Ecosyste.ms: Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / alpa-projects/tensorflow-alpa issues and pull requests

#141 - [FIX] replace device_count by addressable_device_count

Pull Request - State: closed - Opened by ZYHowell about 2 years ago

#140 - [FIX] fix setting cross mesh communicator

Pull Request - State: closed - Opened by ZYHowell about 2 years ago

#139 - [FIX] fix dtype and shape mismatch in grad_acc_rewrite

Pull Request - State: closed - Opened by ZYHowell about 2 years ago

#138 - Fix auto-sharding follow heuristics for t5x & fix nccl all-reduce-start

Pull Request - State: closed - Opened by merrymercy about 2 years ago

#137 - [FIX] Fix unsupported Gather in auto-sharding and shape mismatch

Pull Request - State: closed - Opened by ZYHowell about 2 years ago

#136 - [FIX] add sharding propagation after auto sharding

Pull Request - State: closed - Opened by ZYHowell over 2 years ago

#135 - [Feature] Support auto sharding for TPU

Pull Request - State: closed - Opened by ZYHowell over 2 years ago

#134 - Fix common ancestor in reduce-scatter & Fix replicated iota

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#133 - Fix iota and rng in auto-sharding

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#132 - Convert pipeline_marker and OptimizationBarrier to bitcast

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#131 - [Feature] Support cross mesh nccl allreduce

Pull Request - State: closed - Opened by ZYHowell over 2 years ago - 1 comment

#130 - [Fix] Fix set input sharding

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#129 - [Fix] Fix various bugs for gradient accumulation and reduce-scatter

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#128 - [Fix] fix an assertion in GenerateReduceScatter

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#127 - Add event level synchronization and refactor nccl communicator api

Pull Request - State: closed - Opened by ZYHowell over 2 years ago - 2 comments

#126 - Add set_hlo_module_input_shardings

Pull Request - State: closed - Opened by comaniac over 2 years ago - 1 comment

#125 - Fix swap thunk impl for SPMD

Pull Request - State: closed - Opened by ZYHowell over 2 years ago

#124 - Control flow support

Pull Request - State: open - Opened by HeydrichBeillschmidt over 2 years ago - 4 comments

#123 - Fix NCCL flaky correctness bug by synchronizing device

Pull Request - State: closed - Opened by TarzanZhao over 2 years ago

#122 - Remove indivisble strategies

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#121 - Bypass cudaStreamLegacy related bug that exists in early version of NCCL

Pull Request - State: closed - Opened by TarzanZhao over 2 years ago - 2 comments

#119 - Fix batch dim analysis for broadcast

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#118 - Fix batch dim analysis for reshape

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#117 - Add status/statusOr for Alpa NCCL apis to catch exception

Pull Request - State: closed - Opened by TarzanZhao over 2 years ago

#115 - expose index select to python api

Pull Request - State: closed - Opened by ZYHowell over 2 years ago

#114 - Update random number generation support for weight initialization

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#113 - Use HloModule instead of XlaComputation for alpa compiler's interface

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#112 - Normalize the dot dimensions

Pull Request - State: closed - Opened by merrymercy over 2 years ago

#110 - Allow only to manually trigger the CI

Pull Request - State: closed - Opened by zhisbug almost 3 years ago

#109 - Fix GetShardingStrategy after supporting variadic reduce

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#108 - Support variadic reduce with tuple output

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#107 - Fix cost calculation for merging nodes

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#106 - Clean up pipeline_marker and identity marker

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#105 - Leave more decisions to ILP solver instead of simplying following

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#104 - Statically link the nccl downloaded by cupy

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#103 - Remove redundant dynamic slice

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#102 - Fix device assignment check

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#101 - Remove use_spmd_partitioning checking in MatchReduceScatter

Pull Request - State: closed - Opened by merrymercy almost 3 years ago - 1 comment

#100 - Support new operators: sort/gather/scatter

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#99 - Fix two out-of-bound errors in auto-sharding

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#97 - Refactor compilation

Pull Request - State: closed - Opened by ZYHowell almost 3 years ago

#96 - Fix integer overflow in nccl thunk

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#95 - Reorganize passes and build options

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#94 - Use DFS to slice pipeline stages

Pull Request - State: closed - Opened by zhuohan123 almost 3 years ago

#93 - Add back the disabled replicated strategy for parameters

Pull Request - State: closed - Opened by merrymercy almost 3 years ago

#92 - Use a new heuristic for choosing instructions to follow

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#91 - Disable force_batch_dim_to_mesh_dim when batch_size == 1

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#90 - rename parax -> alpa

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#88 - Fix swing parameters in slice_auto_sharded_stages.cc

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#87 - Return stage name for pipeline stage slicing pass

Pull Request - State: closed - Opened by zhuohan123 about 3 years ago

#86 - Add TransposeFolding in spmd simplification pipeline & fix typos

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#85 - Enumerate 2d partition for activation & Fix depth analysis for pipeline

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#84 - Fix performance regression after rebase

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#83 - Fix the CanImplement check in all-to-all thunk

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#82 - Fix cost model for integer types

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#81 - Add heuristics for intra-op ablation study

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#80 - Fix indivisble errors in reduce-scatter

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#79 - Do not abort in cost model

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#78 - Enumerate 2d partition for activation & Fix depth analysis for pipeline

Pull Request - State: closed - Opened by ZYHowell about 3 years ago - 1 comment

#77 - Raise catchable python error for invalid arguments

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#76 - Fix out-of-bounds bugs in resharding cost

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#75 - Fix hlo cost model : amortize the cost of gradient sync all-reduce

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#74 - Hlo cost model part 2 : estimate the cost of a HloModule

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#73 - Fix cross-mesh resharding for moe

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#72 - Disable mixed mesh shape for indivisible batch size

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#71 - Fix memory issue of iota in embedding

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#70 - Improve force_batch_dim_to_mesh_dim

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#69 - Fix reshape

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#68 - Fix reshape following for MoE

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#67 - Fix accumulation friendly reduce-scatter for MoE

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#66 - Use a larger value for INFINITY_COST

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#65 - Fix reduce-scatter

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#64 - Expose reset memory stats to python api

Pull Request - State: closed - Opened by ZYHowell about 3 years ago

#63 - Gradient accumulation friendly reduce-scatter

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#62 - Support reduce-scatter in pipeline parallel

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#61 - Fix a heuristic in depth analysis for following

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#60 - Eliminate common subcomputation

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#59 - Track Modifications of Tensorflow

Issue - State: open - Opened by merrymercy about 3 years ago

#58 - Force batch dim split on parameter

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#57 - Change layout from dlpack and add api for gpu memory usage

Pull Request - State: closed - Opened by ZYHowell about 3 years ago

#56 - Fix client memory

Pull Request - State: closed - Opened by ZYHowell about 3 years ago

#55 - Improve memory usage of remat

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#54 - add hook and parse sharding proto bytes

Pull Request - State: closed - Opened by ZYHowell about 3 years ago

#53 - Generate ZeRO stage 3

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#52 - Support mixed logical mesh shape strategies for conv nets

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#51 - fix memory usage api

Pull Request - State: closed - Opened by ZYHowell about 3 years ago

#50 - Generate mixed mesh shape startegies for MoE

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#49 - Support mixed 1d and 2d logical mesh shape

Pull Request - State: closed - Opened by merrymercy about 3 years ago

#48 - Fix depth analysis for pipeline and identity marker

Pull Request - State: closed - Opened by merrymercy over 3 years ago

#47 - expose memzero as customcall

Pull Request - State: closed - Opened by ZYHowell over 3 years ago

#44 - infer spmd sharding from proto

Pull Request - State: closed - Opened by ZYHowell over 3 years ago

#40 - Fix conflicts of env var XLA_SKIP_NCCL_COLLECTIVE_IDS

Pull Request - State: closed - Opened by ZYHowell over 3 years ago

#20 - swap insertion compatible with auto-sharding

Pull Request - State: closed - Opened by ZYHowell over 3 years ago - 1 comment

#8 - Rebase to tensorflow mainline

Issue - State: open - Opened by merrymercy over 3 years ago - 4 comments