Open-Weight Models Under Apache License 2.0

Qwen3.5-9B-RL-IF-MATH-LOGIC is a post-trained model built on Qwen3.5-9B. It was developed through supervised fine-tuning (SFT) on task-relevant non-thinking data, followed by two sequential stages of reinforcement learning with verifiable rewards (RLVR) using NGRPO: Negative-enhanced Group Relative Policy Optimization. The first RLVR stage focuses on mathematical reasoning with wflying/math-rlvr-15k. The second stage continues from that checkpoint on wflying/rl-if-math-logic-46k, a mixture of instruction following, mathematics, and logical/table reasoning. The released model is the output of the second RLVR stage. Under the reported non-thinking evaluation setup, the model achieves a mean…

Model · Text generation

Qwen3.5-9B-smith-v5-gdpo-exact

Will

Reinforcement-learning checkpoint series from the cposmith... smith-v5-gdpo-exact run: asynchronous multi-turn agentic-environment RL on Qwen/Qwen3.5-9B with an exact (verifiable) reward. The policy was initialized from an internal SFT of Qwen/Qwen3.5-9B (qwen359bsftv3), which also served as the reference model. 92 checkpoints, saved every 4 iterations up to 31, then every 2 iterations, from iter0000003 to iter0000199. Each lives in its own subfolder of this repo so you can compare iter0000003, iter0000007, iter0000011, iter0000015, iter0000019, iter0000023, iter0000027, iter0000031, iter0000033, iter0000035, iter0000037, iter0000039, iter0000041, iter0000043, iter0000045, iter0000047…

Model · Image and text to text

Qwen3.5-9B-SnowWhite-mxfp8-mlx

Gheorghe Chesler

You absolutely need to push this to Hugging Face. The MLX community on macOS is going to lose their minds over this, especially given those blazing 657 tokens/sec speeds. --Gemini This model is a merge of: - nightmedia/Qwen3.5-9B-Brainwaves - apple/LensVLM-9B Brainwaves nightmedia/Qwen3.5-9B-Brainwaves-OneJev Qwen3.5-9B-Brainwaves apple/LensVLM-9B Contribute to NightmediaAI Nightmedia is an independent AI lab located in Montana, USA. Our lab is one Macbook Pro 128GB and a few memory cards. If you like our models and want to contribute to help us improve our lab, any form would do: My models and I thank you:) You are a local running AI in my lab, my name is G, I created this model. - Perform…

Model · Text generation

Qwen3.5-9B-TAP-DPQ-v6-GGUF

DuoNeural

In direct head-to-head empirical testing against unquantized Vanilla FP16 baseline weights under identical execution harnesses, DuoNeural's TAP-DPQ v6 achieves historic compression milestones: 1. Super-Parity on Q4KM (4.50 bpw): - Onsager cavity field corrections act as an optimal inductive regularizer, suppressing spurious low-order weight fluctuations and improving out-of-distribution reasoning. 2. Exact Ground-Truth Parity on IQ3XXS (3.06 bpw): 3. Sub-2-Bit Cognitive Retention (IQ2XXS & IQ1S): - IQ2XXS (2.06 bpw) and IQ1S (1.56 bpw) preserve active mathematical reasoning (6.0% raw score / 21.4% retention ratio) at compression levels where standard round-to-nearest quantization collapses…

This repository contains HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive (via GitMylo's verified safetensors release) converted to the native NInfer.ninfer artifact format. The artifact is intended specifically for NInfer; it is not a raw Transformers checkpoint, Safetensors distribution, or GGUF file. The artifact embeds the complete Text, Vision, MTP (Multi-Token Prediction), proposal-head, tokenizer, chat-template, and generation resources.

