Where is description of model config file format?

[Expro]

I want to enable flash attention KV quantization for one of models. Where is YAML format for model configuration described?

[DevrMichael]

@Expro

• The YAML model config format is documented in the docs under “Full config model file reference”: Advanced usage → Full config model file reference.
• The source of truth is the code that defines the schema: core/config/model_config.go (see ModelConfig and embedded LLMConfig). That’s where all YAML keys come from.
• For flash-attention and KV cache quantization, use:

flash_attention
cache_type_k
cache_type_v
no_kv_offloading

Example (llama.cpp backend):
name: my-llama
backend: llama.cppparameters:  
model: my-model.gguf
flash_attention: "fa2"      # value per your llama.cpp build
cache_type_k: "q8_0"
cache_type_v: "q8_0"
no_kv_offloading: false

Note: Accepted values for flash_attention and cache_type_* are the same as llama.cpp CLI flags; check the llama.cpp documentation for the exact strings supported by your build.