pytorch/torchtitan

add-moe-model

Add a new HF MoE model to the transformers_modeling_backend experiment. Use when the user provides a HuggingFace model ID and wants to integrate it with Titan's MoE, or invokes /add_moe_model.

Idioma do texto original: inglês

numerics-debugging

Capture and compare per-op activations between two TorchTitan runs to spot numerics divergence (eager vs aot_fx_trace, FSDP vs no-FSDP, before vs after a refactor). Use when the user wants to debug bitwise / numeric drift in training, or invokes…

Idioma do texto original: inglês

cuda-graph-trace-compaction

Make a CUDA-graph PyTorch profiler trace readable in Perfetto by packing replay streams into semantic lanes, and optionally merge per-rank pipeline-parallel traces into one aligned timeline with send/recv flow arrows. Use when a trace captured with CUDA…

Idioma do texto original: inglês

pr-review

Review torchtitan pull requests the way the maintainers do -- question the problem and design first, push for minimal code, explain the change with a concrete example, then check correctness. Use when reviewing a PR or branch, when asked to "review", "review…

Idioma do texto original: inglês

pr-description

Write a torchtitan PR description that convinces a human reviewer -- concrete problem with a simple example, what changed, user-facing before/after, numerics, testing. Use when opening or updating a PR, or when asked to write or improve a PR summary.

Idioma do texto original: inglês

local-compile-perf

Make a torchtitan model train faster with `local_compile` regions and compile-friendly rewrites: measure, find the slow glue, rewrite it, verify. Use when asked to speed up a model, layer or op with torch.compile, or to review such a change.

Idioma do texto original: inglês

fake-distributed-backend

Run TorchTitan with fully fake communication or real pipeline communication plus fake SPMD axes, especially for logical-rank memory debugging. Use for fake process groups, real-PP/fake-SPMD runs, large-model dry runs, or comparing rank-local CUDA memory…

Idioma do texto original: inglês

inference-perf-hillclimb

Profiler-driven hill-climbing to close the inference throughput gap between TorchTitan's unified model (running inside vLLM) and vLLM's native model. Benchmark with generate.py --benchmark, climb optimization rungs (compile / cudagraph / fused kernels),…

Idioma do texto original: inglês

kernel-numerics-verifier

Verify forward and backward numerical correctness of a handwritten low-precision GPU kernel against a high-precision oracle and an eager same-dtype baseline. Use for Triton or CUDA kernel correctness; use numerics_debugging to locate an unknown whole-model…

Idioma do texto original: inglês

torch-bisect

Bisect PyTorch commits to find the regression that breaks TorchTitan. Use when the user wants to bisect PyTorch or invokes /torch_bisect.

Idioma do texto original: inglês