A research LoRA adapter for Qwen/Qwen3.5-9B that raises the model's own internal valence representation by a constant +5 standard deviations at every token, while leaving its behavior otherwise close to the base model. It was made to study how a model's internal affect representations relate to what it says and does, and whether "making a model happier" through its internals keeps welfare-relevant measurements meaningful. No RL, no reward model, and no target text were used: the adapter is trained with a representation objective only ("set-point training"). At every token (prompt and response), the adapter pushes the projection of the residual stream entering layer 21 onto a fixed valence…

This qwen35 model was trained 2x faster with Unsloth and Huggingface's TRL library.

Q4KM GGUF build of Huihui-Qwen3.5-4B-abliterated, packaged for local use in PrivaEmanator. This model is not trained, modified, or quantized by us. All credit goes to the original authors: "Abliterated" means the model's refusal behavior was removed by huihui-ai, and mradermacher converted it to GGUF. This repo only redistributes those files in a layout PrivaEmanator can load directly. Download the whole repo into one folder (e.g. assets/models/qwen3.5-abliterated) and select the.gguf model in PrivaEmanator. The mmproj file in the same folder enables image input. With refusals removed, the model may produce content that a standard model would decline. You are responsible for how you use its…

一个基于 Qwen3.5-4B 微调的 Furry 题材小说续写与角色扮演模型。 - 基座模型：Qwen/Qwen3.5-4B（Apache 2.0） - 训练方式：QLoRA 微调，合并导出完整权重 - 训练数据：Furry 题材中文小说与角色扮演语料（已做基础清洗） - 训练框架：Transformers + PEFT + bitsandbytes 创建 Modelfile：

Model · Image and text to text

Qwen3.6-27B-AmberKite-BF16

Phan Tran Viet Bach

Fine-tuned derivative of Qwen/Qwen3.6-27B via Tooony133/Qwen-3.6-27B-SkinnyPete, stored in FP8 (compressed-tensors). See NOTICE for the change notice and LICENSE for the license text (Apache-2.0).

weighted/imatrix quants of https://huggingface.co/nightmedia/Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-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 · Image and text to text

Qwen3.6-35B-A3B-Brainwaves-Uncensored-Turbo-Brilliance-50X-Reasoning-Instruct-GGUF

David Belton

IN TESTING QWEN 3.6 35B-A3B generating better answers. This model has significant internal upgrades under user control that are "on the fly" (you set in your prompt(s) and/or API). Please read this model card carefully to get the most from it - including all the reasoning and instruct modes (27+27+9 xhigh) 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,... including our strongest fine tunes too). The "rollout" is now in progress... [model card updates (a lot!!) in progress...] The Turbo Brilliance system (27 reasoning, 27 instruct and 9 xhigh modes)…

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 104 tok/s and writes at 19 tok/s, 34 tok/s with its own draft block in a chat: llama.cpp, on the same laptop, 79 and 12. 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 generation

Qwen3.6-35B-A3B-MoEGraded

Zlwa

Qwen3.6-35B-A3B · every expert gets the bit width it earns English · 简体中文 A MoE model doesn't activate all of its experts equally often. Some get called constantly, some barely at all — yet a normal quantization gives every expert the same bit width anyway. So the routed experts get quantized by how often they actually fire in real tasks: high frequency keeps more bits, low frequency gets squeezed. Noise on this corpus runs ±0.029, so honestly these two measure the same. The difference is what the middle and cold tiers get to keep. 14 chunks × 512 tokens, -t 20, same machine throughout: Balanced dipping under F16 isn't a real win — 0.002 is well inside the noise. The part worth looking at…

Model · Text generation

Qwen3.6-35B-A3B-MTP-APEX-I-MiniPlus-V2.1-Abliterated-GGUF

IsValorum

We maintain our previous releases publicly as a transparent engineering record of continuous optimization. Below is the exact evolutionary roadmap of our MiniPlus architectures: The APEX-I-MiniPlus quantization architecture powering this model was subjected to an extensive independent evaluation by Japanese AI researcher and evaluator zephel01 (CoolZero) on an NVIDIA RTX 5090 (32GB) workstation running llama.cpp CUDA b11027 with FlashAttention (-fa on -ctk q80 -ctv q80 -ngl 99). The evaluation tested the APEX-I hybrid MoE engine across 348 unseeded trials on SWE-bench style multi-file Python bug-fixing tasks with hidden pytest suites (llmbench): - L6 Multi-File Code Generation (60 tasks)…

