UPSTREAM PR #18399: Add pre-tokenizer hash for MiniMax-M2.1

[loci-dev]

Mirrored from ggml-org/llama.cpp#18399

Adds pre-tokenizer hash for MiniMaxAI/MiniMax-M2.1.

The model uses the same architecture (MiniMaxM2ForCausalLM) and tokenizer type as MiniMax-M2, but requires an additional hash entry due to differences in how the tokenizer is processed.

Without this fix, conversion fails with:

WARNING: The BPE pre-tokenizer was not recognized!
NotImplementedError: BPE pre-tokenizer was not recognized - update get_vocab_base_pre()

Tested conversion successfully: https://huggingface.co/AaryanK/MiniMax-M2.1-GGUF

[loci-agentic-ai]

Explore the complete analysis inside the Version Insights

I've generated a summary report for your project. Here are the key findings:

Summary Report for llama.cpp PR #715

Project Information:

• Repository: llama.cpp (owner: auroralabs-loci)
• Pull Request: UPSTREAM PR #18399: Add pre-tokenizer hash for MiniMax-M2.1 #715
• Comparing versions: 1b37d0a3-2d73-4ca4-bc16-73c34a215de1 (base) → f4629965-8bc1-43c7-beb7-94074ad4cf49 (target)

Performance Analysis Results:

✅ No significant performance changes detected - No modified functions show performance changes greater than 2% in either response time or throughput time.

Conclusion:
This pull request maintains performance stability and appears safe to merge from a performance perspective. The changes do not introduce any measurable performance regressions.