app : add the llama download subcommand

[angt]

Overview

Add the tool llama-download and the download subcommand of llama.

Additional information

I try to keep the current usage and avoid code duplication, but maybe it's also the opportunity to rework this part ?

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES

[ngxson]

IMO we don't really need a dedicated llama-download binary for it. The goal of llama-* binaries is mostly for compatibility (i.e. users who already use llama-* continue to use it), but this subcommand is new so it's quite redundant to be both a subcommand and a dedicated binary

[CISC]

IMO we don't really need a dedicated llama-download binary for it. The goal of llama-* binaries is mostly for compatibility (i.e. users who already use llama-* continue to use it), but this subcommand is new so it's quite redundant to be both a subcommand and a dedicated binary

Also would not a more appropriate name be llama cache as download suggests to me the files would be placed in a folder of my choosing for more permanent storing.

[angt]

IMO we don't really need a dedicated llama-download binary for it. The goal of llama-* binaries is mostly for compatibility (i.e. users who already use llama-* continue to use it), but this subcommand is new so it's quite redundant to be both a subcommand and a dedicated binary

I agree with that too, @ggerganov ?

[angt]

Also would not a more appropriate name be llama cache as download suggests to me the files would be placed in a folder of my choosing for more permanent storing.

llama cache will be also provided to handle the cache, here we only focus on downloading models, selecting the best tag, without launching the server.

we are approximately reproducing the UX of hf:

Main commands:
  ...
  cache                Manage local cache directory.
  download             Download files from the Hub.
  ...

[ggerganov]

IMO we don't really need a dedicated llama-download binary for it. The goal of llama-* binaries is mostly for compatibility (i.e. users who already use llama-* continue to use it), but this subcommand is new so it's quite redundant to be both a subcommand and a dedicated binary

I agree with that too, @ggerganov ?

Either way is fine. Let's remove the llama-download binary build for now.

[angt]

Done

$ ./build/bin/llama download --help
----- example-specific params -----

-h,    --help, --usage                  print usage and exit
--mmproj-auto, --no-mmproj, --no-mmproj-auto
                                        whether to use multimodal projector file (if available), useful when
                                        using -hf (default: enabled)
                                        (env: LLAMA_ARG_MMPROJ_AUTO)
-m,    --model FNAME                    model path to load
                                        (env: LLAMA_ARG_MODEL)
-mu,   --model-url MODEL_URL            model download url (default: unused)
                                        (env: LLAMA_ARG_MODEL_URL)
-dr,   --docker-repo [<repo>/]<model>[:quant]
                                        Docker Hub model repository. repo is optional, default to ai/. quant
                                        is optional, default to :latest.
                                        example: gemma3
                                        (default: unused)
                                        (env: LLAMA_ARG_DOCKER_REPO)
-hf,   -hfr, --hf-repo <user>/<model>[:quant]
                                        Hugging Face model repository; quant is optional, case-insensitive,
                                        default to Q4_K_M, or falls back to the first file in the repo if
                                        Q4_K_M doesn't exist.
                                        mmproj is also downloaded automatically if available. to disable, add
                                        --no-mmproj
                                        example: ggml-org/GLM-4.7-Flash-GGUF:Q4_K_M
                                        (default: unused)
                                        (env: LLAMA_ARG_HF_REPO)
-hff,  --hf-file FILE                   Hugging Face model file. If specified, it will override the quant in
                                        --hf-repo (default: unused)
                                        (env: LLAMA_ARG_HF_FILE)
-hft,  --hf-token TOKEN                 Hugging Face access token (default: value from HF_TOKEN environment
                                        variable)
                                        (env: HF_TOKEN)
--mtp                                   also download the multi-token prediction (MTP) head, if available
                                        (default: unused)

examples:
  download -hf ggml-org/gemma-3-4b-it-qat-GGUF
  download -hf ggml-org/gemma-3-4b-it-qat-GGUF:Q4_K_M
  download -hf ggml-org/models -hff model.gguf
  download -mu https://example.com/model.gguf -m model.gguf

[julien-c]

do we need the --docker-repo etc flags? not sure if they're actually used widely in practice:)