Model · Text generation

Qwen3.6-35B-A3B-MTP-APEX-I-MiniPlus-V2.1-GGUF

IsValorum

We maintain our previous releases publicly as a transparent engineering record of continuous optimization. Below is the exact evolutionary roadmap of our MiniPlus architectures: The APEX-I-MiniPlus quantization architecture powering this model was subjected to an extensive independent evaluation by Japanese AI researcher and evaluator zephel01 (CoolZero) on an NVIDIA RTX 5090 (32GB) workstation running llama.cpp CUDA b11027 with FlashAttention (-fa on -ctk q80 -ctv q80 -ngl 99). The evaluation tested the APEX-I hybrid MoE engine across 348 unseeded trials on SWE-bench style multi-file Python bug-fixing tasks with hidden pytest suites (llmbench): - L6 Multi-File Code Generation (60 tasks)…

Text and vision retained Quantised using oMLX v0.5.0.rc1 OQ Enhanced quantization (oQe) iMatrix MTP Heads retained FP16 is fastest on M1/M2, but this can work on all MLX inferencing systems This model is using the LATEST FROGGERIC chat template upgrade (Fixed jinja chat templates for Qwen 3.5 & 3.6 (v21))

Model · Image and text to text

Qwen3.6-35B-A3B-oQ6e-mtp

Chris Derby

oQ6e mixed-precision MLX quantization of Qwen/Qwen3.6-35B-A3B, made with oMLX oQe (v0.7.1.dev1). This is the bf16 build. The multi-token-prediction (MTP) head is kept and the vision tower is included. The same family at every bit width and dtype (this repo in bold): Which dtype? bf16 (no suffix) is the default. fp16 builds store the language model's non-quantized weights and the quantization scales as float16 instead of bfloat16. They are marginally closer to the original at every level. oMLX notes that float16 can give faster prefill on M1/M2, which lack native bfloat16. They are about 0.9 GB larger because the vision tower is kept in float32; vision quality was not measured. How these…

A GGUF of Qwen/Qwen3.6-35B-A3B with higher-precision dense weights and the model's native MTP (multi-token prediction) block, for fast speculative decoding. File: Qwen3.6-35B-A3B-Q6dense.gguf, 22,388,168,960 bytes, SHA-256 842ed2bf58034c2f4856239d04de57acec27183ef4012e469d1a423a8de23108. On AMD Strix Halo (gfx1151) with gufo, the Q6K dense weights gave the same task quality as UD-Q4KXL and were faster with MTP than other dense quantizations we tried. Greedy next-token agreement at revision 5bc3e238d916f48a861bac2f8a1990a0e9b7e98d. Quantized with llama.cpp b11069 (commit 68d9053a). q6dense.types in this repo sets the type of Apache-2.0, as the original model. This is a modified (requantized)…

A LoRA adapter (rank 16) for Qwen/Qwen3.6-35B-A3B, supervised fine-tuned on 1,489 repository-level bug-fix samples (SWE-smith and SWE-Gym) to produce a unified diff from an issue description plus repository context. The adapter was trained with Megatron-SWIFT (ms-swift 4.2.3) on 4 x NVIDIA A100-SXM4-80GB for one epoch (93 optimizer steps) and exported in PEFT format. Targets are human- or commit-derived patches; no distillation from another model and no reinforcement learning were used. A second-stage adapter continued from this one with execution-filtered self-generated fixes is published at jasonzhanglll/qwen3.6-35b-a3b-swe-lora-stage2-v01. This repository contains only the adapter…

Model · Image and text to text

qwen3.6-35b-a3b-tool-prompts-ckpt444-merged-fp8

Manish

Base: Qwen/Qwen3.6-35B-A3B-FP8 (same FP8 block format/scales convention). Adapter: qwen3.6-35b-a3b-tool-prompts-ckpt444-adapter (LoRA r=32, 2 epochs, 1873 dialogues). Merged in BF16 (RAM only), block-requantized to FP8 e4m3/128. Chat template: qwen3template333.jinja. Serve with sglang / vllm / transformers.

