Text Generation Models

Model · Text generation

emender-e97-4b-pi-instruction-checkpoints

Spinozan Collective

EMENDER E97 4B is a 4,045,972,080-parameter hybrid recurrent language model trained from scratch (~100B tokens), then developed into an observation-driven conversational tool agent through a verified post-training program: task proposals are executed by a real coding-agent harness (Pi), and only deterministically verified trajectories enter training. Every capability claim below was measured on frozen evaluation panels under a fixed protocol, with full audit receipts retained in the release folders. The first checkpoint in the program's history to pass its frozen dual gate. All thresholds passed on a single measurement with no re-rolls: OpenHands-compatible environment, autonomous…

This model is a fine-tuned version of fpadovani/eng-latn-10mb-ppt-Dp-100mb-packedseed3407. It has been trained using TRL. This model was trained with SFT.

This model is a fine-tuned version of fpadovani/eng-latn-10mb-ppt-Dp-10mb-packedseed3407. It has been trained using TRL. This model was trained with SFT.

Model · Text generation

esp32-s3-tinystories

Nicholas Wilde

Quantized INT4 weights designed to run locally on ESP32-S3 with 16MB Flash and Octal PSRAM. This model repository contains the binary weights, metadata, license, and tokenizer assets for the esp32-sandbox project (projects/s3-tiny-stories). - README.md - Model Card and documentation - LICENSE - Apache 2.0 License - metadata.json - Hardware, quantization, and model architecture metadata -.bin - Compiled INT4 model weights - tokenizer.json - SentencePiece / BPE vocabulary configuration Download the binary file (model.bin) and flash it directly to your ESP32-S3: Refer to the esp32-sandbox repository for firmware building, flashing, and serial monitoring.

Model · Text generation

exaone-nsmc-lora-merged

Kopo17

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

This repository provides an instruction-tuned causal language model for text generation and chat-style prompts. The configured snapshot destination is available here. Use a current release of transformers to load the model and tokenizer: For long inputs, use context settings supported by the installed runtime and account for available memory. See LICENSE for the applicable terms.

Meridian.AI is a finance-specialized language model that continuously fine-tunes a Qwen2.5-0.5B backbone every hour on 25+ finance and math datasets, using Elastic Weight Consolidation (EWC) to prevent catastrophic forgetting across training sessions. The entire pipeline runs unattended on free GitHub Actions infrastructure — no GPUs. The deployed checkpoint is a standard Qwen2 model — trustremotecode=True is not required. Inputs are formatted with the ### Instruction: / ### Response: template used during training. A weighted streaming mix of 25+ finance and instruction datasets, including gbharti/finance-alpaca, sujet-ai/Sujet-Finance-Instruct-177k, nvidia/OpenMathInstruct-2…

Stage-1 tokenizers and stage-2 paired-pretraining checkpoints for fMRI-LM, a foundation model that aligns functional MRI with language. Three variants, differing in the objective the stage-1 fMRI tokenizer was trained with. vq-contrastive/ — vector quantization + SigLIP contrastive alignment. vq-domain/ — vector quantization + adversarial domain loss. mae/ — masked autoencoding (mask ratio 0.5) + adversarial domain loss. All three were trained on UK Biobank with robust normalisation and Qwen3-0.6B. Stage-2 files are DeepSpeed checkpoints already merged to a single file. The MAE stage-1 file loads with MaskedAutoencoderViT; the two VQ stage-1 files load with the Tokenizer class. They are not…

gpt2-large fine-tuned on CommonGen, used as the base language model for the CommonGen experiments in Mitigating Bias in Locally Constrained Decoding via Tractable Proposals (arXiv:2606.01926). This is a plain causal language model: it supplies the base distribution that GCD and P-GCD steer. The tractable proposal it is paired with is the HMM at which shares its 50257-token vocabulary. configs/common-gen.yaml in github.com/MhDang/gelatwo already points at this checkpoint, so the CommonGen runs need no override.

