sleepy/llama.cpp
1
0
Fork 0
Code Issues 11 Pull Requests Actions 109 Packages Projects Releases Wiki Activity

model: support Ministral3 (#17644)

Browse Source 
* conversion script

* support ministral 3

* maybe this is better?

* add TODO for rope_yarn_log_mul

* better ppl (tested on 14B-Instruct)

* Add Ministral3 support to Mistral format

* improve arch handling

* add sizes

* Apply suggestions from code review

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* nits

---------

Co-authored-by: Julien Denize <julien.denize@mistral.ai>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Xuan-Son Nguyen
2025-12-01 12:26:52 +01:00
committed by GitHub
parent 649495c9d9
commit cd3c118908
11 changed files with 342 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
gguf-py/gguf/constants.py
+23
View File
@@ -175,6 +175,7 @@ class Keys:
VALUE_LENGTH_MLA = "{arch}.attention.value_length_mla"
SHARED_KV_LAYERS = "{arch}.attention.shared_kv_layers"
SLIDING_WINDOW_PATTERN = "{arch}.attention.sliding_window_pattern"
TEMPERATURE_SCALE = "{arch}.attention.temperature_scale"
class Rope:
DIMENSION_COUNT = "{arch}.rope.dimension_count"
@@ -444,6 +445,7 @@ class MODEL_ARCH(IntEnum):
MINIMAXM2 = auto()
RND1 = auto()
PANGU_EMBED = auto()
MISTRAL3 = auto()
class VISION_PROJECTOR_TYPE(IntEnum):
@@ -817,6 +819,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.COGVLM: "cogvlm",
MODEL_ARCH.RND1: "rnd1",
MODEL_ARCH.PANGU_EMBED: "pangu-embedded",
MODEL_ARCH.MISTRAL3: "mistral3",
}
VISION_PROJECTOR_TYPE_NAMES: dict[VISION_PROJECTOR_TYPE, str] = {
@@ -3071,6 +3074,26 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.MISTRAL3: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
# TODO
}
gguf-py/gguf/gguf_writer.py
+3
View File
@@ -904,6 +904,9 @@ class GGUFWriter:
def add_attn_temperature_length(self, value: int) -> None:
self.add_uint32(Keys.Attention.TEMPERATURE_LENGTH.format(arch=self.arch), value)
def add_attn_temperature_scale(self, value: float) -> None:
self.add_float32(Keys.Attention.TEMPERATURE_SCALE.format(arch=self.arch), value)
def add_pooling_type(self, value: PoolingType) -> None:
self.add_uint32(Keys.LLM.POOLING_TYPE.format(arch=self.arch), value.value)