# Image and Text to Text Models, Page 10: Open-Weight AI Source: https://savrn.com/models/tasks/image-and-text-to-text/page/10 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Models by Task # Image and Text to Text Models 577 open-weight image and text to text models in the SAVRN Model Hub, with Qwen, Solstice-AI and Isaac Godfried publishing the most. 577Models 278Publishers 665K to 2.8TParameter range 8Licenses ## SAVRN's Take Hand one of these models a photograph or a screenshot, ask a question, and it answers in text. Our hub lists 138 models for the job, 49 of them in the Index. Qwen publishes 28 and Unsloth AI 11, with Google, LM Studio Community and the OpenVLA Collaboration at 4 each. Downloads follow suit: Qwen3-VL-8B-Instruct pulled 19,098,599 last month, ahead of Qwen3.6-35B-A3B-FP8 at 10,009,539 and Google's gemma-4-26B-A4B-it at 9,794,984. All eight leaders share a 262,144-token context and fit on a single MI300X at $1.85 an hour on the Index, so the question is how much of that card each takes. At 16-bit there are two tiers: Qwen3.5-4B, Qwen3-VL-8B-Instruct and Qwen3.5-9B need 11.2 to 23.2 GB, leaving the rest of the card for batch, while gemma-4-26B-A4B-it, both Qwen3.8-27B builds, gemma-4-31B-it and Qwen3.6-35B-A3B-FP8 need 61.9 to 86.3 GB. Halve those for 8-bit; 4-bit puts the whole group under 22 GB. Kimi-K3, at 2.8T parameters, is the outlier: 6,671.8 GB at 16-bit, which no Index host can hold, and 1,668.0 GB at 4-bit across 7 MI325X cards at $14.00 an hour. Licensing is the easy part. 89 of the 138 are Apache 2.0, including all eight leaders, and 16 more are MIT, so most of the catalog deploys without a negotiation. The 20 marked other, Kimi-K3 among them, and the 5 under Gemma terms need a read first; 3 state no license, which we treat as a no. Settle the memory tier first, then the license, then treat the download count as a tally of operators who got there before you. SAVRN Research, 2026-09-18 ## Most Downloaded | Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit | | --- | --- | --- | --- | --- | --- | | [gemma-4-26B-A4B-it](https://savrn.com/models/gemma-4-26b-a4b-it) | Google | 25.8B | apache-2.0 | 9.4M | 1x MI300X, $1.85/hr | | [gemma-4-31B-it](https://savrn.com/models/gemma-4-31b-it) | Google | 31.3B | apache-2.0 | 8.6M | 1x MI300X, $1.85/hr | | [Qwen3.8-27B](https://savrn.com/models/qwen3-8-27b) | Qwen | 27.8B | apache-2.0 | 7.8M | 1x MI300X, $1.85/hr | | [Qwen3.5-4B](https://savrn.com/models/qwen3-5-4b) | Qwen | 4.7B | apache-2.0 | 7.6M | 1x MI300X, $1.85/hr | | [Qwen3.5-9B](https://savrn.com/models/qwen3-5-9b) | Qwen | 9.7B | apache-2.0 | 6.7M | 1x MI300X, $1.85/hr | | [Qwen3.8-Flash-Next-GSQ-RCO-GGUF](https://savrn.com/models/qwen3-8-flash-next-gsq-rco-gguf) | IST Austria Distributed Algorithms and Systems Lab | — | apache-2.0 | 6.4M | — | | [GLM-5.3-Flash](https://savrn.com/models/glm-5-3-flash) | Z.ai | 321.3B | mit | 6.4M | 3x MI355X, $7.77/hr | | [Qwen2.5-VL-7B-Instruct](https://savrn.com/models/qwen2-5-vl-7b-instruct) | Qwen | 8.3B | apache-2.0 | 5M | 1x MI300X, $1.85/hr | | [Qwen3-VL-8B-Instruct](https://savrn.com/models/qwen3-vl-8b-instruct) | Qwen | 8.8B | apache-2.0 | 4.8M | 1x MI300X, $1.85/hr | | [Qwen3.8-27B-FP8](https://savrn.com/models/qwen3-8-27b-fp8) | Qwen | 27.8B | apache-2.0 | 4.3M | 1x MI300X, $1.85/hr | ## Licenses | License | Models | Commercial use | | --- | --- | --- | | [apache-2.0](https://savrn.com/models/licenses/apache-2-0) | 360 | Yes | | other | 105 | Read the license | | [mit](https://savrn.com/models/licenses/mit) | 53 | Yes | | not stated | 40 | Not stated | | [gemma](https://savrn.com/models/licenses/gemma) | 7 | Yes, with conditions | | [openrail](https://savrn.com/models/licenses/openrail) | 3 | Yes, with conditions | | [openmdw-1.1](https://savrn.com/models/licenses/openmdw-1-1) | 2 | Read the license | | [llama2](https://savrn.com/models/licenses/llama2) | 2 | Read the license | ## Who Publishes Them | Publisher | Models | | --- | --- | | [Qwen](https://savrn.com/model-publishers/qwen) | 28 | | [Solstice-AI](https://savrn.com/model-publishers/solstice-ai) | 16 | | [Isaac Godfried](https://savrn.com/model-publishers/isaacmg) | 16 | | [RKNNAI](https://savrn.com/model-publishers/rknnai) | 15 | | [David Belton](https://savrn.com/model-publishers/davidau) | 14 | | [Unsloth AI](https://savrn.com/model-publishers/unsloth) | 13 | ## All 577 Models, Page 10 of 10 Model · Image and text to text ### [smoke_stage1_v3](https://savrn.com/models/smoke-stage1-v3) [Linda](https://savrn.com/model-publishers/lindafei001) Open weights other 8.8B parameters transformers [View model](https://savrn.com/models/smoke-stage1-v3) Model · Image and text to text ### [span-02](https://savrn.com/models/span-02) [He Li](https://savrn.com/model-publishers/lihe07) span-02 is a decision model: it answers typed questions about a state (JSON or text, plus up to 4 images) with a full probability distribution over the options, in the Decision Index /v1/systemone wire format. It is a fine-tune of Gemma 4 26B-A4B (mixture of experts, about 4B active parameters per token), quantized (NVFP4 experts, FP8 dense layers) and served by a custom Rust/CUDA engine for Blackwell GPUs (sm120: RTX 5090, RTX PRO 6000) that is included in this repo (engine/). Before answering it may think. If the top probability of the no-think answer is below 0.8, the model writes a greedy thought. Its length limit comes from a deterministic thought budget: a fixed cost model… Open weights apache-2.0 15.8B parameters 262,144 tokens [View model](https://savrn.com/models/span-02) Model · Image and text to text ### [sph_baseline1_openmobile_sft](https://savrn.com/models/sph-baseline1-openmobile-sft) [Linda](https://savrn.com/model-publishers/lindafei001) Open weights other 8.8B parameters transformers [View model](https://savrn.com/models/sph-baseline1-openmobile-sft) Model · Image and text to text ### [sriq-heretic-spatial-9b-MIXED-10020-preview](https://savrn.com/models/sriq-heretic-spatial-9b-mixed-10020-preview) [Sparse Reasoning Intelligence & Quality](https://savrn.com/model-publishers/sriq-ai) Merged bf16 weights. Reasons in compressed Simplified Chinese and answers in the language it was asked in. No benchmarks were run for this release. Nothing here reports rows that reason in Chinese, verification rate, or looping against the base model. Evaluate it on your own workload before relying on it. - Training saw 2,940 of the dataset's 2,940 rows at 700 steps on an effective batch of 8, about 1.90 of one epoch. - Inherits the base model's limitations. Do not pass enablethinking=False: the template prefills an empty and the trace disappears. Wall clock and cost are measured by the agent from its own start to publish, so they cover the base-model download and the merge upload as well… Open weights apache-2.0 9.7B parameters 262,144 tokens transformers [View model](https://savrn.com/models/sriq-heretic-spatial-9b-mixed-10020-preview) Model · Image and text to text ### [Step-5-Preview-NVFP4-Local](https://savrn.com/models/step-5-preview-nvfp4-local) [Brandon Music](https://savrn.com/model-publishers/brandonmusic) A lossless repack of stepfun-ai/Step-5-Preview-NVFP4 (revision bc246cc2848e17b37f377eda8fd825c5b3018d1f), 17.1 GB smaller. The NVFP4 block scales of the routed experts are stored in b12x's compressed scale format (CSF), the lossless format introduced with nvidia/Qwen3.8-Flash-Next-FP4-Local. Every FP4 weight byte and every other tensor is identical to StepFun's checkpoint, and every scale decodes to StepFun's exact bytes, so this checkpoint computes exactly what Step-5-Preview-NVFP4 computes. Step 5 Preview is StepFun AI's model: a sparse mixture of experts with 600B total and 27B active parameters, text and image input, up to 1M tokens of context. Model card: stepfun-ai/Step-5-Preview.… Open weights other 338.5B parameters 1,048,576 tokens [View model](https://savrn.com/models/step-5-preview-nvfp4-local) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Companion-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-companion-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) TopP: 0.8 ~ 0.95 RepetitionPenalty: 1 ~ 1.1 Space Time 2. 