apple/axlearn issues and pull requests

#634 - Add transformer block output to module output

Pull Request - State: closed - Opened by alex8937 6 months ago

#633 - Adds honeycrisp trainer configs with deterministic inputs

Pull Request - State: closed - Opened by ruomingp 6 months ago

#632 - Makes the TPU log command more prominent.

Pull Request - State: closed - Opened by ruomingp 6 months ago

#631 - Fix flaky test.

Pull Request - State: closed - Opened by apghml 6 months ago

#630 - Add attention_logit_biases in DiTAttentionLayer

Pull Request - State: closed - Opened by weiliu89 6 months ago

#629 - Add core deps to different build targets.

Pull Request - State: closed - Opened by markblee 6 months ago - 1 comment

#628 - Making core optional causing no module names absl error

Issue - State: closed - Opened by samos123 6 months ago - 1 comment

#627 - Add Orbax Checkpointing

Pull Request - State: closed - Opened by jiya-zhang 6 months ago - 1 comment

#626 - Add structure option in DiT attn and feedforward layers

Pull Request - State: closed - Opened by weiliu89 6 months ago

#625 - Adds honeycrisp models

Pull Request - State: closed - Opened by ruomingp 6 months ago

#624 - Make `tree_paths()` consistent with jax.

Pull Request - State: closed - Opened by apghml 6 months ago - 1 comment

#623 - Enable support for Kueue for GKETPUJob

Pull Request - State: closed - Opened by samos123 6 months ago - 1 comment

#622 - allow using on-demand instead of spot only

Pull Request - State: open - Opened by samos123 6 months ago - 2 comments

#621 - Unable to use TPU on GKE using on-demand quota

Issue - State: open - Opened by samos123 6 months ago

#620 - remove nodeSelector provisioner-nodepool

Pull Request - State: open - Opened by samos123 6 months ago

#619 - nodeSelector set by default requires tpu provisioner

Issue - State: open - Opened by samos123 6 months ago

#618 - Add `InvocationContext.functional()`.

Pull Request - State: closed - Opened by apghml 6 months ago

#618 - Add `InvocationContext.functional()`.

Pull Request - State: closed - Opened by apghml 6 months ago

#617 - Fix frame size for pre-emphasis.

Pull Request - State: closed - Opened by markblee 6 months ago

#616 - Change the CloudBuild to Sync by Default

Pull Request - State: closed - Opened by amcw7777 7 months ago

#615 - Supports custom config field validators.

Pull Request - State: closed - Opened by markblee 7 months ago

#615 - Supports custom config field validators.

Pull Request - State: closed - Opened by markblee 7 months ago

#612 - Add option to return module outputs from inference runner

Pull Request - State: closed - Opened by a-metz 7 months ago

#607 - Supports configuring spectrogram implementation in mel frontend.

Pull Request - State: closed - Opened by markblee 7 months ago

#567 - New DataPartitionType DATA

Pull Request - State: closed - Opened by apoorvtintin 8 months ago - 3 comments

#566 - Neuron support in Axlearn

Pull Request - State: closed - Opened by apoorvtintin 8 months ago - 5 comments

#556 - add kv_state arg to forward function in RoFormerQKVLinear

Pull Request - State: closed - Opened by chongw 8 months ago

#542 - Add github action support

Pull Request - State: closed - Opened by samos123 8 months ago - 1 comment

#534 - Decouples from diffusers.

Pull Request - State: closed - Opened by markblee 8 months ago

#464 - Te flash attention

Pull Request - State: closed - Opened by kelvin-zou 9 months ago

#434 - Upgrade jax to 0.4.25

Pull Request - State: closed - Opened by kelvin-zou 10 months ago

#433 - Adjust fuji single host configs.

Pull Request - State: open - Opened by ruomingp 10 months ago

#432 - Addresses Kelvin's comments

Pull Request - State: closed - Opened by ruomingp 10 months ago - 1 comment

#431 - Adds bpe_128k.json and train_spm.py.

