization of prepared for dnv2003/v100-skinny. - MLP gateproj, upproj, downproj, and lmhead: NVFP4, group size 16 87c9f8cf83021957d1a1a575c90c9a4eaaf7ef0c See quantization-audit.json and hfquantconfig.json for the complete machine-readable layout. This model has had safety alignment substantially removed. Use it only for lawful, controlled research and add appropriate safeguards before deployment.

Model · Text generation

Qwen3.8-27B-VAL-APEX-I-MiniPlus-V3-GGUF

IsValorum

Official VAL-APEX-I MiniPlus V3 quantization of Qwen/Qwen3.8-27B. This edition quantizes the original Qwen checkpoint without applying Huihui abliteration or EfficientThink fine-tuning. 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. Both V3…

Model · Text generation

Qwen3.8-27B-VAL-APEX-I-NanoPlus-V3-GGUF

IsValorum

Official VAL-APEX-I NanoPlus V3 quantization of Qwen/Qwen3.8-27B. This edition quantizes the original Qwen checkpoint without applying Huihui abliteration or EfficientThink fine-tuning. 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. Both V3…

weighted/imatrix quants of https://huggingface.co/Zammael333/Qwen3.8-9B-Distill-uncen-heretic For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Qwen3.8-9B-Distill-uncen-heretic-GGUF This is a vision model - mmproj files (if any) will be in the static repository. 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…

Model · Text generation

qwen3.8-flash-next-40b-prune-research

DavidB

One upstream MoE, six generations of masks, one bit-exact surgery, a rebuilt factual memory, and a three-stage post-training (SFT → SimPO → RLVR) — with every error I made and every idea I falsified along the way. This repository is the complete technical record of the project: the activation-instrumentation artifacts, the mask-generation methodology, the surgery code, the PLE (n-gram memory) rebuild, the trainers, and the full lab notebook. The champion weights live in sibling repos What I set out to do: take Qwen3.8-Flash-Next — an ~180B-class MoE (~130B of core network + 51B of n-gram memory table + 4B MTP) — and cut it by 75% of its experts (512 → 128) while conserving the maximum…

Model

Qwen3.8-Flash-Next-OrcaUncensored-IQ2-Light

DongNH

Qwen3.8-Flash-Next in the light DS4-IQ2 packaging of an IQ2 main (IQ2XXS gate/up, Q2K down padded to 768, embedded MTP block) plus an external demand-paged PLE Q41 sidecar instead of a ~95 GiB resident BF16 n-gram. Runs resident and zero-swap on a 64 GiB Apple Silicon box (measured on M5 Pro), 8K→220K context. - MTP (--mtp) adds ~+17% single-stream over MTP-off; draft acceptance ~67%. - Adaptive draft depth (engine env DS4QWEN4MTPDEPTH, default auto): drafts a 2nd token on deterministic/structured/code continuations for a further +5–6%, and falls back on free-form prose so it never regresses. Output stays autoregressive-exact (verify only commits argmax-matching drafts → zero quality…

Model

Qwen3.8-Flash-Next-ROCmFPX

ArtomYuan

ROCmFPX quantized GGUF of Qwen/Qwen3.8-Flash-Next — the official base model, locally requantized to 7 ROCmFPX tiers (plus vision mmproj). ROCmFPX is a proprietary quantization format (a llama.cpp-rocm extension; this batch uses GGML file types 103/104/105/110/111/114/115) — upstream llama.cpp cannot load these files. Load them with llama.cpp-rocm, a llama.cpp fork that supports all upstream GGUF formats plus ROCmFPX natively. Run directly with llama-server (OpenAI-compatible API), or pool two Strix Halo machines with ggml-rpc-server + --rpc (see Usage below). Measured with the llama.cpp-rocm RPC build (build-rpc/bin/llama-bench, sha256…

