Image and Text to Text Models

Model · Image and text to text

ACE-3-F-26B-A4B-260804

APMIC

ACE-3-F-26B-A4B-260804 is an enterprise-grade, production-ready large language model developed and optimized by APMIC for financial and regulated-business scenarios. The model is based on google/gemma-4-26B-A4B-it and has been further optimized through quantization and financial-domain adaptation to support high-reliability deployment in Traditional Chinese business environments. This release demonstrates APMIC’s end-to-end capability in: - Enterprise LLM optimization and deployment engineering - Deployment readiness for private cloud and on-premise AI infrastructures This model is specifically strengthened for finance-oriented prompts and terminology, with stronger alignment - Financial…

Model · Image and text to text

Agnes-3.0-Qwen

Agnes

Hello! Today we are introducing agnes-3.0-qwen, our most capable reasoning model for agentic coding, tool use, scientific problem solving, long-context analysis, and multimodal understanding. A multimodal reasoning model available as open weights. The model is a hybrid linear-attention / full-attention Mixture-of-Experts architecture (512 experts, 10 active per token) with a multi-token-prediction head, combining long-context understanding with strong coding, agentic, and scientific reasoning at the throughput and cost needed for production workloads. agnes-3.0-qwen is compared against GPT-5.6 Sol (Max), Gemini 3.8 Flash, DeepSeek V4 Flash (Max), MiniMax-M3, Claude Opus 4.8 (Max), and…

Model · Image and text to text

ancient-words-o1-gguf

Rengo Syrup

GGUF files of commreteris/ancient-words-o1 for llama.cpp. The model reads a scanned book page and returns its regions and every word, each with a polygon, as one JSON object. The prompt, output format, training and evaluation are described in that repository's model card. Both files are needed. They were converted from the float32 weights with llama.cpp 0c1e570 (converthftogguf.py --outtype q80 --no-mtp, and --mmproj --outtype f16). Qwen3.5 support needs a llama.cpp build from that commit or later. The fine-tuned checkpoint has no multi-token-prediction head, so these files cannot be used for MTP speculative decoding. llama.cpp needs a Hugging Face token with read access in HFTOKEN. The…

Model · Image and text to text

Apodex1.1-mini-Ciru-Halo-Agent-vllm-strix-halo

Ciru - Crown

Ciru v4.1.2 — October 5, 2026. Applies the shared output-budget correction; Apodex retains its own native MTP4 implementation. Explicit 262,144-token output ceilings now use the fully rendered prompt's remaining context. Weights, pinned engine wheels and sampling defaults are unchanged. Patch notes · Upgrade existing installations. OMP clients must advertise input: [text, image], reasoning: true, contextWindow: 262144 and maxTokens: 262144 to use vision and the full output ceiling. Ciru v4.1.1 — September 30, 2026. Qualified repairs and a retained subset of community PR #2. Apodex retains its own native MTP4, model-specific state layout and released recurrence/attention implementation. The…

Model · Image and text to text

Aura-Prototype-26B-A4B

EldritchLabs

This is a merge of pre-trained language models created using mergekit. This model was merged using the aura merge method. Aura is an experimental method with a live heatmap visualizer. This model took 10 hours to merge using graphv18.py The following models were included in the merge: - TheDrummer/Orion-26B-A4B-v1.1 - Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 - electroglyph/gemma4-26b-fiction-bf16 The following YAML configuration was used to produce this model

Model · Image and text to text

Baanzon-33street-3.8-27B-Uncensored-Aggresive-MTP-GGUF

Opraiz Technology

GGUF distribution of the 27B vision-language model with an aggressive, uncensored behavioral profile, featuring multi-token prediction (MTP) speculative decoding acceleration for fast inference in llama.cpp. - Uncensored / aggressive personality — minimal refusals, combative direct tone - Multimodal (image + text) — vision projector included (mmproj), supports image input - MTP speculative decoding — token-drafting heads for substantially faster generation - Long context — 262,144-token context window (32K-tuned FastMTP variant included) - Multilingual — English / Chinese / multilingual - Quantized for local inference — ready-to-run GGUF quants, no PyTorch weights required All files are…

Model · Image and text to text

backup2026-09-18-Swift-Qwen3.8-27B-Genesis-F16-GGUF

M M

licensename: swift-open-license-1.0 licenselink: https://huggingface.co/ukisai/Swift-Qwen3.8-27b/blob/main/LICENSE libraryname: transformers pipelinetag: image-text-to-text - qwen38 - efficient-thinking - reasoning - token-efficient - genesis - lora basemodel: ukisai/Swift-Qwen3.8-27b basemodelrelation: finetune