邮箱: Zeus66@163.com Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-companion-gguf) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Uncensored-Q2_K-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q2-k-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q2-k-gguf) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Uncensored-Q3_K_L-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-l-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-l-gguf) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Uncensored-Q3_K_M-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-m-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-m-gguf) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Uncensored-Q3_K_S-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-s-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q3-k-s-gguf) Model · Image and text to text ### [Suri-Qwen-3.8-27B-Uncensored-Q4_0-GGUF](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q4-0-gguf) [Space Time](https://savrn.com/model-publishers/spacetimee) This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights transformers [View model](https://savrn.com/models/suri-qwen-3-8-27b-uncensored-q4-0-gguf) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27b](https://savrn.com/models/swift-1-5-qwen3-8-27b) [LEBON](https://savrn.com/model-publishers/jeromelebon91) Swift 1.5 Qwen3.8-27B is UkisAI's reasoning-efficient derivative of Qwen3.8-27B. It uses 58.5% fewer thinking tokens while scoring 0.35% higher than the base, for a 1.95× speed-up on several tasks. Swift 1.5 is a direct upgrade from Swift 1.0, our model with 350k+ downloads, delivering stronger overall performance than both base and Swift 1.0 in various tasks, especially coding and agentic, while using fewer thinking tokens. We accomplished that by scaling up the post-training (RL and OPD) from the previous version. We gave base Qwen3.8-27B and Swift 1.5 27B the same prompt: Base Qwen3.8-27B took 104.6 minutes to build its game. Swift 1.5 took 11.39 minutes. We made Swift efficient by… Open weights other 27.8B parameters 262,144 tokens transformers [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27b-heretic](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic) [Akumaburn](https://savrn.com/model-publishers/akumaburn) An abliterated build of ukisai/Swift-1.5-Qwen3.8-27b, UkisAI's reasoning-efficient derivative of Qwen3.8-27B. Refusal directions are removed by directional ablation; nothing else is retrained. identical to the source's), lmhead, embedtokens, all norms, and every GatedDeltaNet recurrent gate. - Size 52G, 1,199 tensors (333 vision, 15 MTP). Produced with Heretic v2.0.0.dev0: 260 TPE trials, seed 42, directional ablation on attn.oproj (16 modules), attn.outproj (48) and mlp.downproj (64). Baseline refusal score before abliteration: 98/100. Trial 140 was selected — Pareto index 1, not index 0. Index 0 (trial 161) scored 22/100 keywords at KL 0.1091; trial 140 scores 23/100 at KL 0.0602. One… Open weights other 27.4B parameters 262,144 tokens vllm [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27b-heretic-SmoothQuant-W8A8-INT8](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic-smoothquant-w8a8-int8) [Akumaburn](https://savrn.com/model-publishers/akumaburn) An INT8 W8A8 quantization of (the abliterated BF16 build of ukisai/Swift-1.5-Qwen3.8-27b), for fast serving on GPUs where native FP8 is unavailable or undesirable. native CUTLASS INT8 tensor-core path. No activation scales are stored on disk; they are computed per token at runtime. vision tower (333 tensors), GatedDeltaNet recurrent gates, all norms. These are FP16, not the source's BF16 — the 16-bit residual precision of this build is float16 throughout (config.json: "dtype": "float16"). with the MTP drafter re-expressed and ablated in the rotated basis. The rotation was verified before calibration by an fp64 whole-graph equivalence check (all 19 paths within 1.6–2.5e-4; every tensor… Open weights other 27.4B parameters 262,144 tokens vllm [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic-smoothquant-w8a8-int8) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27b-heretic-W4A16](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic-w4a16) [Akumaburn](https://savrn.com/model-publishers/akumaburn) A 18G INT4 build of for single-user decoding and low-VRAM deployment. (Marlin on Ampere). - Unrotated, so the DFlash2 drafter works. GatedDeltaNet gates, all norms. 