From the model card
Fields below are copied from the tags and counters on the HuggingFace repository sakamakismile/Qwen3.6-27B-Text-NVFP4-MTP at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- sakamakismile
- Pipeline tag
- text-generation
- Library
- Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Lineage
-
- base model Qwen/Qwen3.6-27B
- quantized from Qwen/Qwen3.6-27B
- Language tags
- English (en), Chinese (zh), Japanese (ja), Korean (ko), French (fr), German (de), Spanish (es), Italian (it), Portuguese (pt), Russian (ru), Arabic (ar)
- Downloads (HF counter at last fetch)
- 974,747
- Likes (HF counter at last fetch)
- 81
- Model card
- https://huggingface.co/sakamakismile/Qwen3.6-27B-Text-NVFP4-MTP
Use cases
- High-throughput text-only inference on H100/B100 GPUs
- Speculative decoding experiments using MTP head
- Multilingual text generation server across 13 supported languages
- Cost-optimized alternative to full multimodal Qwen3.6 deployment
Pros
- Text-only path skips vision encoder overhead for pure text workloads
- NVFP4 provides hardware-level efficiency on Hopper/Blackwell
- MTP speculative decoding can increase effective throughput
- Apache-2.0 license; 13-language coverage
Cons
- Vision capability is entirely absent — no image understanding
- Requires Hopper or Blackwell GPU; unusable on older hardware
- NVFP4 tooling (NIM/TensorRT-LLM) is more complex than standard Transformers
- 8-bit tag suggests mixed precision; actual NVFP4 behavior needs verification
Tags
transformerssafetensorsqwen3_5image-text-to-textqwen3.6nvfp4quantizedmodeloptmtpspeculative-decodingblackwelltext-onlytext-generationconversationalenzhjakofrde