Search results

[float8] add auto_filter_for_recipe to float8

CLA SignedThis label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

float8

topic: new featureUse this tag if this PR adds a new featureUse this tag if this PR adds a new feature

Status: Open (in progress).

pytorch/aonumber 2410

#2410 In pytorch/ao;

· danielvegamyhre opened on Jun 18, 2025

Add round_scales_to_power_of_2 option for float quantization

CLA SignedThis label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

float8

topic: new featureUse this tag if this PR adds a new featureUse this tag if this PR adds a new feature

Status: Open (in progress).

pytorch/aonumber 2323

#2323 In pytorch/ao;

· drisspg opened on Jun 6, 2025

Add a way to do power of 2 scaling

CLA SignedThis label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

float8

topic: new featureUse this tag if this PR adds a new featureUse this tag if this PR adds a new feature

Status: Open (in progress).

pytorch/aonumber 2256

#2256 In pytorch/ao;

· drisspg opened on May 23, 2025

[float8] Investigate if workaround for slow cutlass rowwise GEMM when fast_accum=False is still needed after perf improvments and potentially optimize GEMM further

float8

Status: Open.

#2184 In pytorch/ao;

· danielvegamyhre opened on May 8, 2025

[float8] Support power of 2 scales with PerRow scales for inference

float8

Status: Open.

#2182 In pytorch/ao;

· danielvegamyhre opened on May 7, 2025

cast to mxfp8 across dim1 should be performant

float8

triaged

Status: Open.

#1945 In pytorch/ao;

· vkuzo opened on Mar 24, 2025

[Bug] FSDP2 FP8 compatibility problem with nn.Linear layers (GPU count > out_features)

distributed

float8

triaged

Status: Open.

#1938 In pytorch/ao;

· HIT-cwh opened on Mar 24, 2025

fp8 quantization with FSDP2 error

float8

quantize

triaged

Status: Open.

#1929 In pytorch/ao;

· happynear opened on Mar 20, 2025

Does torchao support FP8 Grouped GEMM?

float8

triaged

Status: Open.

#1928 In pytorch/ao;

· zigzagcai opened on Mar 20, 2025

placeholder for tracking saturated cast logic to float8_e4m3fn

float8

triaged

Status: Open.

#1912 In pytorch/ao;

· vkuzo opened on Mar 17, 2025

Can FP8 GEMM be enabled via module hooks instead of module swapping?

float8

triaged

Status: Open.

#1887 In pytorch/ao;

· zigzagcai opened on Mar 14, 2025

Fix wrong scale eps applied

CLA SignedThis label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

float8

topic: bug fixUse this tag for PRs that fix bugsUse this tag for PRs that fix bugs

Status: Open (in progress).

pytorch/aonumber 1770

#1770 In pytorch/ao;

· alexsamardzic opened on Feb 24, 2025