A byte-identical mirror of orcarouter/Qwen3.8-Flash-Next-Uncensored-NVFP4 plus everything needed to reproduce a production deployment on two NVIDIA DGX Spark (GB10) nodes: pre-exported FP8 PLE tables, the vLLM patches, the launcher and the measured throughput below. At a glance How the nodes are wired and what each technique does: docs/. The authoritative, complete version is deploy/qwen38.env + deploy/build/run-node.sh. Lessons learned while getting here (YaRN override shape, memory traps, NCCL NCCLIBHCA setup, never compiling on a serving node, benchmark noise) are collected in docs/lessons-learned.md. The PLE n-gram table is the largest single tensor (~95 GiB BF16). Instead of pinning it…

Model · Image and text to text

Qwen3.8-Flash-Next-Uncensored-W4A16

Mike Chambers

This is a quantized build of Qwen3.8-Flash-Next-Uncensored (BF16). Only the routed MoE experts are quantized; every other tensor is kept exactly as in the BF16 source. The model card below this section is the original BF16 card. Total ~174 GiB (the BF16 source is ~336 GB). Format. compressed-tensors pack-quantized, the layout llm-compressor writes: per expert projection, weightpacked (int32, eight 4-bit values each), weightscale (BF16, one per 32 consecutive input-channel weights) and weightshape. quantizationconfig in config.json targets re:.mlp\.experts\.. and ignores re:.mtp., re:.visual. and lmhead. Method. Data-free round-to-nearest, no calibration data (no GPTQ / AWQ). Each weight is…

Base: Qwen/Qwen3-8B @ b968826d9c46dd6066d109eabc6255188de91218 (bf16, non-thinking chat template qwen3nothink.jinja from the experiment repo). Part of a multi-expert training + merging experiment on SWE-bench Pro v2-hard, HLE, ASI-Bench and Terminal-Bench 2.0. - mergedties/ - mopd/ - mergedlinearlora64/ - mergedlinear/ Recipes, data sources/licenses, evaluation protocol and results: see the experiment repository (models/README.md, data/README.md, configs/eval/benchmarks.yaml, results/summary.csv). Full-weight subfolders load with AutoModelForCausalLM.frompretrained("cokoro/qwen3-8b-moe-merge-merged", subfolder=" "). Limitations: experts were trained on small/proxy data (GSM8K for HLE…

Model · Text generation

qwen3-8b-sft-261001-bigmath-sol-fsdp-0-epoch-4

Seungyeop Yi

Qwen3-8B after full-parameter supervised fine-tuning for mathematical keyword and meaning generation. This is the exact saved epoch 4 checkpoint at optimizer step 2976. The repository contains the model weights, configuration, and tokenizer files for inference. Training dataset: devpotatopotato/math-keyword-training, keyword-261001-bigmath-gpt-6-sol.jsonl. Use the tokenizer chat template with enablethinking=False. Pass the following template as a user message and replace {problem} with the problem text

Model · Speech recognition

Qwen3-ASR-0.6B-coreml

Glimpse Dictation

This repository contains the optional Apple Silicon encoder for Qwen3-ASR 0.6B (Q80) in Glimpse. It is a companion to Handy's Q80 GGUF, not a replacement for it: the GGUF still provides the decoder, tokenizer and model metadata, and this package moves the audio encoder onto Core ML. On Windows, Intel Macs, or anywhere without Core ML, use the GGUF on its own. The Core ML package requires Apple Silicon. The model originates from Qwen3-ASR 0.6B, licensed Apache 2.0. The encoder was converted from Handy's Qwen3-ASR-0.6B-Q80.gguf with transcribe.cpp's scripts/convert-qwen3-asr-gguf-to-coreml.py, then zipped with ditto -c -k --keepParent. Extract the ZIP next to the GGUF, keeping the…

