Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on May 24, 2024

• Add the layernorm specialized op. (huggingface#2212)

  Show description for 1df2bdd
  LaurentMazare

  authored
  1df2bdd

  Copy full SHA for 1df2bdd

• More efficient cuda implementation for ConvTranspose1d. (huggingface#2211)

  Show description for 6f0b807
  LaurentMazare

  authored
  6f0b807

  Copy full SHA for 6f0b807

Commits on May 23, 2024

• Avoid a contiguous call in the quantized phi 3 model. (huggingface#2209)

  Show description for d54e02d
  LaurentMazare

  authored
  d54e02d

  Copy full SHA for d54e02d

• Simplify the KvCache api. (huggingface#2207)

  LaurentMazare

  authored
  45e235a

  Copy full SHA for 45e235a

• Add a couple kv-cache helper functions. (huggingface#2206)

  LaurentMazare

  authored
  31cf641

  Copy full SHA for 31cf641

• Add Phi-3 Medium (huggingface#2205)

  janimo

  authored
  77ea479

  Copy full SHA for 77ea479

Commits on May 22, 2024

• Add some missing where-cond kernels for metal. (huggingface#2203)

  LaurentMazare

  authored
  72e7ca5

  Copy full SHA for 72e7ca5

Commits on May 21, 2024

• Add RandomNormal ONNX operator (huggingface#2200)

  mokulus

  authored
  7ff921c

  Copy full SHA for 7ff921c

• Remove the deprecated wav crate in favor of hound. (huggingface#2202)

  LaurentMazare

  authored
  9b8537a

  Copy full SHA for 9b8537a

Commits on May 18, 2024

• Use flash-attn in gemma. (huggingface#2195)

  Show description for 7ebc354
  LaurentMazare

  authored
  7ebc354

  Copy full SHA for 7ebc354

• Support flash-attn in quantized phi3. (huggingface#2194)

  LaurentMazare

  authored
  eefc1c7

  Copy full SHA for eefc1c7

• Add a slice_set op. (huggingface#2193)

  Show description for 01545f7
  LaurentMazare

  authored
  01545f7

  Copy full SHA for 01545f7

Commits on May 16, 2024

• Support embedding model gte-Qwen1.5-7B-instruct (huggingface#2190)

  Show description for 349c3e8
  

  yinguobing

  and

  LaurentMazare

  authored
  349c3e8

  Copy full SHA for 349c3e8

• chore: add fix for windows cudarc into the readme (huggingface#2189)

  Cifko

  authored
  bdaa342

  Copy full SHA for bdaa342

Commits on May 15, 2024

• Allow the threshold argumet to be negative in the segment-anything example (huggingface#2187)

  Show description for cc80e06
  vargad

  authored
  cc80e06

  Copy full SHA for cc80e06

Commits on May 12, 2024

• Fix VarBuilder::from_slice_safetensors (huggingface#2180)

  Show description for 13c64f6
  boustrophedon

  authored
  13c64f6

  Copy full SHA for 13c64f6

Commits on May 11, 2024

• Add SliceSafetensors. (huggingface#2179)

  Show description for 21f82a5
  LaurentMazare

  authored
  21f82a5

  Copy full SHA for 21f82a5

• Make it possible to use TF32 accumulation in F32 matmuls. (huggingface#2178)

  Show description for 9cff7bc
  LaurentMazare

  authored
  9cff7bc

  Copy full SHA for 9cff7bc

Commits on May 9, 2024

• Switch cudarc back to dynamic linking. (huggingface#2176)

  LaurentMazare

  authored
  d9bc5ec

  Copy full SHA for d9bc5ec

Commits on May 8, 2024

• Update cudarc requirement from 0.11.0 to 0.11.1 (huggingface#2174)

  Show description for 84328e2
  sidharthrajaram

  authored
  84328e2

  Copy full SHA for 84328e2

Commits on May 6, 2024

• Update cudarc requirement from 0.10.0 to 0.11.0 (huggingface#2165)

  Show description for 82b641f
  

  dependabot[bot]

  and

  LaurentMazare

  authored
  82b641f

  Copy full SHA for 82b641f

Commits on May 5, 2024

• Use write rather than try-write on the metal rw-locks. (huggingface#2162)

  LaurentMazare

  authored
  01794dc

  Copy full SHA for 01794dc

Commits on May 4, 2024

• Force the revision for the phi3-llama quantized models. (huggingface#2159)

  LaurentMazare

  authored
  a75cd81

  Copy full SHA for a75cd81

• Separate quantized phi-3 implementation. (huggingface#2157)

  Show description for b13a82a
  LaurentMazare

  authored
  b13a82a

  Copy full SHA for b13a82a

Commits on May 3, 2024

• Pin the version used for the quantized phi 3 gguf file. (huggingface#2156)

  LaurentMazare

  authored
  59b18d9

  Copy full SHA for 59b18d9

• Bump the version number to 0.5.1. (huggingface#2155)

  Show description for 89f53b9
  LaurentMazare

  authored
  89f53b9

  Copy full SHA for 89f53b9

Commits on May 1, 2024

• Support top-k in tthe llama example. (huggingface#2150)

  LaurentMazare

  authored
  a09d451

  Copy full SHA for a09d451

Commits on Apr 29, 2024

• F16/BF16 bugfix (bis). (huggingface#2143)

  Show description for fa06f5f
  LaurentMazare

  authored
  fa06f5f

  Copy full SHA for fa06f5f

• Bugfix the recent f16/bf16 changes. (huggingface#2142)

  LaurentMazare

  authored
  09d4845

  Copy full SHA for 09d4845

• Bug Fix: When converting a tensor to a variable, clone if the tensor is already a variable. (huggingface#2124)

  Show description for a0d03ad
  

  jsdt

  and

  Jeffrey Dallatezza

  authored
  a0d03ad

  Copy full SHA for a0d03ad

• Fix sigmoid gradient calculation and move sigmoid into a specialized op (huggingface#2114)

  Show description for 3bbb88f
  

  MilkFather

  and

  LaurentMazare

  authored
  3bbb88f

  Copy full SHA for 3bbb88f

• Add a toggle for F16/BF16 accumulation in gemm. (huggingface#2141)

  Show description for ed7b99f
  LaurentMazare

  authored
  ed7b99f

  Copy full SHA for ed7b99f

Commits on Apr 28, 2024

• Add a forward_via_f16 method to the qmatmul op. (huggingface#2138)

  LaurentMazare

  authored
  287013e

  Copy full SHA for 287013e

• Add the cuda dequantize f16 kernels. (huggingface#2137)

  Show description for eb26e24
  LaurentMazare

  authored
  eb26e24

  Copy full SHA for eb26e24

• chore: fix some typos in comments (huggingface#2121)

  Show description for c68ed89
  hardlydearly

  authored
  c68ed89

  Copy full SHA for c68ed89

Pagination

Previous

Next