Skip to content

Conversation

saood06
Copy link
Collaborator

@saood06 saood06 commented May 9, 2025

This ports ggml-org/llama.cpp#9396 and ggml-org/llama.cpp#9117 (I don't think I needed this as the changes in here are basically reverted in 9396).

The issue was that the convert script used generate_extra_tensors for those tensors but there was no code that called that function.

I tested with Llama-3_1-Nemotron-51B-Instruct and it now generates the rope_freqs.weight which was missing previously.

Look at #377 for more information.

ngxson and others added 3 commits May 9, 2025 06:17
This should also fix vocab-only conversion for Phi-3.
MiniCPM3's tokenizer is treated as a SentencePiece tokenizer to avoid
having to run its custom Python code which mixes tokenization
in the same file as tool calls.

gguf-py : add long and short RoPE factors to tensor mappings

Empty, but the key names are used to populate the mappings.
@saood06 saood06 requested a review from ikawrakow May 9, 2025 12:23
@saood06 saood06 merged commit 967a2e1 into main May 9, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants