Ecosyste.ms: Issues

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

GitHub / apple/axlearn issues and pull requests

#989 - Add Goodput documentation

Pull Request - State: open - Opened by jiya-zhang 1 day ago

#988 - Fix Parameter Redefinition in learner_test.py

Pull Request - State: open - Opened by apivovarov 1 day ago

#987 - Fix Missing return statement in various classes

Pull Request - State: open - Opened by apivovarov 1 day ago - 2 comments

#986 - Migrate from Legacy JAX APIs jax.tree_util to jax.tree

Pull Request - State: open - Opened by apivovarov 2 days ago

#985 - pip install command in docuemntation doesn't work

Issue - State: closed - Opened by wangkuiyi 2 days ago - 4 comments

#984 - GoodPut minor fix: only process 0 should start goodput uploader

Pull Request - State: open - Opened by jiya-zhang 3 days ago - 1 comment

#982 - Add support to slice dataset based on proportions.

Pull Request - State: closed - Opened by RsEnts 3 days ago

#981 - Context parallelism for TPU

Pull Request - State: open - Opened by hanzhi713 3 days ago

#980 - Add support for grain.IterDataset in sampling

Pull Request - State: closed - Opened by RsEnts 3 days ago

#979 - Transpose kv cache for better decode performance

Pull Request - State: closed - Opened by changlan 4 days ago - 1 comment

#978 - Allow metrics layers to have state.

Pull Request - State: closed - Opened by markblee 4 days ago

#977 - Ensures that cache_dtype is respected.

Pull Request - State: closed - Opened by markblee 4 days ago

#976 - Add segment_ids option in DiTAttentionLayer

Pull Request - State: closed - Opened by weiliu89 5 days ago

#975 - Remove redundant import logging

Pull Request - State: closed - Opened by apivovarov 5 days ago

#974 - Fix membership checks in tool_use_execution.py

Pull Request - State: closed - Opened by apivovarov 5 days ago

#973 - Replace jnp.ndarray with Tensor from axlearn.common.utils

Pull Request - State: closed - Opened by apivovarov 5 days ago

#972 - Use broadcasting trick for KV update

Pull Request - State: closed - Opened by changlan 7 days ago

#971 - Support system role when calling the Gemini API.

Pull Request - State: closed - Opened by zetaqubit 7 days ago

#970 - Licensing clashes

Issue - State: open - Opened by hyandell 8 days ago

#969 - Making shared_memory configurable

Pull Request - State: closed - Opened by RsEnts 8 days ago

#968 - Don't keep initial key/value inputs in the KV cache.

Pull Request - State: closed - Opened by ds-hwang 8 days ago - 2 comments

#967 - :sparkles: Add cache for CloudBuild API location queries

Pull Request - State: closed - Opened by dswann5 8 days ago

#966 - Fix incorrect number of formatting arguments

Pull Request - State: closed - Opened by changlan 9 days ago

#965 - Reduce the verbosity of variable norm summaries

Pull Request - State: closed - Opened by dunan 11 days ago

#964 - [Enhancement] Add PPO/GRPO

Issue - State: open - Opened by sbhavani 11 days ago - 1 comment

#962 - Sliding window support for GPU flash attention

Pull Request - State: closed - Opened by kelvin-zou 14 days ago

#961 - Skipping empty grain batches during unbatch.

Pull Request - State: closed - Opened by markblee 15 days ago

#960 - Supports loss_weights and live_targets in metrics.

Pull Request - State: closed - Opened by markblee 15 days ago

#959 - Improve gcsfuse io through setting a couple of flags

Pull Request - State: closed - Opened by RsEnts 16 days ago

#958 - SplashAttention performance tuning for v6e

Pull Request - State: closed - Opened by hanzhi713 16 days ago - 1 comment

#957 - Use env id for gcp settings

Pull Request - State: closed - Opened by Ethanlm 16 days ago

#956 - Use InputDispatcher for fuji models

Pull Request - State: closed - Opened by hanzhi713 17 days ago

#955 - Add v6e PCIe overload workaround flag

Pull Request - State: closed - Opened by hanzhi713 17 days ago

#954 - Fix GCSFUSE flags by setting resource limit.

Pull Request - State: closed - Opened by RsEnts 18 days ago

#953 - Explicitly pass module outputs to metrics.

Pull Request - State: closed - Opened by markblee 18 days ago

#952 - Add v6e special meshes

Pull Request - State: closed - Opened by hanzhi713 19 days ago - 1 comment

#951 - Workaround module outputs being dropped.

Pull Request - State: closed - Opened by markblee 19 days ago

#950 - Adds logging to aux loss collection.

Pull Request - State: open - Opened by ruomingp 20 days ago

#949 - Update LoraFusedQKVLinear

Pull Request - State: closed - Opened by qdavid1 20 days ago

#948 - update jax to 0.4.37

Pull Request - State: closed - Opened by matthew-e-hopkins 21 days ago - 2 comments

#947 - Add link to github issue regarding kubernetes-32.0.0

Pull Request - State: closed - Opened by Ethanlm 21 days ago

#945 - J

Issue - State: closed - Opened by julianandresr199 22 days ago

#945 - J

Issue - State: closed - Opened by julianandresr199 22 days ago

#944 - Forward input keys to decoder.

Pull Request - State: closed - Opened by markblee 22 days ago

#944 - Forward input keys to decoder.

Pull Request - State: closed - Opened by markblee 22 days ago

#943 - Legacy flash remat fix

Pull Request - State: closed - Opened by hanzhi713 23 days ago

#942 - Some fixes for flash remat

Pull Request - State: closed - Opened by hanzhi713 24 days ago

#941 - Uses pytest markers instead of module skip.