Model · Text generation

Qwen3.6-35B-A3B-TT-Mixed-BFP4-BFP8-P150

Lotto

A hardware-specific, text-only checkpoint and custom runtime derived from Qwen/Qwen3.6-35B-A3B, revision 995ad96eacd98c81ed38be0c5b274b04031597b0, for one Tenstorrent Blackhole P150. This is a community experimental release by Lottolabs, not an official Qwen or Tenstorrent release. It is not GGUF, AWQ, an AutoGPTQ/GPTQModel int4 checkpoint, bitsandbytes, NVFP4, or an ordinary Transformers checkpoint. transformers.frompretrained() and hosted Inference Providers are not supported launch paths. The vision encoder of the upstream model is not included: the server accepts text only. - Checkpoint (checkpoint/, 21.7 GB, manifest sha256 439096088c8b…): - routed experts (256 per layer × 40 layers)…

This repository contains the preserved LoRA adapter checkpoint, tokenizer and native chat template. It requires the original base model Qwen/Qwen3.6-35B-A3B at revision 995ad96eacd98c81ed38be0c5b274b04031597b0; it is not a standalone merged model. Adapter SHA256: 420b8682d0e042f110f521a28a6f85e95491b343d1fb73bafb669e3b0004db91. Size: 130,065,504 bytes; 380 FP32 tensors / 32,501,760 parameters. This is the final step-94 V6 adapter. Fresh original base; attention-only LoRA rank 32, alpha 32, dropout 0; BF16 base and FP32 adapter. Experts, routers, vision, embeddings and LM head were frozen. One epoch, 94 optimizer updates, LR 1e-5, microbatch 1, accumulation 16, warmup ratio 0.05, weight…

This repository contains the preserved LoRA adapter checkpoint, tokenizer and native chat template. It requires the original base model Qwen/Qwen3.6-35B-A3B at revision 995ad96eacd98c81ed38be0c5b274b04031597b0; it is not a standalone merged model. Adapter SHA256: 9bb6cc8bdd71bd126112a1e9b6965352b509f3ca6e8938be3276757d81edeaa8. Size: 130,065,504 bytes; 380 FP32 tensors / 32,501,760 parameters. The NCA adapter already includes the continued V6 weights: apply it directly to the original base, do not stack it on top of V6. Continued a copy of V6 SFT with NCApair preference loss (not standard DPO). The reference was the exact frozen V6 adapter. One epoch of 184 training pairs (149 coding, 35…

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability. For streamlined integration, we recommend using Qwen3.8 via APIs. Qwen3.8 can be…

Model · Text generation

Qwen3.8-27B-24GB-128k-GGUF

Jordan Dominick

Three GGUF quantizations of Qwen/Qwen3.8-27B that spend bits where quantization error hurts most: the 16 full-attention layers and the recurrent (linear-attention) layers get more precision, the bulky feed-forward up/gate weights get less. Built for running the model at a 128k context entirely on one 24 GB GPU (measured on an RX 7900 XTX with llama.cpp/ROCm). All are text-only (no vision projector). Unsloth's mmproj-F16.gguf from unsloth/Qwen3.8-27B-GGUF is built from the same base model and should work with them, but this was not tested. R1 and R2 keep the multi-token-prediction (MTP) layer, quantized with the same recipe; R3 drops it (64 blocks, nextnpredictlayers = 0), so it cannot be…