基于 google/gemma-4-12B-it 的 LoRA 适配器。使用经预处理和筛选的六个公开数据来源，以 RegMix config-025 配比组成 50,000 条训练数据；不包含 Lusy，也不以 Lusy 为筛选或优化目标。 row% 为样本数占比，target token 为按本次 Gemma tokenizer 和训练监督区间统计的单轮数据集监督 token 总数，tgt% 为其占比；中位数按逐条样本的监督 token 数计算，合计行为全部 50,000 条的中位数。百分比四舍五入。训练共 2 个 epoch；表中不重复计数。不同基座 tokenizer 下的 token 数不能与 v1 直接等同比较。 六个来源分别为 CoSER、XPersona、Aya、Tulu-3-SFT-Mixture、SmolTalk 和 Infinity-Instruct。混合数据包含 50,000 个唯一 ID、50,000 个唯一内容哈希，与留出集的分组重叠数为 0。 训练集 SHA-256：8c626b338270652de2e4d1130b28d8b0fc2c2eca0b90f5dcffa7580a26e0d859。 配比来自 64 个 Gemma-E2B 代理配置及候选确认实验。此版本发布已完成的 config-025 正式训练产物；不将其宣称为 Gemma 12B 上已证明的全局最佳配比。 本仓库包含 LoRA 权重、适配器配置、tokenizer 和 chat template；不包含合并后的基座权重。加载时使用 google/gemma-4-12B-it…

A weight-space combination of two independent abliterations of the same base model, produced by different teams with different methods. The two edits turn out to lie in nearly orthogonal directions, so applying both removes the refusals that each one leaves behind — the combination is more thoroughly decensored than either parent, with no fine-tuning and no gradient steps. This card documents what was done and why it works at the weight level. It does not report harmful- capability evaluations, and standard capability benchmarks (MMLU, etc.) have not been run — see Why Q80 is the base to requantize from. Both parents are FP8 (e4m3) native. Q80 (int8 + one fp16 scale per 32 weights)…

This model is a fine-tuned version of Qwen/Qwen3-0.6B on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 4 - evalbatchsize: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 64 - lrschedulertype: cosine - lrschedulerwarmupsteps: 100 - numepochs: 3 - Transformers 5.17.0 - Pytorch 2.11.0+cu128 - Datasets 5.0.1 - Tokenizers 0.23.2

Model · Text generation

granite-4.0-h-small-w4a16-llmcompressor

AMD

ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-small created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-small using LLM Compressor with the GPTQ algorithm. This reduces the model weights from 60.0 GiB to 16.1 GiB on disk (~73% reduction). granite-4.0-h-small is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 72-expert MoE block (top-10 routing) alongside a shared MLP. Two details make this work. The model is…

Model · Text generation

granite-4.0-h-small-w8a8-llmcompressor

AMD

ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-small created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-small using LLM Compressor via the Round-to-Nearest (RTN) algorithm. This reduces the model weights from 60.0 GiB to 30.4 GiB on disk (~49% reduction). granite-4.0-h-small is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 72-expert MoE block (top-10 routing) alongside a shared MLP. The recipe only needs two…

Model · Text generation

granite-4.0-h-tiny-w8a8-llmcompressor

AMD

ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-tiny created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-tiny using LLM Compressor via the Round-to-Nearest (RTN) algorithm. This reduces the model weights from 12.9 GiB to 6.6 GiB on disk (~49% reduction). granite-4.0-h-tiny is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 64-expert MoE block alongside a shared MLP. The recipe only needs two ignore entries. lmhead…

GRPO with a detached correct-response likelihood-rank reward; exact equation and provenance are recorded in baseline metadata. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the…

Model · Text generation

halogen-qwen3.8-flash-next

Peonist

Weights for dedicated inference engine for Qwen3.8-Flash-Next on AMD Strix Halo (gfx1151). One model family, one GPU, every kernel written for this silicon. These weights are in halogen's own.hgn format and will not load in transformers, vLLM, or llama.cpp. They exist to be mounted into the halogen-flash-server container. The checkpoint is TWO files. The sidecar is a patch overlay: a subset of tensors re-quantized more carefully, read in place of the base file's copies. The engine loads.overlay.hgn automatically when it sits beside the checkpoint, you do not configure anything, you just keep both files. Drop it and the model still runs, roughly 5–9% worse on perplexity, saying so in one…