Pull Request - State: open - Opened by ruomingp 10 months ago

#430 - upgrade to hf 4.40.1

Pull Request - State: closed - Opened by gyin94 10 months ago - 1 comment

#429 - update launch print log

Pull Request - State: closed - Opened by zhiyun 10 months ago

#428 - Adds versions V1,V2,V3 for the Fuji models.

Pull Request - State: closed - Opened by ruomingp 10 months ago - 1 comment

#427 - Fix dataflow deps.

Pull Request - State: closed - Opened by markblee 10 months ago

#426 - Reverts back to jax 0.4.21 temporarily.

Pull Request - State: closed - Opened by ruomingp 10 months ago

#425 - refactor rnn api

Pull Request - State: open - Opened by zhiyun 10 months ago

#424 - Supports tier scheduling in {tpu,gke}_runner.

Pull Request - State: closed - Opened by markblee 10 months ago

#423 - allow _map_label_sequences to set `remove_repeats=False`

Pull Request - State: closed - Opened by zhiyun 10 months ago

#421 - [rnnt] add RNN prediction network

Pull Request - State: closed - Opened by zhiyun 10 months ago

#420 - chore(axlearn): Fix typos

Pull Request - State: open - Opened by tony 10 months ago - 2 comments

#419 - [rnnt] make blank_id field required in transducer

Pull Request - State: closed - Opened by zhiyun 10 months ago

#418 - update absl and fix tests

Pull Request - State: closed - Opened by zhzhyi 10 months ago - 1 comment

#417 - fix absl-py version

Pull Request - State: closed - Opened by zhiyun 10 months ago

#416 - Initial GKE runner and launcher.

Pull Request - State: closed - Opened by markblee 10 months ago

#415 - Re-schedule bastion jobs that change tiers.

Pull Request - State: closed - Opened by markblee 10 months ago

#414 - change blank_token_id -> blank_id in CTCDecoderModel

Pull Request - State: closed - Opened by zhiyun 10 months ago - 1 comment

#413 - Upgrade Jax to 0.4.25

Pull Request - State: closed - Opened by kelvin-zou 10 months ago

#412 - Pass Along soft_label in EncoderModel if Available.

Pull Request - State: closed - Opened by tuzhucheng 10 months ago

#411 - Replace Matmul in Linear Layer with Variable Shape Einsum

Pull Request - State: closed - Opened by jiarui-lu2 10 months ago

#410 - An initial implementation of scheduling tiers.

Pull Request - State: closed - Opened by markblee 10 months ago

#409 - Allow asr input to truncate speech

Pull Request - State: closed - Opened by zhiyun 10 months ago

#408 - make mel energy floor value required

Pull Request - State: closed - Opened by yqwangustc 10 months ago - 1 comment

#407 - Support adaptive drop norm

Pull Request - State: closed - Opened by taolei87 10 months ago

#405 - Check ema before assign

Pull Request - State: closed - Opened by weiliu89 10 months ago

#404 - Add skip_bundle to reuse images.

Pull Request - State: closed - Opened by markblee 10 months ago

#403 - Finish changes to `bind_layer`.

Pull Request - State: closed - Opened by apghml 10 months ago - 1 comment

#402 - Fix error `bad-plugin-value` for `pylint.extensions.docparams`.

Pull Request - State: closed - Opened by apghml 10 months ago

#401 - Add `bind_layer()`.

Pull Request - State: closed - Opened by apghml 10 months ago

#400 - Replace `chex.dataclass` with `struct.PyTreeNode`.

Pull Request - State: closed - Opened by apghml 10 months ago

#399 - Enable using ImageSummary in SpmdEvaler.

Pull Request - State: closed - Opened by apghml 10 months ago - 1 comment

#398 - Adds `Learner.forward_and_backward`.

Pull Request - State: closed - Opened by apghml 10 months ago

#397 - BERT layer_norm_epsilon should be used if passed in.