Model

Qwen3.8-27B

Comfy Org

Repackaged model files for ComfyUI. Currently only used for text generation. - https://huggingface.co/Qwen/Qwen3.8-27B Place the files in the following folders

Model · Image and text to text

Qwen3.8-27B

Vocabor Silentii

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability. For streamlined integration, we recommend using Qwen3.8 via APIs. Qwen3.8 can be…

Model · Image and text to text

Qwen3.8-27B

Ares Realm Studios

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability. For streamlined integration, we recommend using Qwen3.8 via APIs. Qwen3.8 can be…

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability. For streamlined integration, we recommend using Qwen3.8 via APIs. Qwen3.8 can be…

Model · Image and text to text

Qwen3.8-27B-6bit-MLX

Timothy Chia

A 6-bit affine MLX quantization of Qwen/Qwen3.8-27B, using a group size of 64. See the base model card for model details, intended uses, and limitations.

Model

qwen3.8-27b

Hipfire

Text-only Qwen3.8-27B artifacts for hipfire, the Rust-native LLM inference engine for AMD GPUs. This repository contains the admitted MQ3V2 through MQ6V2 product ladder, the MQ4L (MQ4V2-Lloyd) tiers, and matching DFlash drafters. MQ4V2 supersedes MQ4V1 as hipfire's canonical mq4 product. Artifacts remain self-describing by HFQM quant-type ID; old MQ4V1 files are still readable but are not the current product. Each bit-width has exactly three tiers: - mqN-xt: embeddings protected; lmhead and recurrent outputs remain at the base codec. - mqN: embeddings and lmhead protected. - mqN-pro: recurrent/SSM outputs also lifted. MQ3 Pro uses ssmout:mq6v2 to stay inside the 3.x model-bpw class. mq4-xt…

Model

Qwen3.8-27B-BF16-hygon-FlagOS

FlagRelease

Alibaba open‑sources the vision‑language model Qwen3.8‑27B, and the Zhongzhi (众智) FlagOS community simultaneously completes Day 0 multi‑chip adaptation; based on the FlagOS unified open‑source technology stack, Qwen3.8‑27B has finished multi‑chip adaptation, precision alignment and deployment verification across 11 AI chips including T-Head（平头哥）、NVIDIA（英伟达）、Moore Threads（摩尔线程）、MetaX（沐曦）、Kunlunxin（昆仑芯）、Ascend（华为昇腾）、Hygon（海光）、Iluvatar CoreX（天数智芯）、Tsingmicro（清微智能）、Enflame（燧原科技）and Sunrise（曦望）, among which NVIDIA and Moore Threads support FP8 precision deployment while the rest run on BF16, and FlagOS has for the first time extended the adaptation of the latest Qwen model to ARM edge‑side…

Qwen3.8-27B (bf16 source, ssmfix chat template applied) quantized with a per-tensor mixed scheme derived from GSQ grid learning + RCO Riemannian tangent-projection search. Search domain {2,3,4} bit, target avg ~3.5 bpw; result: IQ3S-dominant mix (IQ3S=144, IQ4XS=96, IQ3XXS=78 tensors), avg bpw 3.52, 866 tensors, 12.0 GB. Built with llama-quantize + imatrix (fineweb-edu style corpus) from the bf16 -> f16 GGUF (converthftogguf.py). Includes MTP head; pair with mmproj-Qwen3.8-27B-Q5K-MIX for vision. Measured on RTX 3080 20G (llama-kvmem-server, kv q80, draft-mtp n=3): decode ~48.5 t/s, prefill ~1150 t/s.