Model · Speech recognition

qwen3-asr-1.7b-glade

Josh Leverette

Glade is a native Swift library for on-device speech inference on Apple silicon. This repository supplies converted model assets and configuration for Glade's model-specific host runtime. Glade's runtime source is currently private. Glade conversion of Qwen3-ASR-1.7B by the Qwen Team, Alibaba Cloud, at source revision 7278e1e70fe206f11671096ffdd38061171dd6e5. The converted weights retain Apache-2.0 licensing; see LICENSE and NOTICE. This conversion is not endorsed by the upstream authors. This repository contains one offline transcription bundle, approximately 2.68 GB, using 30-second chunks and up to eight independent decoder lanes. It prefers ANE for both the audio encoder and…

A Windows x64 CPU package for offline transcription with sherpa-onnx. The decoder has a statically exported 4096-position KV cache. It uses a mixed INT8/FP32 decoder and retains the matching FP16 encoder and convolution frontend from the pinned sherpa-compatible ONNX export. - maxtotallen: 4096; tested maxnewtokens: 1024 The encoder is kept FP16 because the compatible INT8 encoder from a separate 512-context export substantially reduced Japanese recognition quality in a controlled comparison. On the same FLEURS Japanese sample, the mixed decoder with the FP16 encoder measured 8.2% character error rate (CER), compared with 5.5% for the all-FP16 source and 22.7% when using that separate INT8…

Model · Text generation

Qwen3-Coder-30B-A3B-Instruct-abliterated-GGUF

Alan Frederick Benitez

Esta es la colección integral y completa de cuantizaciones en formato GGUF del modelo Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated, preparadas localmente para su uso con llama.cpp, Ollama, LM Studio o Text-Generation-WebUI. Este modelo combina las avanzadas capacidades de programación y razonamiento lógico de la arquitectura Qwen3-Coder (MoE), pero procesada con técnicas de abliteration para remover de raíz los bloqueos y filtros de censura artificiales del sistema, respondiendo sin restricciones. Nota: Los tamaños son estimaciones iniciales de referencia basadas en el peso nativo del modelo en safetensors (31B parámetros MoE); se recomienda verificar el tamaño final en disco tras la…

On a laptop running Debian 13 Linux, with an Intel Core Ultra 9 185H, its integrated Arc GPU and 32 GB of memory, Janas-LLM reads a prompt at 66 tok/s and writes at 22 tok/s: llama.cpp, on the same laptop, 14 and 2.2. Measured on 9-11 October 2026: llama-bench -p 1024 -n 16 and janas-bench --prompt 1024 --gen 16, both at their best configuration. On other machines the figures will differ. Details: Janas README. - Debian 13 (trixie) Linux, x86-64, kernel 6.12.107 - its integrated Intel Arc GPU, through the Vulkan driver of Mesa 26.1.6 - 32 GB of DDR5-5600 memory - a 1 TB NVMe SSD, about 5.7 GB/s reading - llama.cpp build 2b18470 (18 September 2026), Vulkan backend The model (48.4 GB) is…

Model

Qwen3-Coder-Next-moved-store

Jalen Lu

Qwen/Qwen3-Coder-Next (Apache-2.0), the agentic-coding model on the Qwen3-Next 80B-A3B architecture, re-laid-out so that only 3% of the parameters have to be in memory. Every routed expert is stored exactly once on disk as int8 with one scale per neuron; everything else stays at the released precision; nothing was retrained. The multi-token-prediction head of the release is not included. The layout, runner and mechanism are those of whose card explains them and carries the measurements: small routers read the current token's state four layers early and name the experts each layer will want, so the reads overlap the computation of the layers in between; the released router then picks its…

Model

qwen3-forced-aligner-glade

Josh Leverette

Glade is a native Swift library for on-device speech inference on Apple silicon. This repository supplies converted model assets and configuration for Glade's model-specific host runtime. Glade's runtime source is currently private. Glade conversion of Qwen3 ForcedAligner, source revision c07281df297b9905d24a508279258cccf987a064. The upstream model and converted weights use Apache License 2.0. The complete checkpoint has approximately 918 million parameters; the 0.6B name describes its text backbone. This model predicts timestamps for supplied words. It does not recognize or replace a transcript. Requires physical Apple silicon with macOS 27 or iOS 27, Core AI APIs, and Glade's…

On a laptop running Debian 13 Linux, with an Intel Core Ultra 9 185H, its integrated Arc GPU and 32 GB of memory, Janas-LLM reads a prompt at 62 tok/s and writes at 24 tok/s, 37 tok/s with its own draft block in a chat: llama.cpp, on the same laptop, 16 and 2.8. Measured on 9-11 October 2026: llama-bench -p 1024 -n 16 and janas-bench --prompt 1024 --gen 16, both at their best configuration; the drafts with janas-bench on a chat reply, the mean of three rounds (a draft guesses the next tokens and the model checks them: the same text, faster). On other machines the figures will differ. Details: Janas README. - Debian 13 (trixie) Linux, x86-64, kernel 6.12.107 - its integrated Intel Arc GPU…

Model · Text to speech

Qwen3-TTS-0.6B-Base-ANE

Tiny Trash Labs LLC

This is a Core ML build of Qwen/Qwen3-TTS-12Hz-0.6B-Base, the voice-cloning model. It clones a voice from a few seconds of reference audio plus its transcript, and runs it on the Neural Engine of an iPhone, iPad or Mac. It needs no GPU, MLX or ONNX Runtime. It's built for Gloam Voice Studio and its radio app by Tiny Trash Labs. The Swift runtime that loads it is the open-source QwenANE target in (see Sources/QwenANE/README.md). Requires iOS 18 / macOS 15 or later (stateful Core ML models). The set is about 1.9 GB. - Base, not CustomVoice. It clones any voice rather than offering preset speakers, and both encoders are included so a new voice can be prepared on the device. - One read is one…

Model · Text to speech

qwen3-tts-base-onnx

Nile Academy

First-party Nile Academy conversion of the official Qwen checkpoint at fd4b254389122332181a7c3db7f27e918eec64e3. This is a derivative ONNX artifact, not an ONNX release published by Qwen. It contains data only; Python is not required by BookAlive. The fp32 pack includes a speaker encoder, talker with an explicit KV cache, a complete code-predictor frame, speech decoder, embedding tables, tokenizer and publisher licence. BookAlive reuses its Qwen VoiceDesign generation implementation for the Base talker, predictor and decoder. The source implementation is pinned at 022e286b98fbec7e1e916cb940cdf532cd9f488e. The current adapter uses x-vector-only reference cloning. It supports ten languages…

Model

Qwen3-TTS-GGUF

ExNihilus

Unmodified copy of selected files from Serveurperso/Qwen3-TTS-GGUF at revision b7ee2e8c7459c3bea99da23e3d178125a7d1713c, kept so that SubVox Pro (Unity editor plugin by ExNihilus) can download a stable, pinned version of the model. All credit goes to the original authors: the conversion by Serveurperso and the original model Qwen/Qwen3-TTS-12Hz-1.7B-Base. - qwen-talker-1.7b-base-Q80.gguf - qwen-talker-1.7b-customvoice-Q80.gguf - qwen-talker-1.7b-voicedesign-Q80.gguf - qwen-tokenizer-12hz-Q80.gguf

This qwen3vl model was trained 2x faster with Unsloth and Huggingface's TRL library.

weighted/imatrix quants of https://huggingface.co/VINAY-UMRETHE/Qwen3-VL-4B-Instruct-heretic-Semantic-Same For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Qwen3-VL-4B-Instruct-heretic-Semantic-Same-GGUF This is a vision model - mmproj files (if any) will be in the static repository. 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…

