Ring distributed backend for Metal

[EricLBuehler]

At its core, a ring-based All Reduce algorithm backend.

This enables tensor parallelism for Metal users!

[github-actions[bot]]

Code Metrics Report

===============================================================================
 Language            Files        Lines         Code     Comments       Blanks
===============================================================================
 C Header                3           62           53            0            9
 CSS                     1          428          366           12           50
 Dockerfile              1           39           22            9            8
 HTML                    1           58           46            4            8
 JavaScript              7         1221          915          169          137
 JSON                   12          107          106            0            1
 Makefile                1            6            5            0            1
 Python                 86         4045         3410          161          474
 Shell                   1           63           26           18           19
 Plain Text              3         3723            0         2413         1310
 TOML                   20          616          558           10           48
 YAML                    2           21           19            2            0
-------------------------------------------------------------------------------
 Jupyter Notebooks       3            0            0            0            0
 |- Markdown             2           77           32           31           14
 |- Python               2          205          178            1           26
 (Total)                            282          210           32           40
-------------------------------------------------------------------------------
 Markdown               56         4936            0         3764         1172
 |- BASH                 9           99           96            0            3
 |- JSON                 1           12           12            0            0
 |- Python               7          121          109            0           12
 |- Rust                22          757          634            1          122
 |- TOML                 2           75           63            0           12
 (Total)                           6000          914         3765         1321
-------------------------------------------------------------------------------
 Rust                  367       129720       115534         2804        11382
 |- Markdown           162         2232           29         1994          209
 (Total)                         131952       115563         4798        11591
===============================================================================
 Total                 564       145045       121060         9366        14619
===============================================================================

[EricLBuehler]

RING_PORT=1500 RING_RIGHT=1501 RING_RANK=0 RING_WORLD_SIZE=2 cargo run --features metal,ring --release '--' -i --throughput plain -m ../hf_models/llama3.2_3b

RING_PORT=1501 RING_RIGHT=1500 RING_RANK=1 RING_WORLD_SIZE=2 cargo run --features metal,ring --release '--' -i --throughput plain -m ../hf_models/llama3.2_3b

[coderabbitai[bot]]

"""

Walkthrough

A new "ring" feature flag is introduced across multiple crates, enabling a TCP socket-based ring all-reduce backend for distributed operations. Conditional logic and compilation paths are updated to support this backend as an alternative to NCCL, with changes in configuration files and core distributed operation modules to recognize and utilize the "ring" feature.

Changes

File(s)	Change Summary
mistralrs-core/Cargo.toml mistralrs-bench/Cargo.toml mistralrs-pyo3/Cargo.toml mistralrs-server/Cargo.toml mistralrs/Cargo.toml	Add "ring" feature flag, linking to downstream crates' "ring" features where needed.
mistralrs-quant/Cargo.toml	Add empty "ring" feature and tokio as a dependency.
mistralrs-core/src/distributed.rs	Add conditional branch in prepare_distributed_mapper for "ring" feature, initializing distributed context; add ring_daemon_replicator and update is_daemon.
mistralrs-core/src/pipeline/normal.rs	Expand distributed device/model loading logic to include "ring" feature in relevant conditionals.
mistralrs-core/src/utils/varbuilder_utils.rs	Broaden mmap loading condition to include "ring" feature.
mistralrs-quant/src/distributed/mod.rs	Add new ops module under "ring" feature implementing TCP ring all-reduce backend; add RingConfig struct and related methods; extend public API to expose RingConfig.
mistralrs-core/src/engine/mod.rs	Add TCP socket-based request replication for "ring" feature in Engine::replicate_request_to_daemons.
mistralrs-core/src/lib.rs	Refactor daemon mode to use ring_daemon_replicator or nccl_daemon_replicator functions instead of inline socket handling.
mistralrs-quant/src/lib.rs	Publicly re-export RingConfig from distributed socket module.
mistralrs-core/src/models/qwen3_moe.rs	Change argument to gather call to use contiguous indices tensor.
mistralrs-core/src/pipeline/sampling.rs	Add new stop reason ToolCalls when tool calls are detected during token streaming or completion.
mistralrs-core/src/sequence.rs	Add ToolCalls variant to StopReason enum and update its Display implementation; change string format for some variants.

Sequence Diagram(s)

sequenceDiagram
    participant User
    participant App
    participant DistributedMapper
    participant RingComm
    participant TCPNetwork

    User->>App: Start distributed model loading (with "ring" feature)
    App->>DistributedMapper: prepare_distributed_mapper()
    DistributedMapper->>RingComm: Initialize ring context (Id, rank, world size)
    RingComm->>TCPNetwork: Establish connections to left/right neighbors
    App->>RingComm: Perform sum_all_reduce operation
    RingComm->>TCPNetwork: Send/receive tensor data in ring
    RingComm-->>App: Return reduced tensor

Possibly related PRs

• EricLBuehler/mistral.rs#1391: Modifies FastMoeMlp::forward method gather calls by removing .contiguous()?, directly related but with opposite change to the same code.

Poem

In circles we gather, a ring now in code,
TCP whispers secrets down distributed road.
Features unfurl in a config parade,
New pathways for tensors, connections well-laid.
With sockets and bytes, our sum travels far—
Hooray for the ring, from your code rabbit star!
🐇✨ """

✨ Finishing Touches

• [ ] 📝 Generate Docstrings

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

• Review comments: Directly reply to a review comment made by CodeRabbit. Example:
  • I pushed a fix in commit <commit_id>, please review it.
  • Explain this complex logic.
  • Open a follow-up GitHub issue for this discussion.
• Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
  • @coderabbitai explain this code block.
  • @coderabbitai modularize this function.
• PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
  • @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
  • @coderabbitai read src/utils.ts and explain its main purpose.
  • @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
  • @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

• @coderabbitai pause to pause the reviews on a PR.
• @coderabbitai resume to resume the paused reviews.
• @coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
• @coderabbitai full review to do a full review from scratch and review all the files again.
• @coderabbitai summary to regenerate the summary of the PR.
• @coderabbitai generate docstrings to generate docstrings for this PR.
• @coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
• @coderabbitai resolve resolve all the CodeRabbit review comments.
• @coderabbitai configuration to show the current CodeRabbit configuration for the repository.
• @coderabbitai help to get help.

Other keywords and placeholders

• Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
• Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
• Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (.coderabbit.yaml)

• You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
• Please see the configuration documentation for more information.
• If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

• Visit our Documentation for detailed information on how to use CodeRabbit.
• Join our Discord Community to get help, request features, and share feedback.
• Follow us on X/Twitter for updates and announcements.