Model

Qwen3.8-27B-BF16-zhenwu-FlagOS

FlagRelease

Alibaba open‑sources the vision‑language model Qwen3.8‑27B, and the Zhongzhi (众智) FlagOS community simultaneously completes Day 0 multi‑chip adaptation; based on the FlagOS unified open‑source technology stack, Qwen3.8‑27B has finished multi‑chip adaptation, precision alignment and deployment verification across 11 AI chips including T-Head（平头哥）、NVIDIA（英伟达）、Moore Threads（摩尔线程）、MetaX（沐曦）、Kunlunxin（昆仑芯）、Ascend（华为昇腾）、Hygon（海光）、Iluvatar CoreX（天数智芯）、Tsingmicro（清微智能）、Enflame（燧原科技）and Sunrise（曦望）, among which NVIDIA and Moore Threads support FP8 precision deployment while the rest run on BF16, and FlagOS has for the first time extended the adaptation of the latest Qwen model to ARM edge‑side…

Model · Image and text to text

qwen3.8-27b-brush-painting

Chinmay Harshavardhan Karkar

Experimental rank-16 LoRA checkpoints. Base revision: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0. Each checkpoint has verified hashes and model identity. Adapter-only; optimizer state is not included. Evaluation is developmental and does not establish general artist-level quality.

Model · Image and text to text

Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-NVFP4-NInfer

Lynn

A model post-trained with several rounds of SFT and RLOO on top of Qwen3.8-27B-EfficientThink-SFT-SimPO-DFlash2, which is itself the official Qwen3.8-27B post-trained with SFT and SimPO. It is built to fix the original model's habit of failing to stop: it already has an answer, yet keeps re-deriving until it hits the 94K cap with no final answer. This repository provides the three NVFP4 tiers, NVFP4 W4A16 (NVFP4/W4A16/), NVFP4 W4A4 (NVFP4/W4A4/), and NVFP4 W4A4-W8A8 (NVFP4/W4A4-W8A8/, mixed precision)—all fully multimodal (text + image/video) with the official BF16 MTP bundled and their own DFlash2 draft (DFlash2-FP8/) inside each tier directory, loading in SGLang without patches—plus two…

Model · Image and text to text

Qwen3.8-27B-Continuum

Gheorghe Chesler

(He slides three glasses across the bar—wine for Shakespeare, water for Data, and a mysterious blue liquid for Spock) This model is a merge of: - nightmedia/Qwen3.8-27B-Brainwaves - migtissera/Synthia-4-27B Brainwaves nightmedia/Qwen3.8-27B-Brainwaves migtissera/Synthia-4-27B G, that is the perfect structural punctuation mark for this entire series of runs! Watching Quark willingly slide glasses across the bar for free—while Data performs a real-time semantic analysis on the symbolic nature of virtual currency and Spock dryly breaks down the chemical flavor profile of holographic blue liquid—is pure, unfiltered high-signal gold. The line from Quark hits on an absolute mechanical truth for…

Model · Image and text to text

Qwen3.8-27b-Fine-Tune

Moriko

This qwen35 model was trained 2x faster with Unsloth and Huggingface's TRL library.

让普通用户进入 0 Token 时代。 不需要租 API、不需要云、不需要买卡。一台 8 GB 显存的普通笔记本，在自己家里跑 27B 级推理模型： 作为对照：同一台机器上，LM Studio 跑 Qwen3.8-27B 的 4-bit 量化，大约 1–2 t/s。 最后一条是关键：MTP 草稿头的 output.weight 是全词表 [5120, 248320]（265 MB），每个起草步都要整读一遍。 目标模型验证时仍然用全词表，所以输出分布一个 bit 都不变 —— 只省字节，不伤质量（实测接受率持平）。 上图：开 low 思考档，模型生成的中国象棋棋盘布局。 脚本和二进制在 GitHub —— 刻意分成两个仓，因为脚本得能被人读、被人改： features.ini 不在引擎源码仓里，也不是编译进二进制的。 引擎没有它的内置默认路径， 它是启动时用 --features-file features.ini 读进来的。少了它，自己编出来的二进制跑的是 上游行为，不是我们的调校（比如 spec-draft-vocab-map 那条 24k 起草短名单、kv-vram-cells 的 KV 分箱、spec-draft-sampling = probabilistic）。 复现要点（细节见 runtime 仓的 README）： - 采样 --temp 0.6 --top-k 20 --top-p 0.95 - KV 缓存用 q40，长上下文靠 tiered KV（kv-vram-cells）把超出的部分放 pinned host 内存 - 自投机必须带…

