GitLab

Put the arguments in a JSON file:
```
{
    "dtype": "half",
    "trust_remote_code": true
}
```

Pass it like this:
```
dynamo-run out=sglang ~/llm_models/Llama-3.2-3B-Instruct --extra-engine-args sglang_extra.json
```

Requested here https://github.com/ai-dynamo/dynamo/issues/290 (`dtype`) and here https://github.com/ai-dynamo/dynamo/issues/360 (`trust_remote_code`).

This lets us do:
```
dynamo-run out=llamacpp <gguf_file>
```

Previously a `--model-config <hf-repo>` was also required, to configure our tokenizer.

That ensures it gets removed when the process stops.

Co-authored-by: Olga Andreeva <oandreeva@oandreeva-mlt.client.nvidia.com>

Co-authored-by: Meenakshi Sharma <163925564+nvda-mesharma@users.noreply.github.com>

It hardly slows the build down, and it makes things run much faster. That allows us to switch to the debug (default) profile for development, and keep the release profile for, well, releasing.

Motivated by changes in https://github.com/ai-dynamo/dynamo/pull/279

Add `AsMut`, `DerefMut` and `IntoIterator` trait impl for the `Tokens` structure.
Signed-off-by: nora-coder-dot <nora6677@gmail.com>
Co-authored-by: nora-coder-dot <nora6677@gmail.com>

Signed-off-by: zibai <zibai.gj@alibaba-inc.com>

Co-authored-by: Ryan Olson <ryanolson@users.noreply.github.com>

Co-authored-by: Dmitry Tokarev <dtokarev@nvidia.com>

This makes the Rust parts all use ring / rustls library instead of local install of openssl. It's a step on the journey to being statically linked.

Pieces:
- `tokenizers` and `mistralrs` now support rustls (mistralrs by default, tokenizers with feature flag).
- Move shared dependencies up into workspace
- New `rand` crate has some renames for future rust
- Ensure the dependency doesn't creep back in by enforcing it with cargo deny.

#### Overview:

This PR enables more aggressive compiler optimizations for the project which should lead to better performance and smaller binary sizes.

In this PR, I decided to use Fat LTO instead of ThinLTO since it provides higher optimization level.

I have made quick tests (AMD Ryzen 5900x, Fedora 41, Rust 1.85.1, the latest version of the project at the moment, `cargo build --release` command) - here are the results about the binary size improvements.

| Binary\Build mode | dynamo-run | libdynamo_llm_capi.so | http | llmctl | metrics | mock_worker |
| --- | --- | --- | --- | --- | --- | --- |
| Release | 55 Mib | 14 Mib | 19 Mib | 14 Mib | 21 Mib | 14 Mib |
| Release + `codegen-units = 1` + ThinLTO | 43 Mib | 11 Mib | 15 Mib | 11 Mib | 17 Mib | 11 Mib |
| Release + `codegen-units = 1` + FatLTO | 38 Mib | 9.2 Mib | 13 Mib | 9.6 Mib | 15 Mib | 9.6 Mib |

#### Details:

Enable `codegen-units = 1` and Fat LTO for better optimizations.

#### Where should the reviewer start?

Just check the `Cargo.toml` file ;)

#### Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

- closes GitHub issue: #278 

Under load it sometimes drops a request. The request gets added to the batch (sequence) and immediately gets a FinishReason Stop. Not sure why. It doesn't happen with the default scheduler (non-paged attention), so switch to that for now.

Co-authored-by: mabdulwahhab <mabdulwahhab@nvidia.com>

Co-authored-by: Anant Sharma <anants@nvidia.com>

Co-authored-by: Dmitry Tokarev <dtokarev@nvidia.com>
Co-authored-by: Meenakshi Sharma <163925564+nvda-mesharma@users.noreply.github.com>

Co-authored-by: Dmitry Tokarev <dtokarev@nvidia.com>

Co-authored-by: mabdulwahhab <mabdulwahhab@nvidia.com>
Co-authored-by: Meenakshi Sharma <163925564+nvda-mesharma@users.noreply.github.com>

Co-authored-by: Meenakshi Sharma <163925564+nvda-mesharma@users.noreply.github.com>

Co-authored-by: Anant Sharma <anants@nvidia.com>