Skip to content

convert : fix context length for nomic-embed-text-v2-moe #13216

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
May 2, 2025
Merged

convert : fix context length for nomic-embed-text-v2-moe #13216

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -504,7 +504,7 @@ def prepare_metadata(self, vocab_only: bool):
def set_gguf_parameters(self):
self.gguf_writer.add_block_count(self.block_count)

if (n_ctx := self.find_hparam(["max_position_embeddings", "n_ctx"], optional=True)) is not None:
if (n_ctx := self.find_hparam(["max_position_embeddings", "n_ctx", "n_positions"], optional=True)) is not None:
self.gguf_writer.add_context_length(n_ctx)
logger.info(f"gguf: context length = {n_ctx}")

Expand Down Expand Up @@ -3523,8 +3523,13 @@ def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path,
if self._tokenizer_is_xlmroberta:
self._xlmroberta_tokenizer_init()

# the HF config claims n_ctx=8192, but it uses RoPE scaling
self.hparams["n_ctx"] = 2048
npos, mtp = self.hparams["n_positions"], self.hparams.get("max_trained_positions", 2048)
if npos == 8192 and mtp == 2048:
self.hparams["n_positions"] = 2048 # nomic-embed-text v1 and v1.5 are trained for 2048 tokens.
elif npos == 2048 and mtp == 2048:
self.hparams["n_positions"] = 512 # nomic-embed-text-v2-moe is trained for 512 tokens.
else:
raise ValueError(f"unrecognized parameters: n_positions={npos}, max_trained_positions={mtp}")

assert self.hparams["activation_function"] == "gelu" if self.is_moe else "swiglu"

Expand Down