CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-08 10:07:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

falcon-h1 specefic vocab resolved

Browse Source
This commit is contained in:
ibrahimkhadraoui
2025-07-07 10:25:57 +04:00
parent 7a25441e13
commit 280dd2dcb7
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
convert_hf_to_gguf.py
View File

@@ -4882,9 +4882,6 @@ class Mamba2Model(TextModel):
pad_vocab = self.hparams.get("pad_vocab_size_multiple", 16) pad_vocab = self.hparams.get("pad_vocab_size_multiple", 16)
# pad using ceiling division # pad using ceiling division
# ref: https://stackoverflow.com/a/17511341/22827863 # ref: https://stackoverflow.com/a/17511341/22827863
# if architecture is FalconH1, don't pad vocab size
if self.hparams.get("architectures", [None])[0] == "FalconH1ForCausalLM":
pad_vocab = 1
vocab_size = -(vocab_size // -pad_vocab) * pad_vocab vocab_size = -(vocab_size // -pad_vocab) * pad_vocab
self.hparams["vocab_size"] = vocab_size self.hparams["vocab_size"] = vocab_size
@@ -6590,6 +6587,9 @@ class FalconH1Model(Mamba2Model):
keys = list(keys) + prefixed keys = list(keys) + prefixed
return super().find_hparam(keys, *args, **kwargs) return super().find_hparam(keys, *args, **kwargs)
def set_vocab(self):
self._set_vocab_gpt2()
def _generate_mup_vector(self, block_id: int) -> torch.Tensor: def _generate_mup_vector(self, block_id: int) -> torch.Tensor:
zxbcdt_multipliers = self.hparams["ssm_multipliers"] zxbcdt_multipliers = self.hparams["ssm_multipliers"]
intermediate_size = self.hparams["mamba_d_ssm"] intermediate_size = self.hparams["mamba_d_ssm"]