4-bit costs 4.5× the rotated W8A8's quantization divergence (0.0264) and 1.5× the unrotated W8A8's (0.0791). Choose this build when the 18G footprint or single-user decode speed matters more than fidelity. Produced with Heretic v2.0.0.dev0: 260 TPE trials, seed 42, directional ablation on attn.oproj (16 modules), attn.outproj (48) and mlp.downproj (64). Baseline refusal score before abliteration: 98/100. Trial 140 was selected — Pareto index 1, not index 0. Index 0 (trial 161) scored 22/100 keywords at KL 0.1091; trial 140… Open weights other 6.3B parameters 262,144 tokens vllm [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b-heretic-w4a16) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27B-Uncensored-Dynamic-MTP-GGUF](https://savrn.com/models/swift-1-5-qwen3-8-27b-uncensored-dynamic-mtp-gguf) [AJ Gazin](https://savrn.com/model-publishers/ajgazin) GGUF quants of ajgazin/Swift-1.5-Qwen3.8-27B-Uncensored-MTP, an abliterated Swift 1.5 Qwen3.8-27B (UkisAI's reasoning-efficient fine-tune of Qwen3.8-27B). For vLLM and size, with Unsloth's importance matrix. - MTP head included in every main GGUF, for self-speculative decoding in llama.cpp. Each quant is one file, Swift-1.5-Qwen3.8-27B-Uncensored-Dynamic-MTP-.gguf; the BF16 is Swift-1.5-Qwen3.8-27B-Uncensored-MTP-BF16.gguf. --spec-type draft-mtp needs a llama.cpp build with MTP support for qwen35. The MTP head loads from the main GGUF; there is no separate draft file. Sampling, as for Swift and Qwen: temperature 1.0, topp 0.95, topk 20, minp 0. The model thinks before answering by default.… Open weights other gguf [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b-uncensored-dynamic-mtp-gguf) Model · Image and text to text ### [Swift-1.5-Qwen3.8-27B-Uncensored-NVFP4](https://savrn.com/models/swift-1-5-qwen3-8-27b-uncensored-nvfp4) [AJ Gazin](https://savrn.com/model-publishers/ajgazin) NVFP4 checkpoint of an abliterated Swift 1.5 Qwen3.8-27B (UkisAI's reasoning-efficient fine-tune of Qwen3.8-27B). For vLLM and SGLang. GGUFs for llama.cpp: (measured on the BF16 source). - Swift's own NVFP4 recipe, unmodified, from ukisai/Swift-Qwen3.8-27B-NVFP4, calibrated with NVIDIA ModelOpt. The module split matches UkisAI's Swift 1.5 NVFP4 exactly. - MTP head and vision tower in BF16, bit-identical to the source. 21.9 GB, NVIDIA ModelOpt mixed-precision format. Needs a vLLM with ModelOpt mixed-precision support. No --quantization flag. Sampling, as for Swift and Qwen: temperature 1.0, topp 0.95, topk 20, minp 0. The model thinks before answering by default. Same format, recipe, module… Open weights other 18.2B parameters 262,144 tokens vllm [View model](https://savrn.com/models/swift-1-5-qwen3-8-27b-uncensored-nvfp4) Model · Image and text to text ### [Swift-1.5-Qwen3.8-Flash-Next-MXFP4-GGUF](https://savrn.com/models/swift-1-5-qwen3-8-flash-next-mxfp4-gguf) [Tsaipifong](https://savrn.com/model-publishers/tsaipifong) Open weights other [View model](https://savrn.com/models/swift-1-5-qwen3-8-flash-next-mxfp4-gguf) Model · Image and text to text ### [Swift-Qwen3.8-27b-heretic-SmoothQuant-W8A8-INT8](https://savrn.com/models/swift-qwen3-8-27b-heretic-smoothquant-w8a8-int8) [Akumaburn](https://savrn.com/model-publishers/akumaburn) An INT8 W8A8 quantization of (the abliterated BF16 build of a merged LoRA finetune of Qwen/Qwen3.8-27B), for fast serving on GPUs where native FP8 is unavailable or undesirable. int-quantized, native CUTLASS INT8 tensor-core path. embedtokens, the vision tower, GatedDeltaNet