LLM inference in C/C++
C++ ggml
convert : support text-only checkpoints in pplx-decider (#30575)
FINAL-Bench0f4376268f2e
diff --git a/conversion/pplx_decider.py b/conversion/pplx_decider.pyindex f9cfaf9d3d..b71bd8a3ad 100644--- a/conversion/pplx_decider.py+++ b/conversion/pplx_decider.py@@ -3,7 +3,7 @@ from __future__ import annotationsimport jsonfrom pathlib import Path-from typing import Any, Callable, Iterable, TYPE_CHECKING+from typing import Any, Iterable, TYPE_CHECKINGimport 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")[email protected]("FINAL-Bench/Darwin-27B-ZTC-v5")class PplxDeciderModel(Qwen3_5TextModel):model_arch = gguf.MODEL_ARCH.QWEN35no_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 tokensreadout = 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