tags:
- mtp
- text-generation-inference
- transformers
- unsloth
- qwen3_5
- tool-use
- heretic
- uncensored
- decensored
- abliterated
- speculative-decoding
- nvidia-blackwell
- rtx-pro-6000
- SM120
- GB202
- B200
- B300
- FP4
- blackwell
- Qwen3.8
- GPTQ
datasets: - HuggingFaceH4/ultrachat_200k
base_model: - DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored
license: apache-2.0
language: - en
pipeline_tag: image-text-to-text
library_name: transformers
Qwen3.8-27B-Ultimate-Heretic-Uncensored-NVFP4
Qwen3.8-27B-Ultimate-Heretic-Uncensored-NVFP4 is a high-efficiency NVFP4 quantization of DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored calibrated using GPTQ with an
imatrix_mseweight observer (targetingLinearlayers while preservinglm_headandvisualmodules in full precision). Built upon the multi-stage-tuned Qwen 3.8 27B architecture with DavidAU's Turbo Brilliance V2 50X AGI system and Heretic reasoning and instruct frameworks, this compressed variant integrates 27 reasoning modes, 27 instruct modes, and over 15 dynamic AI constructs to deliver low-footprint, high-throughput inference while maintaining deep refusal removal, multilingual capability, and token efficiency across complex creative, coding, and analytical workloads.
[!NOTE]
MTP — Speculative Decoding: Yes | Accelerated FP4 Tensor Core Performance
Quantization Recipe
The model is quantized using the GPTQModifier with the NVFP4 quantization scheme.
default_stage:
default_modifiers:
GPTQModifier:
targets: [Linear]
ignore: [lm_head, 're:.*visual.*']
scheme: NVFP4
weight_observer: imatrix_mse
bypass_divisibility_checks: false
requires_calibration_data: true
block_size: 128
dampening_frac: 0.01
batched_quantization: auto
Model Files
| File Name | File Type | File Size | File Link | Description |
|---|---|---|---|---|
| chat_template.jinja | Jinja Template | 248 kB | Link | Chat template for formatting conversations and prompts. |
| config.json | JSON | 10.3 kB | Link | Core model architecture and configuration parameters. |
| generation_config.json | JSON | 214 Bytes | Link | Default text generation parameters (e.g., temperature, top_p). |
| model-00001-of-00004.safetensors | Safetensors | 5 GB | Link | Model weight shard 1/4. Required to load the model. |
| model-00002-of-00004.safetensors | Safetensors | 4.96 GB | Link | Model weight shard 2/4. Required to load the model. |
| model-00003-of-00004.safetensors | Safetensors | 4.97 GB | Link | Model weight shard 3/4. Required to load the model. |
| model-00004-of-00004.safetensors | Safetensors | 4.78 GB | Link | Model weight shard 4/4. Required to load the model. |
| model.safetensors.index.json | JSON | 278 kB | Link | Index file mapping model weights to their respective shards. |
| model_mtp.safetensors | Safetensors | 849 MB | Link | Multi-Token Prediction (MTP) auxiliary model weights, used to speed up generation. |
| processor_config.json | JSON | 1.22 kB | Link | Configuration for the multimodal processor. |
| recipe.yaml | YAML | 334 Bytes | Link | Configuration file detailing the quantization or training recipe (e.g., NVFP4 parameters). |
| tokenizer.json | JSON | 20 MB | Link | Complete tokenizer vocabulary and encoding rules. |
| tokenizer_config.json | JSON | 1.22 kB | Link | Tokenizer-specific configuration parameters. |
License and Attribution
This model is based on and/or incorporates the following open-source projects and models:
- Qwen3.8-27B (Base): https://huggingface.co/Qwen/Qwen3.8-27B
- Qwen3.8-27B-Ultimate-Heretic-Uncensored: https://huggingface.co/DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored
- llm-compressor: https://github.com/vllm-project/llm-compressor
This model is released under the Apache License 2.0.