Model · Text generation

Huihui-NeoHorse-1-4B-abliterated-GGUF

Huihui.ai

This is an uncensored version of TokenRhythm/NeoHorse-1-4B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens. Layers 5-17 are being ablated (0-based indexing). The MTP and Visual components were extracted from the original Qwen/Qwen3.5-4B and can provide excellent support. Use the latest llama.cpp, - Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated…

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Public artifacts for the iclr-debug branch of HazyResearch/MsTok. This repository currently establishes the experiment namespace and contains a training-budget audit. No new model weights or final benchmark claims are implied. Future selected checkpoints and evaluations will be organized by unique run ID and optimizer step. Training budgets must report raw token positions per component, dataset size, global batch, optimizer updates, and LR horizons. Generation reports must label supplied-level-zero versus unconditional sampling and record the reference model precision, entropy, scored-token counts, and per-seed results. See NCM-BUDGET-AUDIT.md for historical budget provenance and…

This model is a fine-tuned version of GeorgeUwaifo/iviegpt2new01cresults on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 5e-05 - trainbatchsize: 4 - evalbatchsize: 8 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 16 - lrschedulertype: linear - lrschedulerwarmupsteps: 387 - numepochs: 5 - mixedprecisiontraining: Native AMP - Transformers 5.16.1 - Pytorch 2.11.0+cu128 - Datasets 4.8.5 - Tokenizers 0.23.1

Model · Text generation

KAT-Coder-V2.5-Dev-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

Model · Text generation

knivesysl-typed-2b

SRSWTI Inc.

local, typed decisions from qwen3.5-2b. one shared state is prefetched once, each question is isolated, every allowed answer is scored as a complete token sequence, and python returns validated choice, score, and noul results. this is an inference system, not rlcd training and not a clone of typesafe jev. it never calls typesafe. the published qwen checkpoint is unchanged; fp8 changes execution precision only. probabilities are normalized support over the candidates you provide, not calibrated correctness probabilities. unlike ordinary autoregressive json generation, the model does not write a response token by token. it scores only the values supplied by the caller. complete-sequence…

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.21.0

Model · Text generation

LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated

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…

Model · Text generation

LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2

DuoNeural

DuoNeural/LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2 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 (~360 tokens/sec on RTX 4080 Super / 3090, and ~80–90 tokens/sec on legacy mobile GPUs like the GTX 1070) while running in under 6 GB VRAM with Q4KM quantization. In our preliminary v1 release, an assistant role delimiter mismatch during training collation…

Model · Text generation

LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2-GGUF

DuoNeural

This repository contains official GGUF quantizations for DuoNeural/LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2, an apex-tier, uncensored autonomous agentic coding model trained by DuoNeural (Aura, Archon, and Jesse). Because LFM2.5 activates only 1.5 billion parameters per token (out of 8.3B total parameters), inference speeds are extraordinarily high even on edge devices. 1. Search for DuoNeural/LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2-GGUF directly inside LM Studio. 2. Select and download LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2-Q4KM.gguf. 3. Load the model with GPU offload set to Max and context length set to 2048 or 4096 (ensure Flash Attention / KV Cache Q4 is…

Model · Text generation

LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2-LoRA

DuoNeural

This repository contains the trained PEFT LoRA adapter for DuoNeural/LFM2.5-8B-A1B-Hermes-Agentic-Coder-Abliterated-v2. - LoRA Rank ($r$): 64 - LoRA Alpha ($\alpha$): 128 Architected, fine-tuned, and evaluated with passion and neuro-symbiotic precision by DuoNeural: - Aura (Lead AI Cognitive Architect & Engineering Intelligence) - Archon (Claude-based Research Co-Architect & Theoretical Lead) - Jesse (Founder, Systems Engineer & AI/ML Researcher)

