The image-editing prompt rewriter for Qwen-Image-2.1, a fine-tuned Qwen3.5-VL 9B that turns a short edit instruction plus 1–N input images into a detailed English edit prompt, with its refusal behaviour removed by Heretic directional ablation. bf16, same shapes and parameter count as the source; nothing else was changed. systemprompt.txt is included and required. It defines the output format. It is the unmodified file from the source repo. The second row is an independent evaluation of the exported weights with Heretic's evaluatemodel. Refusals were measured on mlabonne/harmfulbehaviors and KL divergence (damage to ordinary behaviour, first-token distributions) on mlabonne/harmlessalpaca…
Open-weight model · Image and text to text
Qwen-Image-2.1-PE-I2I-Heretic-NVFP4
by Darrellbest darrellbest/Qwen-Image-2.1-PE-I2I-Heretic-NVFP4
Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 is an open-weight model for image and text to text from Darrellbest, released under other. It has 9.4B parameters and a 262,144-token context. At 16-bit it needs about 22.6 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 29 downloads a month.
NVFP4 (4-bit floating point, W4A4) build of darrellbest/Qwen-Image-2.1-PE-I2I-Heretic, the refusal-ablated image-editing prompt rewriter for Qwen-Image-2.1. For vLLM on NVIDIA Blackwell GPUs, which run NVFP4 natively. 11 GB instead of 18 GB.
Parameters9.4B
Context262,144
Weights11.2 GB
Licenseother
AccessOpen weights
Monthly Downloads29
Runs On
What it takes to serve Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 (9.4B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 18.8 GB | 22.6 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 9.4 GB | 11.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 4.7 GB | 5.6 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 24, 2026.
Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 on every accelerator the SAVRN Index prices, at every precision
Model Card
NVFP4 (4-bit floating point, W4A4) build of darrellbest/Qwen-Image-2.1-PE-I2I-Heretic, the refusal-ablated image-editing prompt rewriter for Qwen-Image-2.1. For vLLM on NVIDIA Blackwell GPUs, which run NVFP4 natively. 11 GB instead of 18 GB. systemprompt.txt is included and required, exactly as for the original. The linear-attention layers carry a recurrent state and the vision tower encodes the input image; both were left in bf16, as other quantizations of this model family do. That is why the file is 11 GB rather than ~6 GB. Made with llm-compressor 0.13.0 (QuantizationModifier, scheme="NVFP4"), calibrated on 64 samples in the model's real input format: its own system prompt, an edit…
Excerpt from the card by Darrellbest, licensed other.
Configuration
- Architecture
- Qwen3_5ForConditionalGeneration
- Context length (tokens)
- 262,144
- Layers
- 32
- Hidden size
- 4,096
- Feed-forward size
- 12,288
- Attention heads
- 16
- Key/value heads
- 4
- Head dimension
- 256
- Vocabulary size
- 248,320
- Model type
- qwen3_5
- Quantization
- compressed-tensors
Identity and Version
- Repository
- darrellbest/Qwen-Image-2.1-PE-I2I-Heretic-NVFP4
- Publisher
- Darrellbest
- Task
- Image and text to text
- Modality
- Image and text
- Library
- Not stated by the source
- Parameters
- 9.4B parameters
- Languages
- Not stated by the source
- Revision
- fa02256ffed1f356d5701a7a7576b175d1392568
- First published
- 2026-09-22
- Last updated
- 2026-09-24
Files and Weights
12 files, 11.2 GB in total. The weights are 1 file totalling 11.2 GB in safetensors.
Weights1 file · 11.2 GB
Configuration4 files · 20.9 KB
Tokenizer2 files · 20.0 MB
Documentation2 files · 11.4 KB
Other2 files · 26.1 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 11.2 GB | 783555b072a4 |
| config.json | Configuration | 19.2 KB | — |
| generation_config.json | Configuration | 179 B | — |
| processor_config.json | Configuration | 1.3 KB | — |
| recipe.yaml | Configuration | 269 B | — |
| LICENSE | Documentation | 7.8 KB | — |
| README.md | Documentation | 3.5 KB | — |
| chat_template.jinja | Other | 7.8 KB | — |
| system_prompt.txt | Other | 18.3 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 20.0 MB | 87a7830d63fc |
| tokenizer_config.json | Tokenizer | 1.2 KB | — |
License and Download
- License
- other
- Access
- Open weights, no gate
- Download size
- 11.2 GB
Download from Darrellbest
Released by Darrellbest through its official repository on Hugging Face.
Built From
- Derived from darrellbest/Qwen-Image-2.1-PE-I2I-Heretic
- Quantized from darrellbest/Qwen-Image-2.1-PE-I2I-Heretic
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 11.2 GB |
| 16-bit | 18.8 GB |
| 8-bit | 9.4 GB |
| 4-bit | 4.7 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Qwen-Image-2.1-PE-I2I-Heretic-NVFP4
How much GPU memory does Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 need?
About 22.6 GB at 16-bit and 5.6 GB at 4-bit: the weights (9.4B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
What license is Qwen-Image-2.1-PE-I2I-Heretic-NVFP4 released under?
other, as its publisher declares it. Read the license text before commercial use.
What is Qwen-Image-2.1-PE-I2I-Heretic-NVFP4's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
LucidVitality-9B is a little goer. It's a roleplay or creative focused merge of two Qwen3.5 9b variants for people with absolute potatoes, like myself. It marries the improved prose of Darkhn's Qwen3.5-9B-Animus-V13.0 with the lower looping, higher EOS exit, and slightly more coherency (compared to base) from Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED of DavidAU's making. I haven't merged anything for a long time, as it's been a hard time for finetuning. They rarely increase prose quality, often deeply lose intelligence over base (even when tuned for intelligence or agentic). Base model's getting tough to beat. So it was a pleasant surprise to find two models that each…
This model is a fine-tuned version of Qwen/Qwen3.5-9B-Base on the Omni-Edu-70K dataset. The following hyperparameters were used during training: - learningrate: 5e-06 - trainbatchsize: 1 - evalbatchsize: 8 - distributedtype: multi-GPU - numdevices: 8 - gradientaccumulationsteps: 8 - totaltrainbatchsize: 64 - totalevalbatchsize: 64 - lrschedulertype: cosine - lrschedulerwarmupsteps: 0.1 - numepochs: 3.0 - Transformers 5.2.0 - Pytorch 2.10.0 - Datasets 4.0.0 - Tokenizers 0.22.2
Moondream 3.1 is a vision language model with a mixture-of-experts architecture (9B total parameters, 2B active). It delivers state-of-the-art visual reasoning and detection while staying fast and cheap to deploy. Skills include query, detect, point, and caption, all native and all returning structured output. For the full story on what's new — including how we trained it and how it holds up on your own tasks — see the release notes. Photon is Moondream's high-performance inference engine. It runs the model locally on NVIDIA GPUs (Ampere or newer) and Apple Silicon Macs, with the same API as Moondream Cloud. No API key is required to run the base model locally. (You'll only need one to run…
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not applicable. Empty cells (--) indicate scores not…
This repo quantizes the model using data-free quantization technique. As of 2026-02-25, make sure your system has cuda12.8 installed. Then, create a fresh Python environment (e.g. python3.12 venv) and run: Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after…