vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-11-01 11:21:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

convert : support text-only checkpoints in pplx-decider (#30575)

Browse source 
* pplx-decider-text-only

convert : support text-only checkpoints in pplx-decider

* Apply suggestions from code review

* convert : remove unused import
This commit is contained in:
FINAL-Bench 2026-10-16 15:56:00 +09:00 • committed by GitHub
parent 3dbcb7fca5
commit 0f4376268f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 3 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

13
conversion/pplx_decider.py
View file

@ -3,7 +3,7 @@ from __future__ import annotations
import json
from pathlib import Path
from typing import Any, Callable, Iterable, TYPE_CHECKING
from typing import Any, Iterable, TYPE_CHECKING
import torch
@ -35,6 +35,7 @@ def _load_pplx_decider_hparams(dir_model: Path) -> dict[str, Any]:
@ModelBase.register("PplxDeciderModel")
@ModelBase.example("perplexity-ai/pplx-decider-v1-27b")
@ModelBase.example("FINAL-Bench/Darwin-27B-ZTC-v5")
class PplxDeciderModel(Qwen3_5TextModel):
model_arch = gguf.MODEL_ARCH.QWEN35
no_mtp = True # the checkpoint has no MTP head
@ -77,14 +78,6 @@ class PplxDeciderModel(Qwen3_5TextModel):
for name in ("choice", "score", "noul"):
self.gguf_writer.add_decision_temperature(name, self.hparams["decision"]["temperature"])
@classmethod
def filter_tensors(cls, item: tuple[str, Callable[[], Tensor]]) -> tuple[str, Callable[[], Tensor]] | None:
name, gen = item
# the checkpoint is the bare backbone, its text tensors have no "model." prefix
if name.startswith("language_model."):
name = "model." + name
return super().filter_tensors((name, gen))
def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
yield from super().generate_extra_tensors()
from safetensors.torch import load_file
@ -92,7 +85,7 @@ class PplxDeciderModel(Qwen3_5TextModel):
# the readout has one row per option label, store it as an LM head that is zero for the other tokens
readout = load_file(self.dir_model / "readout.safetensors")["weight"]
token_ids = self.hparams["decision"]["token_ids"]
n_vocab = self.hparams["text_config"]["vocab_size"]
n_vocab = self.hparams["vocab_size"]
assert readout.shape[0] == len(token_ids) == len(set(token_ids))
lm_head = torch.zeros(n_vocab, readout.shape[1], dtype=readout.dtype)
lm_head[token_ids] = readout