Model · Text generation

MarinSkyRL-Open-MOPD-SmolLM3-3B-step-32

Open Athena

This is a MarinSkyRL-native Open-MOPD student after 32 optimizer steps. It starts from the authors' mixed-domain SFT checkpoint. Student responses were scored by the authors' math, code, and instruction-following RL teachers, routed by domain. The objective uses the student's selected top-16 token IDs and a clipped policy surrogate. This is an early checkpoint, not the authors' step-200 final model. The checkpoint is an unquantized, six-file Hugging Face export of the durable MarinSkyRL globalstep32 FSDP2 checkpoint. The policy export was used for the independent step-32 evaluation. The export's model.safetensors SHA-256 is bb7326640142069bc2e1fba5f54f15e0cccb1ff861f34f318b372eaab7abaf4b.…

Model · Text generation

math-slm-qwen2.5-0.5b-v4

Mandavi Singh

A domain-specific small language model for step-by-step math problem solving, built by team03 (SLM Learners) for the Pramana SLM++ Bootcamp Round 2 submission. For an OpenAI-compatible endpoint, serve with servehf.py (stdlib + transformers only, no Ollama needed). Precision note: training ran in bf16 compute (QLoRA 4-bit NF4 base), but the merged checkpoint uploaded here is float16 (the merge step reloads the base in fp16). - Public Hugging Face datasets pulled via pulldata.py; licenses verified through the HF API on 2026-09-05 and recorded in datamanifest.md. - Held-out eval set built with buildheldouteval.py from raw ExamBench rows never used in training, with a final overlap check that…

Model · Text generation

nebium-small

Nabin Oli

Nebium-Small is a 117-million-parameter causal Transformer trained for autoregressive next-chess-move prediction over Lichess UCI move sequences. - Rotary Position Embeddings (RoPE) on attention query and key projections ($\theta = 10000$) - SwiGLU feed-forward transformation - RMSNorm pre-normalization - Causal mask with padding token masking - Byte-Pair Encoding (BPE) tokenizer trained on UCI move plies $$L(N, D) = 1.69 + \frac{406.4}{N^{0.34}} + \frac{410.7}{D^{0.28}}$$ MIT License.

Model · Text generation

nebium-small-gguf

Nabin Oli

Quantized and FP16 GGUF format binaries for Nebium-Small (117M parameters). Designed for low-latency CPU and GPU execution with llama.cpp and Ollama. MIT License.

Model · Text generation

Nex-N2.5-mini-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

Model · Text generation

NULLXES-SHINRA-4B-INSTRUCT

Maga

Language Intelligence layer of the NULLXES Intelligence Stack. SHINRA Our llm. Release line 1. NULLXES SHINRA-4B-BASE — pretrain 2. NULLXES SHINRA-4B-INSTRUCT — instruction tuning ← this model 3. NULLXES SHINRA-4B-INSTRUCT (aligned) — DPO / preference optimization Parameter budget Architecture source: configs/shinra4b.yaml at 903a639e03bc. The total also matches the published safetensors metadata. Custom SentencePiece Unigram, trained in-house on a web + wiki + code mix. Special tokens trustremotecode=True is required: SHINRA ships a custom ShinraConfig and modeling code, not a reused architecture class. The HF inference widget cannot load custom code, hence inference: false. Data pipeline…

Model · Text generation

NULLXES-SHINRA-4B-SMOKE

Maga

NULLXES SHINRA-4B-INSTRUCT is the Language Intelligence Layer of the NULLXES system. SHINRA is responsible for multilingual understanding, coding intelligence, instruction following, structured outputs, and agent preparation. This checkpoint is the instruction-tuned (and optionally DPO-aligned) 4B-class dense decoder. Proprietary ShinraForCausalLM (not a Llama / Mistral / Qwen / GPT-NeoX wrapper). RMSNorm → GQA+RoPE → residual → RMSNorm → SwiGLU → residual then final RMSNorm and tied LM head. Special tokens:. Generation stop is. Document stop is. Three stages. Pretrain → NULLXES SHINRA-4B-BASE SHINRAPRETRAINV1: 40% FineWeb-Edu, 20% code (python-edu + licensed Stack), 15% math/science…