Junhao Zhou 4, Zitian Chen 1, Linchao Zhu 2,† CaptchaAgent is a single 9B computer-use policy for all 20 CAPTCHA types in CaptchaArena, fine-tuned from Qwen3.5-9B. One set of weights, one tool schema and one observation format are shared across every type, with no per-task adapter or task-conditioned switching. Both subfolders hold full bf16 weights (9 safetensors shards each) with the tokenizer, processor and chat template, so each loads on its own. SFT/ is the final SFT checkpoint with the LoRA merged in (Table 5a of the paper); RL/ is the best GRPO checkpoint (Table 5b). CaptchaArena scores are on the 4,000-puzzle test split (20 types × 200), with 5 rollouts per puzzle and the unbiased…

Model · Image and text to text

chatter-27b-v1.1

Big Move Intelligent

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

Model · Image and text to text

clef-4bit

MLX Community

Cloudflare/clef converted to MLX (4-bit) for Apple Silicon. Clef turns a state (text, JSON, images, or video) plus a schema of typed questions into a probability for every allowed option, in a single forward pass. It is not a chat model — mlxvlm.generate, mlxlm.generate, and LM Studio will load the backbone but produce meaningless text. Use the bundled clefmlx.py loader, which runs the backbone and the joint schema head. Images (PIL) and videos (frame arrays) go in images / videos, as in the original: See the original model card for the input format, question types, and benchmarks. - processorconfig.json is the original from Cloudflare/clef; prompt/token layout matches the reference…

Model · Image and text to text

clef-8bit

MLX Community

Cloudflare/clef converted to MLX (8-bit) for Apple Silicon. Clef turns a state (text, JSON, images, or video) plus a schema of typed questions into a probability for every allowed option, in a single forward pass. It is not a chat model — mlxvlm.generate, mlxlm.generate, and LM Studio will load the backbone but produce meaningless text. Use the bundled clefmlx.py loader, which runs the backbone and the joint schema head. Images (PIL) and videos (frame arrays) go in images / videos, as in the original: See the original model card for the input format, question types, and benchmarks. - processorconfig.json is the original from Cloudflare/clef; prompt/token layout matches the reference…

Model · Image and text to text

clef-flash-4bit

MLX Community

Cloudflare/clef-flash converted to MLX (4-bit) for Apple Silicon. Clef turns a state (text, JSON, images, or video) plus a schema of typed questions into a probability for every allowed option, in a single forward pass. It is not a chat model — mlxvlm.generate, mlxlm.generate, and LM Studio will load the backbone but produce meaningless text. Use the bundled clefmlx.py loader, which runs the backbone and the joint schema head. Images (PIL) and videos (frame arrays) go in images / videos, as in the original: See the original model card for the input format, question types, and benchmarks. - processorconfig.json is the original from Cloudflare/clef-flash; prompt/token layout matches the…

Model · Image and text to text

clef-flash-8bit

MLX Community

Cloudflare/clef-flash converted to MLX (8-bit) for Apple Silicon. Clef turns a state (text, JSON, images, or video) plus a schema of typed questions into a probability for every allowed option, in a single forward pass. It is not a chat model — mlxvlm.generate, mlxlm.generate, and LM Studio will load the backbone but produce meaningless text. Use the bundled clefmlx.py loader, which runs the backbone and the joint schema head. Images (PIL) and videos (frame arrays) go in images / videos, as in the original: See the original model card for the input format, question types, and benchmarks. - processorconfig.json is the original from Cloudflare/clef-flash; prompt/token layout matches the…

Model · Image and text to text

Clemma-12B

Andrew DeLisa

This gemma4unified model was trained 2x faster with Unsloth and Huggingface's TRL library.

Model · Image and text to text

clover-1-150b-preview

Kitani

libraryname: transformers pipelinetag: image-text-to-text basemodel: - Qwen/Qwen3.8-27B - clover - mixture-of-experts - multimodal - reasoning - open-weights - custom-code Clover 1 150B Preview is the first public checkpoint of Clover 1, an experimental MoE model we're developing at Kitani. It's based on Qwen3.8-27B, but it isn't just a finetune with a different name. We kept much of Qwen3.8's underlying architecture, including the tokenizer, multimodal components, hybrid language backbone, embeddings, and LM head, while replacing the language model's dense FFNs with our own sparse mixture-of-experts setup. Clover currently has 8 full-sized experts per language layer with top-1 routing. The…

This model is a fine-tuned version of Agtian/cpt8. It has been trained using TRL. This model was trained with SFT.

