microsoft / microsoft/BitNet

When I implemented Llama3-8B-1.58-100B-tokens with TL1 on ARM, a malloc() error was occured.

Open
#143 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
40.3k
Forks
3.7k
PR merge metrics
No merged PRs in 30d

Description

Environment

  • CPU: NVIDA Grace CPU(72thread)
  • Model: Llama3-8B-1.58-100B-tokens with TL1
  • Prompt: AI is going to
  • N_Predict: 128
  • Threads: 1, 2, 4, 8, 16, 32, 64, 72
  • Context Size: 2048
  • Temperature: 0.8

When I implemented Llama3-8B-1.58-100B-tokens with TL1 on ARM, a malloc() error was occured.

  • Error Type
    • Type 1: malloc() error and nothing output
    • Type 2: malloc() error and correct output(generated text)
    • Type 3: double malloc() error and correct output(generated text)
Thread Size Error Message Output Error Type
1thread malloc(): invalid next size (unsorted) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '1', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Nothing Type 1
2thread free(): invalid next size (normal) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '2', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 2
4thread double free or corruption (!prev) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '4', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 3
8thread double free or corruption (!prev) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '8', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 3
16thread double free or corruption (!prev) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '16', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 3
32thread double free or corruption (!prev) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '32', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 3
64thread free(): invalid next size (normal) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '64', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 2
72thread double free or corruption (!prev) Error occurred while running command: Command '['build/bin/llama-cli', '-m', '/root/BitNet/models/Llama3-8B-1.58-100B-tokens/ggml-model-tl1.gguf', '-n', '128', '-t', '72', '-p', 'AI is going to', '-ngl', '0', '-c', '2048', '--temp', '0.8', '-b', '1']' died with <Signals.SIGABRT: 6>. Generated Text Type 3

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Reproduce the reported llama-cli command on the NVIDA Grace ARM environment with the Llama3-8B-1.58-100B-tokens TL1 model, varying the listed thread counts. The issue names no source files or tests, so first locate the inference and ARM execution entry points involved in the command. Done means all thread counts complete without malloc or free corruption errors and still produce output.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.