recurrent gates, all norms. Note these are FP16, not the source's BF16 — the 16-bit residual precision of this build is float16 throughout (config.json: "dtype": "float16"). (llm-compressor), with the MTP drafter ablated in the rotated basis. The MTP head is unquantized but is not byte-identical to the BF16 build's: it carries the same residual-stream rotation and RMSNorm fold as the rest of the model, so the two heads… Open weights other 27.4B parameters 262,144 tokens vllm [View model](https://savrn.com/models/swift-qwen3-8-27b-heretic-smoothquant-w8a8-int8) Model · Image and text to text ### [Swift-Qwen3.8-27B-MIX_GGUF](https://savrn.com/models/swift-qwen3-8-27b-mix-gguf) [Vmarcelo](https://savrn.com/model-publishers/vmarcelo) Three custom mixed-tensor quantizations of ukisai/Swift-Qwen3.8-27b, a fine-tune of Qwen/Qwen3.8-27B — the dense 27B multimodal model with hybrid GatedDeltaNet + Gated Attention and a Multi-Token Prediction (MTP) head. Built with llama.cpp 58367713a, using a purpose-built importance matrix and per-tensor quantization overrides. All three are designed around the 16 GB VRAM budget of consumer GPUs such as the AMD Radeon RX 9070 XT, and all keep the MTP head embedded so speculative decoding works out of the box — no second file to download or load. Both quants follow the same philosophy: protect what the model cannot afford to lose, and pay for it out of the FFN, which is two-thirds of the… Open weights other 262,144 tokens gguf [View model](https://savrn.com/models/swift-qwen3-8-27b-mix-gguf) Model · Image and text to text ### [Swift1.5-Qwen3.8-Flash-Next-q4-MTPLX](https://savrn.com/models/swift1-5-qwen3-8-flash-next-q4-mtplx) [Elvis McFee](https://savrn.com/model-publishers/ecmcfee) This is a quantized MLX pack of ukisai/Swift1.5-Qwen3.8-Flash-Next — UkisAI's reasoning-efficient finetune of Qwen3.8-Flash-Next — produced by applying the MTPLX "Optimized Speed" quantization recipe from Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed to the Swift-1.5 weights. It was forged locally with mtplx forge (MTPLX 2.12.2, qwen4-exp lane) from source revision 0bd4fe22431372cdad1979267d3ab45aa7e6150a, full-load verified before stamping, and runs with MTP draft-acceptance active and the n-gram table offloaded to SSD — the same target class as the original pack: M5 Ultra 96 GB or M5 Max 128 GB Macs with fast NVMe storage. Note vs the original Youssofal pack: this build runs qsa8bit… Open weights other 20.9B parameters 262,144 tokens mlx-lm [View model](https://savrn.com/models/swift1-5-qwen3-8-flash-next-q4-mtplx) Model · Image and text to text ### [Synin-V1.1-Flash](https://savrn.com/models/synin-v1-1-flash) [Synin AI Lab](https://savrn.com/model-publishers/synin) We introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. The model natively processes images and text, and generates text autoregressively. Architecture. DeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organized as a 20-layer causal encoder followed by a 20-layer decoder. With CED, the decoder's global KV cache is projected from the final encoder hidden states rather than derived from each decoder layer's own hidden states. This allows the model to activate only 8B parameters per token during prefill and 16B during decode, substantially… Open weights mit 763.2B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/synin-v1-1-flash) Model · Image and text to text ### [Ternary-Bonsai-2-4B-onwx](https://savrn.com/models/ternary-bonsai-2-4b-onwx) [Ryu Gyo](https://savrn.com/model-publishers/ryugyosoft) Prism ML の Ternary-Bonsai-2-4B（Qwen3.5-4B を三値の重みで学習し直したモデル）を、onwx 用に変換したものです。onwx は LLM を AMD Ryzen AI の NPU（XDNA2）だけで 動かすアプリで、このリポジトリにはモデルと、このモデル専用の NPU プログラムが入っています（アプリは別にインストールします）。 重みは三値（−1・0・+1）を 2 ビットのまま NPU に載せ、行列計算・DeltaNet・アテンション・LM ヘッドまで 1 トークンの計算をすべて NPU の 32 個の AI エンジンで行います。同じ大きさの int4 のモデルよりダウンロードもメモリも小さく、生成も速くなります。画像のエンコーダー（ViT）は CPU で動きます。NPU 向けのプログラムはビルド済みなので、PC の上でコンパイルを待つことはありません。 （onwx 0.7.0、Ryzen AI Max+ 395（Windows）で NPU の電力モードを performance にして測定。