AI Briefing
KOSign in

llama.cpp Adds Multi-Shard Loading and External PLE File Support

·2026.11.10 16:26

Key point

The update adds a --shard argument to combine model fragments from different locations and supports loading PLE files from separate GGUFs for Gemma4, Qwen3.8Next, and GLM5.3Flash.

Details

  • New --shard command-line argument allows loading model shards from multiple locations, enabling the combination of fragments from various providers or different quantizations.
  • Support added for loading PLE (engram) files from a separate GGUF file using the --ple flag, which is an alias for --shard.
  • Conversion support included for building both PLE-less and PLE-only GGUF files, currently working for Gemma4, Qwen3.8Next, and GLM5.3Flash.
  • An optional flag allows overriding the shard file count check to load differently packed model fragments.
inferenceopen-sourceggufllama.cppmodel-loading
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

PyTorchKRllama.cpp v1.0.0, First Major Release in Three and a Half YearsHuggingFace Blogllama.cpp and ggml reach v1.0.0 with stable API and ABI guaranteesRedditMACaQ: Open-source quantizer with custom vLLM kernels and lower low-bit KLDHacker Newsggml v1.0.0 Released with FP8 Support and Backend OptimizationsRedditTontaube V1.1 released: Open-weights TTS with 4-bit CPU support and 70% preference over Fish Audio S2 ProHacker NewsHuawei releases MACaQ, a model-agnostic quantizer for local inference