Pull Request - State: closed - Opened by tuzhucheng 10 months ago

#396 - Pin scipy.

Pull Request - State: closed - Opened by markblee 10 months ago

#395 - Relax is_valid_job_name check.

Pull Request - State: closed - Opened by markblee 10 months ago

#394 - Fix `MetricAccumulator` bug.

Pull Request - State: closed - Opened by apghml 10 months ago

#393 - Flush vertexai tensorboard more frequently to workaround OutOfSpace bug.

Pull Request - State: open - Opened by markblee 10 months ago

#392 - Fix conversion of `ImageSummary._value` to tuple in `metrics.py`.

Pull Request - State: closed - Opened by apghml 10 months ago

#391 - Support --index-url for gcs bundler.

Pull Request - State: closed - Opened by markblee 11 months ago

#390 - Workaround qrm API segfaults.

Pull Request - State: closed - Opened by markblee 11 months ago

#389 - Fix dataflow bundler specs.

Pull Request - State: closed - Opened by markblee 11 months ago

#388 - Pin google-api-core.

Pull Request - State: closed - Opened by markblee 11 months ago

#387 - Cleanup all procs when encountering an exception.

Pull Request - State: closed - Opened by markblee 11 months ago

#386 - Adds additional validation to jobspecs.

Pull Request - State: closed - Opened by markblee 11 months ago

#385 - update bastion to use google storage instead of gsutil

Pull Request - State: closed - Opened by zhzhyi 11 months ago

#384 - Dataflow changes

Pull Request - State: open - Opened by jiya-zhang 11 months ago

#383 - Add more Adastar optimizer stats

Pull Request - State: closed - Opened by taolei87 11 months ago

#382 - [misc] Update urls to lingvo/jax with fixed commit

Pull Request - State: closed - Opened by zhzhyi 11 months ago

#381 - Update attrs dependency

Pull Request - State: closed - Opened by WeersProductions 11 months ago

#380 - Add more docs on CLI and installation.

Pull Request - State: closed - Opened by markblee 11 months ago

#379 - Add another layer of timeouts to workaround jax bug.

Pull Request - State: closed - Opened by markblee 11 months ago

#378 - Add ruff directives.

Pull Request - State: closed - Opened by apghml 11 months ago

#377 - Clean Unschedulable jobs.

Pull Request - State: closed - Opened by apghml 11 months ago

#376 - configure gcloud when configuring axlearn

Issue - State: closed - Opened by jiya-zhang 11 months ago - 1 comment

#375 - Preserve field types when using config_for_function.

Pull Request - State: closed - Opened by apghml 11 months ago

#374 - Decouples from flax.

Pull Request - State: closed - Opened by markblee 11 months ago

#373 - Makes adastar_optimizer compute `raw_update_norm` …

Pull Request - State: closed - Opened by ruomingp 11 months ago

#372 - add topic tag "deep-learning"

Issue - State: open - Opened by Beliavsky 11 months ago

#371 - Supports configuring distributed initialization timeout.

Pull Request - State: closed - Opened by markblee 11 months ago

#370 - Draft PR to add GoodPut Measurement

Pull Request - State: closed - Opened by jiya-zhang 11 months ago

#369 - Generalizes the bastion runner pattern.

Pull Request - State: closed - Opened by markblee 11 months ago

#368 - Don't load evalers for pretrained builder.

Pull Request - State: closed - Opened by markblee 11 months ago

#367 - Don't override user-supplied tf loglevel.

Pull Request - State: closed - Opened by markblee 11 months ago

#366 - Create separate pre-commit job

Pull Request - State: closed - Opened by madrob 11 months ago

#365 - Allow specifying --cache-from to speedup builds.

Pull Request - State: closed - Opened by markblee 11 months ago

#364 - Workaround GCP bad gateway.

Pull Request - State: closed - Opened by markblee 11 months ago

#363 - bump up transformers to 4.38.2

Pull Request - State: closed - Opened by gyin94 11 months ago

GitHub / apple/axlearn issues and pull requests