Ryzen AI 300 シリーズ（HX 370 など）も NPU は同じ 32 個の AI エンジンで、同じ方式で動きます。標準の電力モードでは NPU のクロックが下がって 2 割ほど遅くなります。電力モードは管理者の PowerShell で C:\Windows\System32\AMD\xrt-smi.exe configure --pmode performance。Windows の Copilot+ の AI 機能が NPU… Open weights apache-2.0 262,144 tokens [View model](https://savrn.com/models/ternary-bonsai-2-4b-onwx) Model · Image and text to text ### [Ternary-Bonsai-2-9B-onwx](https://savrn.com/models/ternary-bonsai-2-9b-onwx) [Ryu Gyo](https://savrn.com/model-publishers/ryugyosoft) Prism ML の Ternary-Bonsai-2-9B（Qwen3.5-9B を三値の重みで学習し直したモデル）を、onwx 用に変換したものです。onwx は LLM を AMD Ryzen AI の NPU（XDNA2）だけで 動かすアプリで、このリポジトリにはモデルと、このモデル専用の NPU プログラムが入っています（アプリは別にインストールします）。 重みは三値（−1・0・+1）を 2 ビットのまま NPU に載せ、行列計算・DeltaNet・アテンション・LM ヘッドまで 1 トークンの計算をすべて NPU の 32 個の AI エンジンで行います。同じ大きさの int4 のモデルよりダウンロードもメモリも小さく、生成も速くなります。画像のエンコーダー（ViT）は CPU で動きます。NPU 向けのプログラムはビルド済みなので、PC の上でコンパイルを待つことはありません。 （onwx 0.7.0、Ryzen AI Max+ 395（Windows）で NPU の電力モードを performance にして測定。Ryzen AI 300 シリーズ（HX 370 など）も NPU は同じ 32 個の AI エンジンで、同じ方式で動きます。標準の電力モードでは NPU のクロックが下がって 2 割ほど遅くなります。電力モードは管理者の PowerShell で C:\Windows\System32\AMD\xrt-smi.exe configure --pmode performance。Windows の Copilot+ の AI 機能が NPU… Open weights apache-2.0 262,144 tokens [View model](https://savrn.com/models/ternary-bonsai-2-9b-onwx) Model · Image and text to text ### [Tess-4.5-27B-exl3](https://savrn.com/models/tess-4-5-27b-exl3) [Everson Alves da Silva](https://savrn.com/model-publishers/khronnuz) Open weights apache-2.0 [View model](https://savrn.com/models/tess-4-5-27b-exl3) Model · Image and text to text ### [test-time-co-evolution-9B](https://savrn.com/models/test-time-co-evolution-9b) [YangHe](https://savrn.com/model-publishers/hahayhe) Open weights apache-2.0 9.4B parameters transformers [View model](https://savrn.com/models/test-time-co-evolution-9b) Model · Image and text to text ### [ThinkingCap-Qwen3.8-27B-exl3](https://savrn.com/models/thinkingcap-qwen3-8-27b-exl3) [Everson Alves da Silva](https://savrn.com/model-publishers/khronnuz) EXL3 quant of ThinkingCap-Qwen3.8-27B, a finetune of Qwen/Qwen3.8-27B. Requires ExLlamaV3 1.5.1. The bit maps match turboderp's Qwen3.8-27B self-calibrated packs. This architecture is dense. There is no n-gram table, and -hq does not raise any layer. - 4.00 bpw: -b 4 -hb 5 -mb 4 -vb 6. Decoder layers are 4 bits. The output head is 5 bits. The MTP block is 4 bits. The vision tower is 6 bits. - 3.00 bpw: -b 3 -hb 4 -mb 3 -vb 4. Decoder layers are 3 bits. The output head is 4 bits. The MTP block is 3 bits. The vision tower is 4 bits. - 5.00 bpw: -b 5 -hb 6 -mb 4 -vb 6. Decoder layers are 5 bits. The output head is 6 bits. The MTP block is 4 bits. The vision tower is 6 bits. - 6.00 bpw: -b 6… Open weights other 262,144 tokens [View model](https://savrn.com/models/thinkingcap-qwen3-8-27b-exl3) Model · Image and text to text ### [Valen-0.8B-ONNX](https://savrn.com/models/valen-0-8b-onnx) [Valen](https://savrn.com/model-publishers/valen-team) This is the ONNX version of Valen-Team/Valen-0.8B. Qwen3.5 with a two-layer MLP-Mixer decision head. Supports text, images and videos, and returns Choice, Noul and Score decisions. Multiple questions can share one state encoding with execution="sharedstate". Trained on millions of samples. Install Python 3.10+, PyTorch, Transformers 5.4.0, torchvision, Pillow and av. Flash Attention 2 is optional with a compatible CUDA build. The model contains its tokenizer, processor and custom inference code; a separate base-model download is unnecessary. Use attnimplementation="flashattention2" for Flash Attention, or execution="question" for independent questions. Video defaults to 16 frames.