llama : register the DT3 file type

Adds LLAMA_FTYPE_MOSTLY_DT3 at the end of the ftype enum, the loader
name/guess mappings, the quantization fallbacks (same as the other
ternary types), and the llama-quantize table entry. The table entry
warns that the in-tree quantizer is only the reference one: DT3 models
with the measured quality are produced by the external PTQTP pipeline.
This commit is contained in:
Millaguie
2026-08-10 23:33:14 +02:00
parent 698f37f40b
commit 693eb7d719
4 changed files with 8 additions and 2 deletions
+1
View File
@@ -156,6 +156,7 @@ extern "C" {
LLAMA_FTYPE_MOSTLY_NVFP4 = 39, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q1_0 = 40, // except 1d tensors
LLAMA_FTYPE_MOSTLY_Q2_0 = 41, // except 1d tensors
LLAMA_FTYPE_MOSTLY_DT3 = 42, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};