From: Sigbjørn Skjæret <redacted>
Date: Mon, 24 Nov 2025 14:50:55 +0000 (+0100)
Subject: convert : allow quantizing lora again (#17453)
X-Git-Tag: upstream/0.0.7446~299
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=b61de2b2df4ff07e6d6de96320fb311d96908b7a;p=pkg%2Fggml%2Fsources%2Fllama.cpp

convert : allow quantizing lora again (#17453)
---

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 6cbaee03..d24a4682 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -565,7 +565,7 @@ class ModelBase:
                             gguf.MODEL_TENSOR.ALTUP_PREDICT_COEF,
                         )
                     )
-                    or not new_name.endswith(".weight")
+                    or new_name[-7:] not in (".weight", ".lora_a", ".lora_b")
                 ):
                     data_qtype = gguf.GGMLQuantizationType.F32
 
diff --git a/convert_lora_to_gguf.py b/convert_lora_to_gguf.py
index 57c6cd0d..b0adde8a 100755
--- a/convert_lora_to_gguf.py
+++ b/convert_lora_to_gguf.py
@@ -242,7 +242,7 @@ def parse_args() -> argparse.Namespace:
         help="path to write to; default: based on input. {ftype} will be replaced by the outtype.",
     )
     parser.add_argument(
-        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0", "auto"], default="f16",
+        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0", "auto"], default="f32",
         help="output format - use f32 for float32, f16 for float16, bf16 for bfloat16, q8_0 for Q8_0, auto for the highest-fidelity 16-bit float type depending on the first loaded tensor type",
     )
     parser.add_argument(