Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Nov 4, 2025

• [mxfp8 moe training] compute prefix sum of group sizes inside kernel intead of precomputing (#3285)

  danielvegamyhre

  authored
  01374ebView commit details

  Copy full SHA for 01374eb

  Browse repository at this point

• fix unit test to use no grad (#3283)

  liangel-02

  authored
  8e3b3daView commit details

  Copy full SHA for 8e3b3da

  Browse repository at this point

• updating flatten/unflatten functions (#3282)

  liangel-02

  authored
  ba3ac9fView commit details

  Copy full SHA for ba3ac9f

  Browse repository at this point

Commits on Nov 3, 2025

• Use nn_module_stack instead

  Show description for 9266734
  tugsbayasgalan

  authored
  9266734View commit details

  Copy full SHA for 9266734

  Browse repository at this point

• [mxfp8 moe training] make scaling mode configurable and make rceil default (#3271)

  danielvegamyhre

  authored
  f657903View commit details

  Copy full SHA for f657903

  Browse repository at this point

• Set model parameters to requires_grad=False (#3272)

  Show description for 2a89491
  

  Jesse Cai

  authored
  2a89491View commit details

  Copy full SHA for 2a89491

  Browse repository at this point

Commits on Oct 31, 2025

• [mxfp8] fix test nan != nan issue (#3273)

  danielvegamyhre

  authored
  315e9b4View commit details

  Copy full SHA for 315e9b4

  Browse repository at this point

• add bias handling for a_1_128_w_128_128 float8 scaling (#3259)

  Show description for f856d36
  vkuzo

  authored
  f856d36View commit details

  Copy full SHA for f856d36

  Browse repository at this point

• add a_1_128_w_128_128 (DeepSeek) float8 scaling for inference (#3257)

  Show description for b49178c
  vkuzo

  authored
  b49178cView commit details

  Copy full SHA for b49178c

  Browse repository at this point

• Update torchao + unsloth integration on README (#3267)

  Show description for e418066
  andrewor14

  authored
  e418066View commit details

  Copy full SHA for e418066

  Browse repository at this point

• Add per tensor fp8 quantization support for conv3d (#3215)

  Show description for 258387a
  jerryzh168

  authored
  258387aView commit details

  Copy full SHA for 258387a

  Browse repository at this point

Commits on Oct 30, 2025

• move float8 blockwise kernels out of prototype (#3256)

  Show description for 1e473ed
  vkuzo

  authored
  1e473edView commit details

  Copy full SHA for 1e473ed

  Browse repository at this point

• properly skip float8 inference tests without fbgemm (#3255)

  Show description for d3bec87
  vkuzo

  authored
  d3bec87View commit details

  Copy full SHA for d3bec87

  Browse repository at this point

• [mxfp8 moe training] update readme and tests (#3260)

  danielvegamyhre

  authored
  0f05b40View commit details

  Copy full SHA for 0f05b40

  Browse repository at this point

Commits on Oct 29, 2025

• [CPU] Improve INT8 SDPA template (#3230)

  Show description for 3577306
  Xia-Weiwen

  authored
  3577306View commit details

  Copy full SHA for 3577306

  Browse repository at this point

• [mxfp8 moe training] initialize zero tensor differently to avoid d2h sync (#3253)

  danielvegamyhre

  authored
  d84f5b8View commit details

  Copy full SHA for d84f5b8

  Browse repository at this point

Commits on Oct 28, 2025

• Add quantize_ nn.Parameter support (#3083)

  Show description for 50a555b
  

  Jesse Cai

  authored
  50a555bView commit details

  Copy full SHA for 50a555b

  Browse repository at this point

• [mxfp8 moe training] bench and profile mxfp8 a2a fwd and bwd separately (#3203)

  danielvegamyhre

  authored
  dffb3a0View commit details

  Copy full SHA for dffb3a0

  Browse repository at this point

• [mxfp8 moe training] simplify e8m0 -> fp32 calc (#3201)

  danielvegamyhre

  authored
  d089c6aView commit details

  Copy full SHA for d089c6a

  Browse repository at this point

• [mxfp8 moe training] integrate triton quant/dequant kernels into mxfp8 all to all (#3197)

  danielvegamyhre

  authored
  6a62549View commit details

  Copy full SHA for 6a62549

  Browse repository at this point

• [mxfp8 moe training] add triton kernel for mxfp8 dequantization (#3195)

  danielvegamyhre

  authored
  7537d99View commit details

  Copy full SHA for 7537d99

  Browse repository at this point

• create a separate test for mx and nv serialization (#3251)

  Show description for 41a0778
  vkuzo

  authored
  41a0778View commit details

  Copy full SHA for 41a0778

  Browse repository at this point

Commits on Oct 27, 2025

• nvfp4: support inference_mode and rank 3 (#3240)

  Show description for 03c2d28
  vkuzo

  authored
  03c2d28View commit details

  Copy full SHA for 03c2d28

  Browse repository at this point

• mx: support inference_mode and rank 3+ (#3238)

  Show description for 14eff10
  vkuzo

  authored
  14eff10View commit details

  Copy full SHA for 14eff10

  Browse repository at this point

Commits on Oct 25, 2025

• Only convert to int4 preshuffled tensor in H100 (#3245)

  Show description for f303f4c
  jerryzh168

  authored
  f303f4cView commit details

  Copy full SHA for f303f4c

  Browse repository at this point

Commits on Oct 24, 2025

• [moe training] update readme with links, cleanup (#3239)

  danielvegamyhre

  authored
  e9c7beaView commit details

  Copy full SHA for e9c7bea

  Browse repository at this point

Commits on Oct 23, 2025

• Fix Wq size check

  Show description for 53b5efd
  RandySheriff

  authored
  53b5efdView commit details

  Copy full SHA for 53b5efd

  Browse repository at this point

• docs: fix qat description in README.md (#3212)

  orangeH25

  authored
  204cd48View commit details

  Copy full SHA for 204cd48

  Browse repository at this point

• Add Unsloth + QAT blog to latest news (#3227)

  andrewor14

  authored
  f3fc5e7View commit details

  Copy full SHA for f3fc5e7

  Browse repository at this point

Commits on Oct 22, 2025

• Update README.md (#3225)

  Show description for 54a7b85
  jerryzh168

  authored
  54a7b85View commit details

  Copy full SHA for 54a7b85

  Browse repository at this point

• [mxfp8 moe training] make compile vs triton for dim0 cast configurable (#3219)

  danielvegamyhre

  authored
  7e68d5eView commit details

  Copy full SHA for 7e68d5e

  Browse repository at this point

Commits on Oct 21, 2025

• enable custom MKN in inference roofline script (#3224)

  Show description for beee153
  vkuzo

  authored
  beee153View commit details

  Copy full SHA for beee153

  Browse repository at this point

• mxfp8 inference roofline: add fusion to observed (#3223)

  Show description for 6452b4a
  vkuzo

  authored
  6452b4aView commit details

  Copy full SHA for 6452b4a

  Browse repository at this point

• fix mxfp8 matmul benchmark (#3221)

  Show description for 13595c5
  vkuzo

  authored
  13595c5View commit details

  Copy full SHA for 13595c5

  Browse repository at this point

• [moe training] change api _scaled_grouped_mm -> _quantize_then_scaled_grouped_mm (#3218)

  danielvegamyhre

  authored
  82ae011View commit details

  Copy full SHA for 82ae011

  Browse repository at this point