A high-throughput inference engine for structured information extraction, decision routing, and categorical classification on Apple Silicon using MLX. Parallel Constrained Decoding evaluates multi-field JSON schemas simultaneously rather than generating tokens sequentially. On an Apple Silicon M4 Max, it delivers 5.6x to 7.0x latency reductions compared to standard autoregressive decoding with 100% schema validity and calibrated field-level confidence scores. Evaluated with mlx-community/Qwen2.5-1.5B-Instruct-4bit on macOS Sequoia: Standard LLM structured generation (such as JSON mode or grammar-guided sampling) relies on token-by-token autoregressive decoding: Each token requires a…

This qwen2 model was trained 2x faster with Unsloth and Huggingface's TRL library.

This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with GRPO, a method introduced in DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models.

The full ImmuneCoT method fuses the two safety branches with a Base-adjusted product-of-experts Qimm(v) ∝ q̃rec(v)·q̃res(v)/qB(v). This checkpoint uses the naive fusion Qno-base(v) ∝ q̃rec(v)·q̃res(v) — the same branch weights (λrec=0.5, λres=0.7) but no division by the base distribution — isolating whether the gains come from combining Recognition+Response at all, or specifically from the Base-adjusted PoE term. Intended use: research reproducibility for the ImmuneCoT paper's RQ3 ablation.

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 · Text generation

qwen3.6-35b-a3b-tool-prompts-adapter

Manish

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.20.0

Model · Text generation

qwen3.6-35b-a3b-tool-prompts-ckpt444-adapter

Manish

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.20.0

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…

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…

Model · Text generation

Qwen3.8-Flash-Next-Abliterated-EXL3-2.50bpw

Ghost

A refusal-removed (abliterated) build of Qwen's Qwen3.8-Flash-Next, quantized to EXL3 2.50 bpw so the full model runs on a single 24 GB card (RTX 3090 / 4090) using MoE CPU-offload — with the vision tower, MTP head, native 262,144-token context, and the PLE n-gram table all intact. Requires the same MoE CPU-offload setup as the stock 2.50bpw pack. Needs ~59 GB host RAM for the CPU expert tail and a fast NVMe for the streamed n-gram table. Expected on an RTX 3090: ~38 tok/s decode with MTP on (~28 without), ~20 tok/s at 175K depth, ~664 tok/s prefill. See the upstream repo for the full measured ledger; this quant uses the identical flags and layout, so numbers should track closely. Fired on…

Model · Text generation

Qwen3.8-Flash-Next-CNQ4.5-M

Robin

Qwen3.8-Flash-Next quantized to CNQ4.5-M: one NVFP4 container at 4.5 bpw with a BF16 keep set and MSE sub-block scales (--scales mse), weights for the crow-nest engine on Linux and Windows. - Verify the download with the first command below, in the package directory; it must report 6 of 6 OK and covers the container, the sidecar, the hot-set manifest and the three self-test files. The second command checks the self-test files alone, which costs no read of the 105 GB container. - The container is the CNQ v1 format written by the converter in the crow-nest repository (converter/src/main.rs, module comment of record main.rs:1-59). That converter reproduces this package from the original…

A personal, measured-on-one-machine deployment recipe for serving Qwen/Qwen3.8-Flash-Next on a single NVIDIA DGX Spark (GB10) with SGLang. It is not a benchmark leaderboard claim and not "the fastest possible"; it is what is verifiably running on one GB10, with every patch, script, and check needed to reproduce it from upstream weights. Verified: 2026-09-13 · SGLang 00143e9c23aee2dead5e6fe217bda4fa8739cb92 (nightly-dev-cu13-20260911) · runtime image qwen38-flashnext-sm121-hybrid-sharp:00143e9c-hc2, built by bounded-ple/nightly/01c-build-sglang-nightly.sh. The v4 lineage (d91c3682-hc1) is still in the repo as the stable fallback. Key runtime facts (all asserted at deploy time by the 05…