… Open weights apache-2.0 262,144 tokens transformers [View model](https://savrn.com/models/valen-0-8b-onnx) Model · Image and text to text ### [vectrayx-vision-1b-checks](https://savrn.com/models/vectrayx-vision-1b-checks) [Juan Salas Santillana](https://savrn.com/model-publishers/jsantillana) This repository contains the model described in the paper VectraYX-Vision-1B: A Sub-2B Spanish/LATAM Cybersecurity Vision-Language Model with Structured Visual Reasoning and Native Tool Use. VectraYX-Vision-1B is a sub-2B vision-language model specialized for Spanish/LATAM cybersecurity imagery. It couples a frozen SigLIP-so400m encoder to a 1.04B Spanish/LATAM security decoder via an MLP, and is designed to answer in Spanish, emit structured reasoning via native tokens, and invoke tools via the Model Context Protocol ( ). Open weights [View model](https://savrn.com/models/vectrayx-vision-1b-checks) Model · Image and text to text ### [ViSkill-FrozenLake](https://savrn.com/models/viskill-frozenlake) [Hongxing Li](https://savrn.com/model-publishers/hongxingli) This repository contains ViSkill-FrozenLake and its visual skill library, introduced in ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills. ViSkill-FrozenLake is a visual agent model for FrozenLake, built upon Qwen2.5-VL-3B-Instruct. It represents successful interactions as reusable visual skill cards. Geometry-aware retrieval guides the agent, while PPO with skill-guided rewards and online skill distillation jointly improve the policy and skill library. Model files are stored at the repository root, with the accompanying visual skill library in skilllibrary/. For the full ViSkill agent pipeline, including skill retrieval and environment interaction, please refer to our code… Open weights apache-2.0 4.1B parameters 128,000 tokens transformers [View model](https://savrn.com/models/viskill-frozenlake) Model · Image and text to text ### [ViSkill-Sokoban](https://savrn.com/models/viskill-sokoban) [Hongxing Li](https://savrn.com/model-publishers/hongxingli) This repository contains ViSkill-Sokoban and its visual skill library, introduced in ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills. ViSkill-Sokoban is a visual agent model for Sokoban, built upon Qwen2.5-VL-3B-Instruct. It represents successful interactions as reusable visual skill cards. Geometry-aware retrieval guides the agent, while PPO with skill-guided rewards and online skill distillation jointly improve the policy and skill library. Model files are stored at the repository root, with the accompanying visual skill library in skilllibrary/. For the full ViSkill agent pipeline, including skill retrieval and environment interaction, please refer to our code… Open weights 4.1B parameters 128,000 tokens transformers [View model](https://savrn.com/models/viskill-sokoban) Model · Image and text to text ### [visual-jev-120k-qwen35-0.8b-lora](https://savrn.com/models/visual-jev-120k-qwen35-0-8b-lora) [Harsh Nandwana](https://savrn.com/model-publishers/harshnandwana) This PEFT adapter was trained on a 120,000-row stratified subset, not the entire training split, of Visual Jev decisions v1 at revision 687c745c34846d104ee85af802b9fd444a854f5d. It answers a task-specific question about an image with a normalized box, choice, or short text. The two published prediction logs contain the same 1,000 held-out test IDs. The 120k adapter improved all five selected-subset scores, including mean grounding IoU from 0.530 to 0.598, box choice accuracy from 94.0% to 97.5%, and spatial decision accuracy from 90.0% to 94.5%. This is not a full test-split benchmark. Modal's billing report measured $8.23 metered for the 120k staging, training, benchmark, and automatic… Open weights apache-2.0 peft [View model](https://savrn.com/models/visual-jev-120k-qwen35-0-8b-lora) Model · Image and text to text ### [visual-jev-budget20-qwen35-0.8b-lora](https://savrn.com/models/visual-jev-budget20-qwen35-0-8b-lora) [Harsh Nandwana](https://savrn.com/model-publishers/harshnandwana) This PEFT adapter was trained on a 40,000-row stratified subset, not the entire training split, of Visual Jev decisions v1 at revision 687c745c34846d104ee85af802b9fd444a854f5d. It answers a task-specific question about an image with a normalized box, choice, or short text. The GitHub composite shows selected held-out predictions with annotated photos. Its companion JSON retains full predictions. Published