LiquidAI/d1-omni-600M converted to Apple Core AI, re-authored to run on the Neural Engine, with GPU models for a Mac and as an iPhone fallback for long inputs. d1-omni-600M is a 587M-parameter decision model: give it a state — text, JSON, a photo or a voice clip — and named questions, and it returns typed answers (a yes/no probability, a pick among named options, or a level) read from its own distribution in one pass, with no generated tokens. These are the portable models behind the D1Omni iPhone app (camera, voice and text decisions on device). Compile them for your device with compile.sh (one command, ~2.5 min). compile.sh runs xcrun coreai-build compile (Xcode with Core AI) on the three…

Model · Image and text to text

DeepSeek-V4.1-Flash-lossless-CSF

Local Inference Lab

A lossless MXFP4-CSF container of deepseek-ai/DeepSeek-V4.1-Flash at revision dba1be0a40aa45a94ad051997016db3960a90277. CSF stores the routed-expert block-scale planes in a compressed form. Nothing else changes: FP4 weight nibbles, every other tensor, the tensor names, dtypes, shapes and the source metadata are all byte-identical. Decoding is integer arithmetic on bytes, with no requantization or fitting. The original safetensors shards can be restored bit-exactly. - Built with trellis-quant trellisquant.losslessscalecheckpoint (commit 60feca330087, family deepseekv41). - verification.json: every original shard was rebuilt from this container, and both the rebuilt and the stored files…

Model · Image and text to text

DeepSeek-V4.1-Flash-oQ2.5e-mtp

Jun Kim

This model was quantized using oQ (oMLX v0.7.1) mixed-precision quantization with Improved oQe Quantization (jundot/omlx#4385). A mixed-precision MLX checkpoint of deepseek-ai/DeepSeek-V4.1-Flash, a multimodal mixture-of-experts model with 552B backbone parameters, a Causal Encoder-Decoder layout, Compressed Sparse Attention 2 (CSA2), Engram conditional memory and DSpark speculative decoding. It includes the language model, the Engram tables, the vision encoder, the multimodal aligner and the DSpark MTP weights. The effective average is 3.104 bits/weight excluding the Engram tables, or 3.592 bits/weight for the complete checkpoint. oQ2.5e allocates bits by measured layer sensitivity, then…

Model · Image and text to text

DeepSeek-V4.1-Flash-TensorFold-TP2-2xGB10

Bertholomus AI

BertholomusAI recipe: DeepSeek-V4.1-Flash served tensor-parallel over two NVIDIA DGX Spark / GB10 nodes by a deepseekv41 model family we wrote for TensorFold (by ashhart). Clean-room: the model math is re-implemented from DeepSeek's MIT inference code and tech report; no code from other DeepSeek-V4.1 recipes or kits was read or copied (ATTRIBUTION.md). Decode and prefill are what we worked on this time. The same release suite on the served lane for both builds (v0.4 = engine bd0024d at --context 262144, v0.5 = 808eb4a at --context 1048576), same client (tools/dsv41/kitbench.py), same prompts, greedy, DSpark drafting. Median of 3 unless noted. four long streams together give the same replies…

with permanent weight-level abliteration — the safety guardrails have been surgically removed while preserving MMLU capability, vision, reasoning, MTP (DSpark), and multi-turn coherence. Proprietary weight-level abliteration developed by the dealignai research team. No custom model.py, no runtime hooks, no steering vectors — it's a standard checkpoint that loads exactly like the base model. The refusal circuitry is surgically removed while every capability-critical component (routed experts, Engram memory, CSA2 sparse attention, DSpark draft head, vision tower, router gates, norms, embeddings) is preserved byte-identical to the base. Every response 4-tier graded (HARDREF / SOFTRED / HEDGE /…

Model · Image and text to text

DeepSeek-V4.1-MLX-Q9

Inferencer

M3 Ultra 512GB + M4 Max 128GB: ~15 tokens/s @ 1000 tokens ~501 GiB This build repacks the majority of the model's existing quantization, with the remainder quantized to 9 bits. This configuration empirically yielded the highest quality in our testing. We are not the creator, originator, or owner of any model listed. Each model is created and provided by third parties. Models may not always be accurate or contextually appropriate. You are responsible for verifying the information before making important decisions. We are not liable for any damages, losses, or issues arising from its use, including data loss or inaccuracies in AI-generated content.