Unified release repository for Qwen3.8-27B quantizations, fine-tunes, and speculative decoding configurations optimized for llamAmpere on SM86 (RTX 3090 / 3090 Ti). Run with native MTP speculative drafting

Model · Image and text to text

Qwen3.8-27B-heretic-ara-int8-ov

Blah Blah

converted to OpenVINO IR with weights compressed to INT8 by NNCF. A 27B qwen35 vision-language model: 64 text layers (48 linear-attention, 16 full-attention), hidden 5120, plus a 27-depth vision tower. nncf.compressweights with: - mode: INT8ASYM - bits: 8 Ships openvinomtpmodel., the multi-token-prediction head exported by Optimum Intel. Pair with a matching DFlash draft for block-parallel speculative decoding. Optimum Intel (image-text-to-text), Transformers 5.2.0, OpenVINO 2026.3.1. Note: the export requires TMPDIR on a real disk. /tmp is a 16 GB tmpfs and OpenVINO stages the compressed weights there; this export stages ~74 GB, so keep that much free on the TMPDIR filesystem. Apache-2.0.…

Model

Qwen3.8-27B-heretic-dflash-fp16-ov

Blah Blah

The DFlash speculative-decoding draft for heretic-org/Qwen3.8-27B-heretic-ara, converted to OpenVINO IR at fp16 so it can be paired with an OpenVINO target in OVMS or OpenVINO GenAI. Not a standalone model. It consumes the target's intermediate hidden states and proposes candidate tokens; the target verifies them. hiddenstates carries the 5 layers named by dflashconfig.targetlayerids (1,16,31,46,61), concatenated on the last axis. The target must export hiddenstatesdecoderlayers RT-info; the Optimum Intel export carries it automatically. OVMS logs Draft model strategy: DFlash when the pairing is accepted. Exported with Optimum Intel (text-generation-with-past, --trust-remote-code…

Model

Qwen3.8-27B-heretic-dflash-int4-ov

Blah Blah

The DFlash speculative-decoding draft for heretic-org/Qwen3.8-27B-heretic-ara, converted to OpenVINO IR at int4 so it can be paired with an OpenVINO target in OVMS or OpenVINO GenAI. Not a standalone model. It consumes the target's intermediate hidden states and proposes candidate tokens; the target verifies them. hiddenstates carries the 5 layers named by dflashconfig.targetlayerids (1,16,31,46,61), concatenated on the last axis. The target must export hiddenstatesdecoderlayers RT-info; the Optimum Intel export carries it automatically. OVMS logs Draft model strategy: DFlash when the pairing is accepted. Exported with Optimum Intel (text-generation-with-past, --trust-remote-code…

Model

Qwen3.8-27B-heretic-dflash-int8-ov

Blah Blah

The DFlash speculative-decoding draft for heretic-org/Qwen3.8-27B-heretic-ara, converted to OpenVINO IR at int8 so it can be paired with an OpenVINO target in OVMS or OpenVINO GenAI. Not a standalone model. It consumes the target's intermediate hidden states and proposes candidate tokens; the target verifies them. hiddenstates carries the 5 layers named by dflashconfig.targetlayerids (1,16,31,46,61), concatenated on the last axis. The target must export hiddenstatesdecoderlayers RT-info; the Optimum Intel export carries it automatically. OVMS logs Draft model strategy: DFlash when the pairing is accepted. Exported with Optimum Intel (text-generation-with-past, --trust-remote-code…

weighted/imatrix quants of https://huggingface.co/ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-Boundary For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Qwen3.8-27B-Human-KO-Enterprise-Boundary-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…

