Skip to content

Eval bug: Missing files/functionality (no way to calibrate) #1

@KGardevoir

Description

@KGardevoir

Name and Version

(This is a cherrypick of the triattention branch onto a more recent version of feature/turboquant-kv-cache)
$ llama-cli --version
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 11911 MiB):
Device 0: NVIDIA RTX 4000 Ada Generation Laptop GPU, compute capability 8.9, VMM: yes, VRAM: 11911 MiB
version: 9078 (4e20d9f84)
built with GNU 11.5.0 for Linux x86_64

Operating systems

Linux

GGML backends

CUDA

Hardware

NVIDIA RTX 4000 Ada

Models

Qwen3.6-35B-A3B-Q4_K_M

Problem description & steps to reproduce

On the head version of triattention there is reference to a flag --triattention-calibrate, neither option exists. In the documentation there is reference to scripts/calibrate-triattention.py and scripts/validate-calibration.py but these are also missing.

First Bad Commit

No response

Relevant log output

Logs

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions