Misc. bug: seq_rm does not support seq_id < 0, documentation claiming the value is supported

[MrJackSpade]

Name and Version

Applicable on Commit (fd1234c)

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

libllama (core library)

Command line

Problem description & steps to reproduce

 // Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
 // Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
 // seq_id < 0 : match any sequence
 // p0 < 0     : [0,  p1]
 // p1 < 0     : [p0, inf)
 LLAMA_API bool llama_memory_seq_rm(
         llama_memory_t mem,
           llama_seq_id seq_id,
              llama_pos p0,
              llama_pos p1);

Above documentation claims that seq_id < 0 can be provided to clear all sequences in a position, however this fails due to the following ASSERT at runtime

bool llama_kv_cache_unified::seq_rm(llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
    GGML_ASSERT(seq_id >= 0 && (size_t) seq_id < seq_to_stream.size());

    auto & cells = v_cells[seq_to_stream[seq_id]];
    auto & head  = v_heads[seq_to_stream[seq_id]];

    uint32_t new_head = cells.size();

First Bad Commit

This ASSERT appears to have been added as part of (225e7a1)

Relevant log output

[ggerganov]

@MrJackSpade Can you confirm that #15226 fixes the issue?

[MrJackSpade]

@ggerganov I can confirm that the assert failure no longer occurs when using -1 as a parameter.