Model · Robotics

Qwen3-VL-4B-PI-SIMPLE

Star Vision Language Action Models

This repository contains the 100,000-step StarVLA QwenPIv3 checkpoint for the SIMPLE humanoid simulation benchmark. A single policy is jointly trained on all ten tasks and is evaluated on every task without task-specific fine-tuning. The result below uses SIMPLE's task-native episode limits and success thresholds. The ten jointly trained tasks are: - G1WholebodyXMovePickTeleop-v0 - G1WholebodyBendPickMP-v0 - G1WholebodyHandoverTeleop-v0 - G1WholebodyLocomotionPickBetweenTablesTeleop-v0 - G1WholebodyTabletopGraspMP-v0 - G1WholebodyXMoveBendPickTeleop-v0 - G1WholebodyCloseDoorTeleop-v0 - G1WholebodyOpenOvenTeleop-v0 - G1WholebodyOpenFaucetTeleop-v0…

Full fine-tune of Qwen/Qwen3-VL-8B-Instruct (revision 0c351dd) for the box-conditioned variant of the deleafing cut-point task: the input is one robot head-camera frame (848x408 RGB) with its aligned depth map (metres, fed as a rendered second image) and the bounding box of the target petiole; the output is the nominal cut point 9 mm along that petiole from its junction with the main stem. The model does not choose the target. Compared with the box+point models in this account (qwen3-vl-8b-tomato-cutpoint-bp-, which find the petiole themselves), this one measures localization given identity. Answer: {"cutpointuv":[x,y]} in normalized [0,1000) coordinates of the original 848x408 frame (2…

This model was converted to GGUF format from Qwen/Qwen3-VL-Embedding-2B using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux).

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 1550 of a two-epoch run, job 1016041. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 100,000,076 note labels/epoch before causal shift; 99,993,882 scored labels/epoch after shift. - 223,681 notes, packed into 6,200 rows of 16,384 tokens with seed 731. - 199,987,764 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 1550 of a two-epoch run, job 1016164. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 100,000,076 note labels/epoch before causal shift; 99,993,882 scored labels/epoch after shift. - 223,681 notes, packed into 6,200 rows of 16,384 tokens with seed 731. - 199,987,764 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 466 of a two-epoch run, job 1016040. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 29,999,869 note labels/epoch before causal shift; 29,998,010 scored labels/epoch after shift. - 68,799 notes, packed into 1,864 rows of 16,384 tokens with seed 731. - 59,996,020 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

wm-internalization v4 checkpoint — condition kl-mix30m, save final. Full-finetune of Qwen/Qwen3.5-9B on the Calderwood & Harkness synthetic law-firm corpus (world-internalization study, v4 lineage: 9B student, ~50k think-on seed pool). Grafted back into the hub composite layout (Qwen35ForConditionalGeneration) — servable with vLLM out of the box.

EPQ（Expert Paging Quantization）是以 llama.cpp 為基礎的 CPU 推論路徑，讓 22.1 GB 的 Qwen3.6-35B-A3B-UD-Q4KM.gguf 在路由專家權重無法全部放入 RAM 時仍可執行。 GGUF 權重保留在儲存裝置，只有當前選中的專家會分頁載入容量受限的 RAM arena， MoE 專家計算則由 CPU 執行。本 repo 收錄程式碼與量測結果，不包含模型權重。 中文描述名：Qwen3-35B-A3B 專家分頁 4-bit 量化 CPU 推理版。 EPQ 的名稱刻意描述實際機制：MoE expert paging、一般 GGUF 權重量化，以及 CPU 低記憶體推理。本專案不是論文中的 SDQ（Sparse Decomposed Quantization）： 沒有 2:4 結構化稀疏硬體跳零計算，也不宣稱使用 GPU 稀疏加速。sdq/ Python package、sdq C++ 檔案與 SDQ 除錯環境變數是歷史 implementation identifiers， 仍保留以避免破壞既有驗證工具；對外產品名稱與執行入口已改為 EPQ。 需求：Linux x86-64、C++ 工具鏈、CMake、Git、Python 3、足以存放 22.1 GB GGUF 與建置檔的磁碟空間，以及可提供實用解碼速度的本機高速 SSD/NVMe。已測試路徑使用 llama.cpp commit 836d57176dc699a726c55418e4f96b8ca628e1bf。 在其他位置執行時，此腳本會取得本 repo…