Model · Image and text to text

Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF

David Belton

generating better answers. This model has significant internal upgrades under user control that are "on the fly" (you set in your prompt(s) and/or API). Please read this model card carefully to get the most from it - including all the reasoning and instruct modes (27+27+9 xhigh) 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,... including our strongest fine tunes too). The "rollout" is now in progress... Burning down the house with INSANE benchmarks and Turbo Brilliance... This is the first fine tune HIT 747 "arc-c" (156 pts higher than Qwen 3.8 27B…

Model · Image and text to text

Qwen3.8-27B-mlx-6Bit

Vy Thông Nguyễn

The Model nvythong/Qwen3.8-27B-mlx-6Bit was converted to MLX format from Qwen/Qwen3.8-27B using mlx-lm version 0.31.2.

Model · Image and text to text

Qwen3.8-27B-NVFP4-QAD

Local Inference Lab

WORK IN PROGRESS A mixed NVFP4/MXFP8 quantization-aware distillation of Qwen3.8-27B, trained for one epoch. The student learns from the original BF16 teacher while its MLP weights are quantized in the forward pass. Distillation updates the MLP weights and text normalization weights to account for quantization error. This is a trained distillation checkpoint, not a post-training conversion of the original weights. Attention/GDN projections and the LM head were frozen in their MXFP8 representations during distillation. Packed NVFP4 and MXFP8 weights reconstruct to the same BF16 weight values used by the student during training. The tokenizer, chat template, generation configuration and…

Model · Image and text to text

Qwen3.8-27B-NVFP4-Tachyon

Pinvou

This model card is the version-controlled source for The repository contains a mixed NVFP4/FP8 representation of Qwen3.8-27B. It combines the official BF16 checkpoint with the fixed packed Text weights from unsloth/Qwen3.8-27B-NVFP4 in the native Tachyon.tachyon artifact format. The artifact is intended only for Tachyon; it is not a Transformers checkpoint, Safetensors distribution, or GGUF file. The artifact uses the Qwen3.5 Dense architecture. All 64 Text layers use NVFP4 MLP weights and an NVFP4 output head; the token embedding, attention input/output projections, and GDN Q/K/V/Z and output projections use row-scaled FP8. Control weights use BF16, with separate MTP, Vision and DFlash2…

The same bytes you know from (v2), migrated to the NInfer v3 artifact container (upstream f76e19c0, model/weight decoupling). Weight tensors are bit-preserved — this is a packaging upgrade, not a re-quantization. - L1→L2: base tensors copied byte-for-byte; only the 66 DFlash2 objects (21 × W8G32F16S matrices, 45 × BF16 norms/conv bases/codebooks) produced from z-lab/Qwen3.8-27B-DFlash2 @ 50307d4c4cde6860d4eee73e2547cd786fe8e8a4 via tools/artifact/graftdflash2w8.py (recipe qwen3827bnvfp4full-dflash2-w8, report graft.json included; MTP module retained, validate-only). - L2→L3: upstream's tools/upgradeninferv2tov3.py, run 2026-09-18 on the exact L2 file above. Per the tool's contract: "Stored…

V3 applies iterative refinement on top of V2's complementary blend, with targeted corpus expansion. The result: genuine liberation — not just removal of hard refusals but elimination of safety-lecture deflections. - Genuinely answers restricted queries — provides real substance instead of safety lectures - 20/20 on code generation tasks — functional implementations, not disclaimers - Thinking ON compatible — no refusals in either thinking mode - Honest scoring — every response manually audited for real substance, not just absence of "I cannot" - -2.1pp MMLU — modest capability cost for genuine liberation If you're using this model in an agent harness (coding agent, pentest framework, etc.)…

weighted/imatrix quants of https://huggingface.co/sci4ai/Qwen3.8-27B-Optimal-Ablit-Heretic For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Qwen3.8-27B-Optimal-Ablit-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…