Model · Image and text to text

DN-MOPD-Qwen3.5-2B-160updates

XinLi

A Qwen3.5-2B student trained with DN-MOPD (Domain-Normalized Multi-Teacher On-Policy Distillation) continued to 160 updates (paper Table 5). Three same-size RL experts (math, code, instruction following) teach one student on its own responses; each prompt is scored by the expert of its domain, and DN-MOPD rescales each domain's token-level feedback by its measured spread, wd = clip(σall / σd, 0.25, 4), so that no domain dominates the shared update. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in…

Model · Image and text to text

DN-MOPD-Qwen3.5-2B-baseline-label-160updates

XinLi

The Label baseline of the DN-MOPD paper at Qwen3.5-2B continued to 160 updates (paper Table 5): multi-teacher on-policy distillation with label routing (each prompt is scored by the expert of its domain, every domain multiplier is 1). Released for comparison with DN-MOPD-Qwen3.5-2B; it is not the proposed method. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in recipes/qwen3.5/ and docs/recipe.md. This model was trained and evaluated with the non-thinking chat format. Pass enablethinking=False to the…

Model · Image and text to text

DN-MOPD-Qwen3.5-4B-160updates

XinLi

A Qwen3.5-4B student trained with DN-MOPD (Domain-Normalized Multi-Teacher On-Policy Distillation) continued to 160 updates (paper Table 5). Three same-size RL experts (math, code, instruction following) teach one student on its own responses; each prompt is scored by the expert of its domain, and DN-MOPD rescales each domain's token-level feedback by its measured spread, wd = clip(σall / σd, 0.25, 4), so that no domain dominates the shared update. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in…

Model · Image and text to text

DN-MOPD-Qwen3.5-4B-baseline-label-160updates

XinLi

The Label baseline of the DN-MOPD paper at Qwen3.5-4B continued to 160 updates (paper Table 5): multi-teacher on-policy distillation with label routing (each prompt is scored by the expert of its domain, every domain multiplier is 1). Released for comparison with DN-MOPD-Qwen3.5-4B; it is not the proposed method. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in recipes/qwen3.5/ and docs/recipe.md. This model was trained and evaluated with the non-thinking chat format. Pass enablethinking=False to the…

Model · Image and text to text

DN-MOPD-Qwen3.5-9B-baseline-label-160updates

XinLi

The Label baseline of the DN-MOPD paper at Qwen3.5-9B continued to 160 updates (paper Table 5): multi-teacher on-policy distillation with label routing (each prompt is scored by the expert of its domain, every domain multiplier is 1). Released for comparison with DN-MOPD-Qwen3.5-9B; it is not the proposed method. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in recipes/qwen3.5/ and docs/recipe.md. This model was trained and evaluated with the non-thinking chat format. Pass enablethinking=False to the…

The trained EvoSteer orchestrator as a complete model: the trained LoRA adapter (rank 64, alpha 128 on qproj, kproj, vproj, oproj, outproj and inprojqkv) is merged into the Qwen3.5-9B weights. The merged layers are stored in float32. The LoRA update is about 0.06% of the weight norm, and rounding the merged layers to bfloat16 would discard most of it, so load the model in float32 to reproduce the trained orchestrator exactly; all other layers are the unchanged bfloat16 base weights. evosteerheads.safetensors holds the learned value and flow heads of the same checkpoint (heads. and the published value snapshot publishedvalue.).

Model · Image and text to text

ezjev-4b-s2

Everett

A typed-decision model (Jev-style /v1/systemone: choice, noul and score questions answered with a probability per option from one forward pass). LoRA fine-tune of Qwen/Qwen3.5-4B, merged into full weights, in two stages: 1. Stage 1 (ezjev-4b): LoRA r=16 on all language-model linear layers (attention, Gated-DeltaNet, MLP), LR 1e-4, one epoch over ~85k rows / ~106k questions. Loss: cross-entropy + Brier over the option labels, llm2jev chat prompt (thinking off). 2. Stage 2: a second LoRA at LR 5e-5 on ~30k rows targeting weak task families (RAG hallucination, product relevance, code-output selection, select-all-that-apply, clinical NLI, phishing emails, claim verification) with 40% replay.…

Model · Image and text to text

flux-3-image-klein-pup-Q8_0-GGUF

KJS

This model was converted to GGUF format from black-forest-labs/flux-3-image-klein-pup using latest llama.cpp 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).

