millaguie/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

llama : register the DT3 file type

Browse Source 
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
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model-loader.cpp
+2
View File
@@ -58,6 +58,7 @@ const char * llama_ftype_name(llama_ftype ftype) {
case LLAMA_FTYPE_MOSTLY_Q6_K: name = LLAMA_FTYPE_PREFIX "Q6_K"; break;
case LLAMA_FTYPE_MOSTLY_TQ1_0: name = LLAMA_FTYPE_PREFIX "TQ1_0 - 1.69 bpw ternary"; break;
case LLAMA_FTYPE_MOSTLY_TQ2_0: name = LLAMA_FTYPE_PREFIX "TQ2_0 - 2.06 bpw ternary"; break;
case LLAMA_FTYPE_MOSTLY_DT3: name = LLAMA_FTYPE_PREFIX "DT3 - 3.5 bpw dual-plane ternary"; break;
case LLAMA_FTYPE_MOSTLY_IQ2_XXS: name = LLAMA_FTYPE_PREFIX "IQ2_XXS - 2.0625 bpw"; break;
case LLAMA_FTYPE_MOSTLY_IQ2_XS: name = LLAMA_FTYPE_PREFIX "IQ2_XS - 2.3125 bpw"; break;
case LLAMA_FTYPE_MOSTLY_IQ2_S: name = LLAMA_FTYPE_PREFIX "IQ2_S - 2.5 bpw"; break;
@@ -747,6 +748,7 @@ llama_model_loader::llama_model_loader(
case GGML_TYPE_Q6_K: ftype = LLAMA_FTYPE_MOSTLY_Q6_K; break;
case GGML_TYPE_TQ1_0: ftype = LLAMA_FTYPE_MOSTLY_TQ1_0; break;
case GGML_TYPE_TQ2_0: ftype = LLAMA_FTYPE_MOSTLY_TQ2_0; break;
case GGML_TYPE_DT3: ftype = LLAMA_FTYPE_MOSTLY_DT3; break;
case GGML_TYPE_IQ2_XXS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XXS; break;
case GGML_TYPE_IQ2_XS: ftype = LLAMA_FTYPE_MOSTLY_IQ2_XS; break;
case GGML_TYPE_IQ2_S: ftype = LLAMA_FTYPE_MOSTLY_IQ2_S; break;