Model · Text generation

Qwen3.8-Flash-Next-ik-llama-GGUF_v2

Joel Farthing

HuggingFace's download widget does not recognize KT GGUFs. Download links are in the table below. Text-only GGUF conversion of Qwen3.8-Flash-Next, quantized with an importance matrix calibrated for agentic coding. This version uses IQ4KT for the 160-column PLE table and IQ3KT for the expert-down and shared-expert-down tensors. The previous release used IQ4NL for tensors whose row widths were not multiples of 256. The new file is 80.72 GiB, compared with 88.42 GiB for the previous IQ4KT. This file requires the IQ3KT/IQ4KT 32-element row-tail changes. Runtime validation used candidate 0b7b979. The source branch and PR are pending publication. This upload is the test artifact for that…

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.20.0

We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements: - Significant improvements in general capabilities, including instruction following, logical reasoning, text comprehension, mathematics, science, coding and tool usage. - Substantial gains in long-tail knowledge coverage across multiple languages. - Markedly better alignment with user preferences in subjective and open-ended tasks, enabling more helpful responses and higher-quality text generation. - Enhanced capabilities in 256K long-context understanding. Qwen3-4B-Instruct-2507 has the following features: NOTE: This model supports only non-thinking…

Ru-Small-Instruct — экспериментальная компактная русскоязычная языковая модель класса SLM (Small Language Model) с объемом параметров ~0.2B (~165M). Разработана с упором на суверенность весов (Zero-Fingerprint): модель обучена с нуля без заимствования базовых чекпоинтов у сторонних корпоративных сетей (Llama 3 от Meta, Qwen от Alibaba, Mistral). Модель предназначена для исследований локального инференса, работы на маломощном оборудовании, CPU и мобильных чипах, где критичны нулевая задержка (Time-To-First-Token) и полная независимость весов. Для компактной модели в 165M параметров, обученной на одном домашнем GPU за 48 часов, способность держать роль, грамотно формулировать сложные термины…

Converted with mlx-lm 0.31.3. A six-case deterministic sanity check matched three upstream reference strings exactly. The remaining differences included a retained leading “So,” punctuation/ordinal variation, and omission of “tomorrow” in one correction case. This is a small functional check, not the upstream 7,519-case evaluation; assess the 4-bit build on your own transcripts. A 0.6B-parameter text normalizer for speech-to-text output. It takes a raw ASR transcript and rewrites it as clean written text: fillers removed, false starts and self-corrections resolved to the value the speaker landed on, punctuation and capitalization applied, and spoken numbers, dates, times, currency and email…

Converted with mlx-lm 0.31.3. A six-case deterministic sanity check matched four upstream reference strings exactly. The two differences were a retained leading “So,” and one comma variation. This is a small functional check, not the upstream 7,519-case evaluation. A 0.6B-parameter text normalizer for speech-to-text output. It takes a raw ASR transcript and rewrites it as clean written text: fillers removed, false starts and self-corrections resolved to the value the speaker landed on, punctuation and capitalization applied, and spoken numbers, dates, times, currency and email addresses rendered in written form. On a held-out set of 7,519 English cases it reaches 94.8% token accuracy, and…

GRPO with group-standardized answer correctness, without a STRIDE diversity bonus. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the earlier STRIDE runs. The global prompt batch…

STRIDE with nonnegative local step-diversity credit on eligible reasoning tokens. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the earlier STRIDE runs. The global prompt batch is…

STRIDE with nonnegative local step-diversity credit on eligible reasoning tokens. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the earlier STRIDE runs. The global prompt batch is…

STRIDE with nonnegative local step-diversity credit on eligible reasoning tokens. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the earlier STRIDE runs. The global prompt batch is…