Pull Request - State: open - Opened by markblee 24 days ago

#940 - Add GKE A3 Ultra support

Pull Request - State: open - Opened by samos123 24 days ago - 1 comment

#939 - Flash Attention for Neuron

Pull Request - State: open - Opened by apoorvtintin 24 days ago - 10 comments

#938 - Repeat KV heads in Flash Attention

Pull Request - State: closed - Opened by changlan 24 days ago

#937 - AOT compilation for v6e

Pull Request - State: closed - Opened by changlan 25 days ago

#936 - Adds mesh rule for a3-megagpu-8g.

Pull Request - State: closed - Opened by markblee 25 days ago

#935 - Avoid a top-level import of tokenizers.

Pull Request - State: closed - Opened by markblee 27 days ago

#934 - Makes causal lm metrics configurable.

Pull Request - State: closed - Opened by markblee 27 days ago - 1 comment

#933 - Supports flexible input partition specs.

Pull Request - State: closed - Opened by markblee 27 days ago - 1 comment

#932 - Enable GCP Workload Monitoring

Pull Request - State: open - Opened by Perseus14 28 days ago

#931 - Add ReadOptions args to _make_autoregressive_inputs

Pull Request - State: closed - Opened by RsEnts 28 days ago

#930 - AdaptiveLayerNormModulation raises ValueError, instead of assert.

Pull Request - State: closed - Opened by ds-hwang 29 days ago - 1 comment

#929 - Fix aot compilation with grain inputs.

Pull Request - State: closed - Opened by markblee 30 days ago

#928 - Add prefill hidden states as module outputs.

Pull Request - State: closed - Opened by markblee 30 days ago - 1 comment

#927 - Cache AoT compilation result

Pull Request - State: closed - Opened by hanzhi713 30 days ago

#926 - Allow external positions to be inputed in RoPE embedding layer

Pull Request - State: closed - Opened by Firenze11 about 1 month ago

#925 - Support remat for FlashAttention

Pull Request - State: closed - Opened by hanzhi713 about 1 month ago

#924 - Fix tf iter unit test

Pull Request - State: closed - Opened by hanzhi713 about 1 month ago

#923 - DiT now supports sequence conditions.

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 1 comment

#922 - Enabled running Pallas Flash Attention on CPU.

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 3 comments

#921 - [Grain] Minor Fix for Version Update

Pull Request - State: closed - Opened by zxybazh about 1 month ago

#920 - Introduce `BaseAttentionBias.has_value()`.

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 1 comment

#918 - Draft: Refactor JobSet for Pathways

Pull Request - State: open - Opened by jiya-zhang about 1 month ago - 1 comment

#917 - fix broken apt install google-perftools

Pull Request - State: closed - Opened by samos123 about 1 month ago - 1 comment

#916 - TRN2 Meshes and Configurations

Pull Request - State: open - Opened by apoorvtintin about 1 month ago - 16 comments

#914 - Optional `positions` support in decoder and attention layers

Pull Request - State: closed - Opened by changlan about 1 month ago

#913 - Enable cudnn attention dropout

Pull Request - State: closed - Opened by hanzhi713 about 1 month ago - 5 comments

#912 - V6e support

Pull Request - State: closed - Opened by kelvin-zou about 1 month ago - 3 comments

#911 - Update lora input linear adapter output dim.

Pull Request - State: closed - Opened by JianyuWangV about 1 month ago - 1 comment

#910 - Allow parallel gpu tests

Pull Request - State: closed - Opened by hanzhi713 about 1 month ago

#909 - Introduce the scale enum flag in Embedding layer for LLM embedding.

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 1 comment

#908 - Optimize TPU Flash Attention (20x XLA compilation speed-up on 32k long context)

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 7 comments

#907 - The codebook of the KmeansVectorQuantizer should be initialized with scale=1/sqrt(dim).

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 1 comment

#906 - `AdaptiveLayerNormModulation` now supports sequence conditions.

Pull Request - State: closed - Opened by ds-hwang about 1 month ago - 1 comment

#905 - Flash2 and supports cross attention and dropout

Pull Request - State: closed - Opened by hanzhi713 about 1 month ago - 1 comment

#904 - Remove version pin of typing-extensions

Pull Request - State: closed - Opened by wangkuiyi about 2 months ago - 2 comments

#903 - Fix softmax scale arg passing

Pull Request - State: closed - Opened by hanzhi713 about 2 months ago

#901 - Little clean-up in frontend.

Pull Request - State: closed - Opened by ds-hwang about 2 months ago - 1 comment

#899 - Implements FlashDecoding with Sparsity Support

Pull Request - State: closed - Opened by hanzhi713 about 2 months ago

#899 - Implements FlashDecoding with Sparsity Support

Pull Request - State: closed - Opened by hanzhi713 about 2 months ago

#898 - Special remat for Neuron

Pull Request - State: closed - Opened by apoorvtintin about 2 months ago - 17 comments

#891 - N

Issue - State: closed - Opened by julianandresr199 2 months ago

#890 - use "true" and "false" instead of 0 and 1

Pull Request - State: open - Opened by samos123 2 months ago - 1 comment

#887 - Add default compiler options for v6e

Pull Request - State: closed - Opened by samos123 2 months ago

#886 - [DO-NOT-MERGE] PR encompassing all changes needed to support neuron on Axlearn

Pull Request - State: closed - Opened by apoorvtintin 2 months ago - 1 comment

#885 - Add meshes and config for TRN2/1 for Fuji models

Pull Request - State: closed - Opened by apoorvtintin 2 months ago - 2 comments