How to use multi GPUs - Githubissues

I try to setting lazy_unpickle=True, out_shard_size=parse_kmb("1B"). This does not solve my question. I got error as:

Traceback (most recent call last):
  File "/mnt/data/Projects/LLM-Merging/llm_merging/main.py", line 49, in <module>
    merge_method.merge()
  File "/mnt/data/Projects/LLM-Merging/llm_merging/merging/DellaLlama31.py", line 61, in merge
    run_merge(
  File "/mnt/data/Projects/LLM-Merging/mergekit/merge.py", line 96, in run_merge
    for _task, value in exec.run(quiet=options.quiet):
  File "/mnt/data/Projects/LLM-Merging/mergekit/graph.py", line 197, in run
    res = task.execute(**arguments)
  File "/mnt/data/Projects/LLM-Merging/mergekit/merge_methods/generalized_task_arithmetic.py", line 160, in execute
    deltas = torch.stack([tv["delta"] for tv in tvs], dim=0)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 5.87 GiB. GPU 0 has a total capacity of 23.65 GiB of which 4.61 GiB is free. Including non-PyTorch memory, this process has 19.03 GiB memory in use. Of the allocated memory 6.85 GiB is allocated by PyTorch, and 11.74 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

My code is:

COPY_TOKENIZER = True  # you want a tokenizer? yeah, that's what i thought
LAZY_UNPICKLE = True  # experimental low-memory model loader
LOW_CPU_MEMORY = False  # enable if you somehow have more VRAM than RAM+swap
run_merge(
  merge_config,
  out_path=OUTPUT_PATH,
  options=MergeOptions(
    lora_merge_cache=LORA_MERGE_CACHE,
    cuda=torch.cuda.is_available(),
    copy_tokenizer=COPY_TOKENIZER,
    lazy_unpickle=LAZY_UNPICKLE,
    low_cpu_memory=LOW_CPU_MEMORY,
    out_shard_size=parse_kmb("1B"),
  ),
)

arcee-ai / mergekit

How to use multi GPUs #420