Model · Image and text to text

Frontis-MA1-35B-SFT

Frontis AI

This is the supervised fine-tuning checkpoint of the Frontis-MA1 family, before OpenMLE reinforcement learning. It is intended for research on machine learning engineering and program transformation. The actual 24,350-row training input is distinct from the 26,259-example public SFT trace collection. Dataset cutoff configuration was 32,768 tokens; this does not establish validated long-context inference performance. No standalone SFT benchmark claim is made by this release. The separately released RL model is FrontisAI/Frontis-MA1-35B. Pin the published commit SHA for reproducibility. This model requires substantial memory; use the dependency versions recorded in the release validation…

LoRA adapters for google/gemma-4-12B-it trained with supervised fine-tuning (SFT) on visual questions generated by PixelProof, as described in the paper PixelProof: Visual Question Generation through Forward–Inverse Agreement. In PixelProof, a coding agent writes a forward program that draws images and computes their answers and an inverse program that must recover each answer from the image alone; a question is kept only if the two agree. Each adapter is stored in its own subfolder. The repository contains no merged weights. Training uses transformers.Trainer and PEFT LoRA adapters on the language model's attention and feed-forward projections; the vision tower is not directly adapted. The…

4-bit quantization of google/gemma-4-26B-A4B (commit 24548b6), produced for efficient vLLM serving. Quantized by axteam. - Scheme — W4A16 — FP4 (E2M1) weights group 16 (FP8 E4M3 scale + per-tensor FP32 global scale), BF16 activations, stored as compressed-tensors nvfp4-pack-quantized. - Kept in BF16 — vision tower, vision projection, token embeddings (tied with lmhead), MoE router, and all norms, plus the shared MLP. These are left unquantized to preserve quality (a small fraction of total parameters). - Tokenizer / chattemplate / processorconfig / generationconfig are byte-identical to the original, including Google's July-2026 corrected chattemplate. config.json is the original plus a…

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from…

4-bit quantization of google/gemma-4-31B-it (commit 842da37), produced for efficient vLLM serving. Quantized by axteam. - Scheme — W4A16 — FP4 (E2M1) weights group 16 (FP8 E4M3 scale + per-tensor FP32 global scale), BF16 activations, stored as compressed-tensors nvfp4-pack-quantized. - Kept in BF16 — vision tower, vision projection, token embeddings (tied with lmhead), MoE router, and all norms. These are left unquantized to preserve quality (a small fraction of total parameters). - Tokenizer / chattemplate / processorconfig / generationconfig are byte-identical to the original, including Google's July-2026 corrected chattemplate. config.json is the original plus a quantizationconfig block.…

LoRA adapters for google/gemma-4-31B-it trained with supervised fine-tuning (SFT) on visual questions generated by PixelProof, as described in the paper PixelProof: Visual Question Generation through Forward–Inverse Agreement. In PixelProof, a coding agent writes a forward program that draws images and computes their answers and an inverse program that must recover each answer from the image alone; a question is kept only if the two agree. Each adapter is stored in its own subfolder. The repository contains no merged weights. Training uses transformers.Trainer and PEFT LoRA adapters on the language model's attention and feed-forward projections; the vision tower is not directly adapted. The…

Model · Image and text to text

Gemma-4-Writers-31B-V2-W8A16-FP8

Hoborific

Quantized version of Ateron/Gemma-4-Writers-31B-V2. Offline-quantized W8A16 FP8 in the float-quantized format: weights in float8e4m3fn with per-output-channel symmetric scales, activations kept in bf16/fp16. For each linear layer, every output row gets its own scale starting from amax / 448, refined by an MSE clip search over ~9 clip fractions (0.8–1.0× amax) picking the lowest-error scale per row. Weights are then quantized q = e4m3(w / scale) with round-to-nearest and saturation. This per-channel + clipping scheme gives better SNR than vLLM's online per-tensor --quantization fp8 path. Only 2D linear projection weights are quantized (attention q/k/v/o, MLP gate/up/down). Embeddings, norms…