train and evaluation questions largely use five fixed templates. COCO instance annotations supply grounding, box choice, and left/right questions; Visual Genome v1.2 supplies selected color and relation labels. Objects, boxes, and targets vary by photo, and a photo yields only supported… Open weights apache-2.0 peft [View model](https://savrn.com/models/visual-jev-budget20-qwen35-0-8b-lora) Model · Image and text to text ### [VLX-Seek-1.5-10B-GGUF](https://savrn.com/models/vlx-seek-1-5-10b-gguf) [Asher](https://savrn.com/model-publishers/asher-1) Open weights apache-2.0 ggml [View model](https://savrn.com/models/vlx-seek-1-5-10b-gguf) Model · Image and text to text ### [ward-network-14b](https://savrn.com/models/ward-network-14b) [Rojas](https://savrn.com/model-publishers/dennis1315) This qwen35 model was trained 2x faster with Unsloth and Huggingface's TRL library. Open weights apache-2.0 262,144 tokens transformers [View model](https://savrn.com/models/ward-network-14b) Model · Image and text to text ### [XORTRON-CriminalComputing-RICO-v4](https://savrn.com/models/xortron-criminalcomputing-rico-v4) [Sonny DeSorbo](https://savrn.com/model-publishers/darkc0de) To achieve optimal performance, use the following settings: Thinking: temperature=1.0, topp=0.95, topk=20, minp=0.0, presencepenalty=0.0, repetitionpenalty=1.0 Instruct: temperature=0.7, topp=0.80, topk=20, minp=0.0, presencepenalty=1.5, repetitionpenalty=1.0 This model is a part of The XORTRON Criminal Computing project; an ongoing research experiment and exercise in AI safety and alignment. XORTRON models are intentionally developed to study, evaluate, and document the potential of advanced AI systems to facilitate real-world criminal activity, abuse, and other high-risk conduct. Because these models may produce unusually capable, operationally relevant, or otherwise sensitive outputs… Open weights 27.8B parameters 262,144 tokens transformers [View model](https://savrn.com/models/xortron-criminalcomputing-rico-v4) Model · Image and text to text ### [ZDTaichu5.0-9B-GGUF](https://savrn.com/models/zdtaichu5-0-9b-gguf) [Zidongtaichu](https://savrn.com/model-publishers/taichuai) ZDTaichu5.0-9B is a multimodal foundation model for general visual understanding, spatial reasoning, agentic tool use, and embodied-AI research. It combines a Qwen3.5-9B language backbone with a C-RADIOv4-H vision encoder, supports text, images and videos with any-resolution visual input. Within the 10B-scale general-purpose VLMs compared in this release blog, ZDTaichu5.0-9B retains first-tier general visual understanding while supporting spatial reasoning, high-level embodied VLM reasoning, and agent tasks under the reported evaluation settings. Rather than trading broad visual competence for specialization, it layers a more comprehensive spatial, embodied, and agent capability profile on… Open weights [View model](https://savrn.com/models/zdtaichu5-0-9b-gguf) ## Questions ### Which Image and text to text models are most downloaded? By monthly downloads reported by the Hugging Face Hub: . ## Other Tasks - [Text Generation](https://savrn.com/models/tasks/text-generation) 1548 - [Feature Extraction](https://savrn.com/models/tasks/feature-extraction) 398 - [Robotics](https://savrn.com/models/tasks/robotics) 398 - [Text Classification](https://savrn.com/models/tasks/text-classification) 331 - [Text to Image](https://savrn.com/models/tasks/text-to-image) 232 - [Speech Recognition](https://savrn.com/models/tasks/speech-recognition) 187 - [Text to Speech](https://savrn.com/models/tasks/text-to-speech) 145 - [Image Classification](https://savrn.com/models/tasks/image-classification) 139 - [Audio Classification](https://savrn.com/models/tasks/audio-classification) 130 - [Sentence Similarity](https://savrn.com/models/tasks/sentence-similarity) 103 - [Reinforcement Learning](https://savrn.com/models/tasks/reinforcement-learning) 100 - [Token Classification](https://savrn.com/models/tasks/token-classification) 100 [See all](https://savrn.com/models/tasks)