bartowski/Qwen3.8-27B-GGUF · Hugging Face - Updated (Per-tensor layout)

Social·Community

bartowski adds per-tensor layout maps to GGUF quantization uploads

A long-time GGUF uploader announced a blog post on Hugging Face covering research into new tensor type layouts and upcoming changes to the shape of models they publish. The post affects users who download their quantized model uploads.

A long-time GGUF uploader announced a blog post on Hugging Face covering research into new tensor type layouts and upcoming changes to the shape of models they publish. The post affects users who download their quantized model uploads.

Unsloth announced its models have surpassed 500 million downloads on Hugging Face. Qwen3.8-27B GGUF has become the organization's most-downloaded model ever.

Releases·Tool Release

Two new quantization releases target Qwen3.8-27B: GSQ-RCO GGUFs and QUASAR NVFP4

Two independent teams released aggressively quantized versions of Qwen3.8-27B within two days. One ships GGUF checkpoints at 2.5-3.0 bits per weight using two new methods, GSQ and RCO, with both the search and quantizer learned. The other releases an NVFP4 W4A4 checkpoint trained via quantization-aware distillation with the QUASAR QAT algorithm, deployable on vLLM with NVIDIA Blackwell GPUs. Both matter to practitioners running large models at low precision locally or on modern accelerators.

Two independent teams released aggressively quantized versions of Qwen3.8-27B within two days. One ships GGUF checkpoints at 2.5-3.0 bits per weight using two new methods, GSQ and RCO, with both the search and quantizer learned. The other releases an NVFP4 W4A4 checkpoint trained via quantization-aware distillation with the QUASAR QAT algorithm, deployable on vLLM with NVIDIA Blackwell GPUs. Both matter to practitioners running large models at low precision locally or on modern accelerators.

Developments·Paper

WarpQuant: INT3 PTQ for Qwen3.8-27B with Hadamard rotation and Output-Fisher recovery

WarpQuant is a new post-training quantization method that separates the compression domain from the recovery domain, achieving 3.6165 text bpw on Qwen3.8-27B. The release provides loadable safetensors with quantized values in BF16, though it is not yet Ollama-ready GGUF.

WarpQuant is a new post-training quantization method that separates the compression domain from the recovery domain, achieving 3.6165 text bpw on Qwen3.8-27B. The release provides loadable safetensors with quantized values in BF16, though it is not yet Ollama-ready GGUF.

A model card on Hugging Face for Qwen/Qwen3.8–2.4T-A95B lists a 2.4-trillion-parameter count, which appears to be a typo. The card was fetched on August 16, 2026, and the discrepancy has drawn attention.