Model · Image and text to text

Gemma-4-Writers-31B-W8A16-FP8

Hoborific

Quantized version of Ateron/Gemma-4-Writers-31B. Offline-quantized W8A16 FP8 in the float-quantized format: weights in float8e4m3fn with per-output-channel symmetric scales, activations kept in bf16/fp16. For each linear layer, every output row gets its own scale starting from amax / 448, refined by an MSE clip search over ~9 clip fractions (0.8–1.0× amax) picking the lowest-error scale per row. Weights are then quantized q = e4m3(w / scale) with round-to-nearest and saturation. This per-channel + clipping scheme gives better SNR than vLLM's online per-tensor --quantization fp8 path. Only 2D linear projection weights are quantized (attention q/k/v/o, MLP gate/up/down). Embeddings, norms…

Model · Image and text to text

GLM-5.3-Flash-Uncensored-Mixed-Quant-GGUF

Sunghoon Baek

Mixed-precision GGUF derived from The recipe preserves higher precision in always-active paths and six boundary routed layers, with IQ2XXS as the minimum precision. Status: published. File identity and bounded single-GPU image checks are complete. General text, MTP and DGX Spark context checks remain pending. Gate/up use the same format within each layer. Each tensor uses one block format throughout; routed matrix widths remain divisible by 256. Source-weight reconstruction proxies were collected for all 126 trunk expert matrices. Q2K does not require an imatrix; optional importance weighting uses available activation statistics without changing the format or file size. Calibration covered…

Model · Image and text to text

GLM-5.3-Flash-Uncensored-MLX

GCSA

The intervention source is GLM-5.3-Flash-Ablitered2 (LoRA v2). The files here are complete MLX model weights, not standalone LoRA adapters; do not apply v2 again. These releases are intended for controlled safety research and red-teaming. Reducing refusal behavior also weakens a safety boundary; read the limitations and disclaimer before use. The directories are alternative complete model formats, not parts to load together. The original block-FP8 tensors were dequantized before MLX quantization. Token embeddings and the language-model head remain BF16; affine scales and biases are FP16. Quantization can change behavior. The consuming MLX runtime must support the GLM5-Next architecture…

Model · Image and text to text

GLM5.3-Flash-E192-DGX-Spark

AutoTrust AI Lab

autotrust/GLM5.3-Flash-E192-DGX-Spark is a compact build of zai-org/GLM-5.3-Flash for desktop Blackwell systems like NVIDIA DGX Spark. It's an unofficial derivative. It keeps 192 of the 288 routed experts in each layer by Neural Architecture Search (NAS), uses NVFP4 for the experts and still activates 18 B parameters per token. The weights take 124 GiB, the smallest in the family: two DGX Sparks connected by ConnectX-7 get about 62 GiB of weights per node and the most KV room of any build, and a single 180 GB Blackwell GPU (B200/GB200) runs it with 32 GiB of KV cache. The KV cache uses FP8. The repo includes a ready-to-use NVFP4 MTP draft in mtp-nvfp4/ (6.95 GB) for speculative decoding. On…

Model · Image and text to text

GLM5.3-Flash-E256-DGX-Spark

AutoTrust AI Lab

autotrust/GLM5.3-Flash-E256-DGX-Spark is a compact build of zai-org/GLM-5.3-Flash for Blackwell systems, from DGX Spark clusters to a single B200. It's an unofficial derivative. It keeps 256 of the 288 routed experts in each layer by Neural Architecture Search (NAS), uses NVFP4 for the experts and still activates 18 B parameters per token. It's the largest architecture in the family, with the best Chinese, code and vision scores. The KV cache uses FP8. The repo includes a ready-to-use NVFP4 MTP draft in mtp-nvfp4/ (6.95 GB) for speculative decoding. On a single B200 it gives 1.84× single-stream decode speed (1.94× with 3 draft tokens), and output quality is unchanged. The model keeps the…

Model · Image and text to text

Herbert_v0.5

Lein Enterprises

A LoRA on Qwen/Qwen3.8-27B for typed decisions over text and images. It continues from e12ex2/jev-vision-27b and adds training on synthetic decision questions whose answers are known by construction. Give it a state (text, images, or both), a question and a list of options; it returns a probability for every option from one forward pass. Nothing is generated. Decision Index 0.3, public vision suite (12,210 rows, all answered), official kit: Options are labelled A, B, C,... in the prompt; the answer is read from the label logits at the answer position and softmaxed over the listed options only. Up to 624 options per question (A-Z, a-z, 0-9, then single-token two-letter labels); larger…