mosaicml/llm-foundry

Pull Requests Commits

snarayan21 committed 1 year ago

fa9e4975

Raise DatasetTooSmall exception if canonical nodes is less than num samples (#1518)

irenedea committed 1 year ago

Verified 5465db4f

Peft fsdp (#1520)

dakinggg committed 1 year ago

Verified 6d932602

Raise MisconfiguredDatasetError from original error (#1519)

irenedea committed 1 year ago

Verified 6849eaa2

Sort callbacks so that CheckpointSaver goes before HuggingFaceCheckpointer (#1515)

irenedea committed 1 year ago

Verified 8a8de18d

Put 2.3 image back in release examples (#1513)

dakinggg committed 1 year ago

Verified e8eca4fa

Fix cross attention for blocks (#1512)

gupta-abhay committed 1 year ago

Verified 1adff747

Allow passing key_value_statest for x-attn through MPT Block (#1511)

gupta-abhay committed 1 year ago

Verified 4ab483f6

Update ci-testing to 0.2.2 (#1503)

dakinggg committed 1 year ago

Verified 02802c53

Update mlflow requirement from <2.16,>=2.14.1 to >=2.14.1,<2.17 (#1506)

dependabot[bot] committed 1 year ago

Verified 01a514c5

[WIP] Only torch 2.4.0 compatible (#1505)

snarayan21 committed 1 year ago

Verified 867a405b

Torch 2.4 in docker images (#1491)

snarayan21 committed 1 year ago

Verified 31c8ba2c

Update to ci-testing 0.2.0 (#1500)

dakinggg committed 1 year ago

Verified b21cc0c1

Enable cross attention layers (#1495)

gupta-abhay committed 1 year ago

Verified d0dc82d2

Fix formatting of loss spike & high loss error messages (#1498)

joyce-chen-uni committed 1 year ago

Verified 07bec5a7

Removing error message for reusing kv cache with torch attn (#1497)

ShashankMosaicML committed 1 year ago

Verified 6d9d79bc

tokenizer is never built when converting finetuning dataset (#1496)

eldarkurtic committed 1 year ago

Verified 100f40f4

Align logged attributes for errors and run metadata in kill_loss_spike_callback.py (#1494)

joyce-chen-uni committed 1 year ago

Verified 85161819

Enable passing in external position ids (#1493)

gupta-abhay committed 1 year ago

Verified bf6cfdf8

Track loss spikes and high losses during training (#1473)

joyce-chen-uni committed 1 year ago

Verified 0db44252

master_weights_dtype not supported by ComposerHFCausalLM.__init__() (#1485)

eldarkurtic committed 1 year ago

Verified 51e92c49

Bump composer to 0.24.1 + FSDP config device_mesh deprecation (#1487)

snarayan21 committed 1 year ago

Verified 249db937

Fix device mismatch when running hf.generate (#1486)

ShashankMosaicML committed 1 year ago

Verified 2978c60b

fix hf checkpointer (#1489)

milocress committed 1 year ago

Verified abdf7cf9

Bump hf-transfer from 0.1.3 to 0.1.8 (#1480)

dependabot[bot] committed 1 year ago

Verified cef39d19

Use Streaming version 0.8.1 (#1479)

snarayan21 committed 1 year ago

Verified 273c2c01

Make sample tokenization extensible (#1478)

gupta-abhay committed 1 year ago

Verified 65936cfb

Add date_string when applying tokenizer chat template (#1474)

snarayan21 committed 1 year ago

Verified e235f420

Add prepare fsdp back (#1477)

dakinggg committed 1 year ago

Verified e84d97e5

Remove extra bos for prompt/response data with llama3.1 (#1476)

dakinggg committed 1 year ago

Verified 01418f8b

Older